Hindawi Publishing CorporationBioMed Research InternationalVolume 2013 Article ID 687607 13 pageshttpdxdoiorg1011552013687607

Research ArticleComputer-Assisted System with Multiple Feature Fused SupportVector Machine for Sperm Morphology Diagnosis

Kuo-Kun Tseng1 Yifan Li1 Chih-Yu Hsu2 Huang-Nan Huang3

Ming Zhao1 and Mingyue Ding4

1 Department of Computer Science and Technology Harbin Institute of Technology Shenzhen Graduate SchoolShenzhen Guangdong 518055 China

2Department of Information and Communication Engineering Chaoyang University of Technology Taichung 41349 Taiwan3Department of Mathematics Tunghai University Taichung 40704 Taiwan4Huazhong University of Science and Technology Wuhan 430074 China

Correspondence should be addressed to Chih-Yu Hsu tccnchsugmailcom

Received 15 April 2013 Revised 4 July 2013 Accepted 23 July 2013

Academic Editor Lei Chen

Copyright copy 2013 Kuo-Kun Tseng et alThis is an open access article distributed under the Creative CommonsAttribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Sperm morphology is an important technique in identifying the health of sperms In this paper we present a new system andnovel approaches to classify different kinds of sperm images in order to assess their health Our approach mainly relies on a one-dimensional feature which is extracted from the spermrsquos contour with gray level information Our approach can handle rotationand scaling of the image Moreover it is fused with SVM classification to improve its accuracy In our evaluation our method hasbetter performance than the existing approaches to sperm classification

1 Introduction

With the development of modern computer technologymedical imaging has played an important role in clinicaldiagnosis and treatment Medical image analysers are facingthe challenge of precisely extracting information from themedical image with the help of computer-assisted systemsSince people have become more and more concerned aboutthe health of the next generation morphology would be oneimportant technique to identify the health of sperms Toexamine whether or not the sperms are healthy it is essentialto inspect the sperms to assess their appearance Currentlysperm quality is mostly judged by experts and doctorsBecause of the numerous types of sperm shape the efficiencyand accuracy relying on human assessment are not idealAs computer morphology technology develops quantitativeanalysis of sperm morphology is demanded to assist doctorsin their diagnoses Thus this research is intended to design ahelpful sperm classification system

Sperm morphology is an image classification problemin sperm imaging It first detects a segment of the spermimage after which feature extraction and analysis is possible

for example sperm length width and size followed byfurther classification according to sperm features [1] As aresult solving the problem of sperm image recognition andclassification can be valuable for aspects of sperm diagnosis

Our spermmorphology system is equippedwith amicro-scope connected to a computer to observe the real-timesperm imageThemicroscope helped us to take photos of thesperm images and input them into our computer With theinput we managed to obtain all the results and conclusionsThe system and its equipment are shown in Figure 1

In addition to system implementation this research hasmade the following contributions

(1) We proposed two approaches to transform the spermcontour into a one-dimensional waveform as an anal-ysis featureThe first algorithm takes advantage of thedistance between two points on the edge to producea waveform The second computes the distance fromthe geometric centre to the edge as the vertical valueof the waveform

(2) After extraction we proposed an SVM classificationon these waveforms with rank and grey level features

2 BioMed Research International

Figure 1 The sperm morphology diagnosis system with a micro-scope

According to our comprehensive survey this has notyet been used in sperm classification

(3) We also conducted a complete comparison We com-pared our approaches with the K-nearest neighbourScale-Invariant Feature Transform (SIFT) and theelliptic model The experiment results show betterperformance than previous methods

In our evaluation we applied our approach to a spermdatabase The results show that our idea is feasible and givesbetter performance than the existing approaches

The rest of this paper comprises four parts Section 2introduces some other research studied to help ourwork Sec-tion 3 proposes the architecture and algorithm introducingthe details of our algorithm which leads to a more detailedunderstanding of our approach Section 4 provides the resultsand discussion proving that our work is feasible and finallythe conclusion is given in Section 5

2 Related Works

For this research on sperm morphology we reviewed therelated works on segmentation extraction shape descriptorand the classification algorithm as shown in Figure 2

21 Segmentation First segmentation takes place so we haveto look at several pieces of the literature [2ndash6] related toimage segmentation The first article [3] presents a methodfor optic nerve head segmentation and its validation Themethod is based on theHough transform and anchored activecontour model The results were validated by comparingthe performance of different classifiers and showed that thisapproach is suitable for automated diagnosis of and screeningfor glaucoma

Considering that there is no guarantee that the sperms weobserved will appear with the posture and position we needit is absolutely necessary for us to investigate how to deal withactive contours In research [2] they introduce a geometricalvariation frame that uses active contours to segment andobtain features from images at the same time

To obtain a better result of image segmentation paper[4] enhanced the lane for interactive image segmentation byincremental path map construction a modified version of

Related work

Segmentation

Nerve headsegmentation [6]

Image segmentation[5 9]

Heart segmentation[10]

Bonesegmentation [11]

Extraction

Sperm morphologyassessment [3]Snakes active

contour model [4]T-snakes topologyadaptive snakes [7]

United snakes [8]

Shapedescriptor

Symmetry auto-detection [2]

Classification

Analysis of humanheartbeat [13]

K-nearest neighbor

SIFT [22]

Elliptic model[23 24]

Shape featureextraction [12]

SVM methods[14 15]

[16 17 18 19 2021]

Figure 2 Related work

the live lane that can extract objects from an image interac-tively with efficiency and repeatability It guarantees a strictlybounded response time and follows the target boundary withlittle digression

Furthermore in paper [5] a new method was proposedfor the local assessment of boundary detection by a simu-lated search Its boundary detection can be optimized perlandmark during model training The success of the methodwas demonstrated for cardiac image segmentation and it wasshown to improve the capture range and accuracy of theboundary detection Another paper [6] evaluated variousimage features and different search strategies for fitting activeshape models (ASMs) to bone object boundaries in digitizedradiographs It proposed an improved search procedure thatis more robust against outlier configurations in the boundarytarget points

22 Extraction The next topic is extraction Because theresults of the sperm-head contour extraction have an essentialinfluence on the classification we studied some issues whichprovide further information As it is of great importanceto obtain the sperm-head contour precisely we studiedarticles on how to abstract contours The first was ldquoSperm

BioMed Research International 3

morphology assessment using Davidrsquos classificationrdquo [7]Thispaper aimed to compare assessment of spermmorphology byusing Davidrsquos classification (DC) based on the strict criteria(computer-assisted sperm analysis (CASA) SC) for theirability to predict fertilization in a selected in vitro fertilization(IVF) population Their results showed that the DC spermmorphology analysis was less indicative of fertilization thanCASA SC

After the DC analysis we reviewed the paper on thetechnology of extracting objectsrsquo edges The first article [8]ldquoSnakesrdquo is an energy-minimizing spline guided by externalconstraint forces and influenced by image forces that pull ittowards features such as lines and edges Snakes are activecontour models they lock onto nearby edges localizingthem accurately Scale-space continuation can be used toenlarge the capture region surrounding a feature Snakesprovide a unified account of a number of visual problemsincluding detection of edges lines and subjective contoursmotion tracking and stereo matching We have used snakessuccessfully for interactive interpretation in which user-imposed constraint forces guide the snake near features ofinterest

As the snake may not do its job well enough for currentresearch we reviewed some improved algorithms such asT-snakes [9] topology adaptive snakes In this paper theypresent a new class of deformable contours (snakes) and areapplied to the segmentation of medical images They enabletopological flexibility among other features The resultingtopology adaptive snakes or ldquoT-snakesrdquo can be used to seg-ment some of the most complex-shaped biological structuresfrom medical images in an efficient and highly automatedmanner

Moreover other authors [10] present a framework calledunited snakes which has two key features First it unifies themost popular snake variants expanding the range of objectmodelling capabilities Second it embodies the idea of thetechnique known as live wire or intelligent scissors The twotechniques can be combined advantageously by introducingan effective hard constraint mechanism They apply unitedsnakes to several different medical image analysis tasksdemonstrating the generality accuracy and robustness of thetool

23 Shape Descriptor The third topic the most related workis called shape descriptor We focused on how to transformit into a one-dimensional feature To achieve the goal westudied further related articles The first paper [11] presentsa new symmetry autodetection approach The symmetry canbe detected automatically by using corner detection Duringthe process the contour can be transferred to a waveform

For additional study of the shape descriptor paper [12]focused on presenting the existing approaches of shape-basedfeature extraction Paper [13] introduced the extraction ofwaveform features by reduced binary features used to reducecomplexity and storage

24 Classification Another common image matchingapproach the K-nearest neighbour method as a compared

target proposes a method to fuse real-value K-nearestneighbour classifiers by feature grouping [14] The real-valueK-nearest neighbour classifier can approximate continuous-valued target functions In addition it is sensitive to featureperturbation Therefore when the multiple real-value K-nearest neighbour classifier is fused by feature groupingthe performance of the fusion will be better than thesingle classifier Another K-nearest neighbour method [15]presents a novel improvement to the K-nearest neighbourmean classifier (K-NNMC) K-NNMC finds the K-nearestneighbours for each class of training patterns separately andfinds the means for each of these K-neighbours (class-wise)Classification is undertaken according to the nearest meanpattern In experiments using several standard datasets ithas been shown that the proposed classifier provides betterclassification accuracy over the conventional K-nearestneighbour method and thus it is a suitable method to beused in data mining applications

As we have been using the SVM as an advanced methodto improve the performance of our approach in order toachieve a deep understanding of SVM we paid attention torelated SVM research Research work [16] presents a newvalid edge detection algorithm based on an SVM to avoid thedisadvantages of traditional image edge detection methodsIn another SVM related work [17] the authors compared theperformance of artificial-immune-system- (AIS) based algo-rithms to a Gaussian kernel-based SVMTheir experimenta-tion indicates that the AIS-based classification paradigm hasthe intrinsic property of dealing more efficiently with highlyskewed datasets In addition research [18] takes advantage ofSVM to characterize the sperm population structure relatedto freezability The SVM was generated using sperm motilityinformation captured by CASA from thawed semen ThisSVM method was used to characterize the motile spermsubpopulations for Iberian red deer

Research [19] provided more information on the effect ofSVM on sperm research An automated quantitative methodthat objectively classifies five distinct motility patterns ofmouse sperm using the SVM method was developed Itsparameters are associated with the classified tracks and wereincorporated into established SVM algorithms to generate aseries of equations These equations were integrated into abinary decision tree that sequentially sorts uncharacterizedtracks into distinct categories

Once we had finished reading about the sperm relatedSVM wemoved on to research [20] which reveals the advan-tages of the proposed mixed-feature model and presentsthe capability of identifying human facial expressions fromstatic images The subsequent framework is a multistagediscrimination model based on global appearance featuresextracted from two-dimensional principal component anal-ysis (2DPCA) and local texture represented by a local binarypattern (LBP) The experimental results indicate that theproposed mixed-feature model is feasible and outperformsthe single-feature model

We then tried to look for research on feature extractionand the SVM classifier [21] This paper introduces a newmethod for the early detection of colon cancer using a com-bination of feature extraction based on wavelets for Fourier

4 BioMed Research International

Original image

Segment sperm head

Extract contour

Build wave

End

SVM classification

Calculate graylevel value

Calculate gray()

Calculatecombined value

GLV GLV

0

10

12

14

16

18

20

0 10 20 30 40 50 60 70 80 90

35 40 45 50 55 60 65 70

7580859095

100105110115120

() ()

Number

Number Number

Number

1

2

97

152

11

12

114

113

1

2

038088547

040800834

11

12

044727911

044294079

Figure 3 The flow chart of the sperm classification system

transform infrared spectroscopy (FTIR) and classificationwith SVM

One popular robust image-matching approach is SIFT Inpaper [22] it performs a reliable matching between differentviews of an object or scene Its features are invariant to imagescale and rotation In that sense its images can be matchedwith high reliability against a large database of features frommany images

With regard to the elliptic model it tried to estimate thecontour of a sperm by an ellipse shape Research work [23]used an ellipse to classify sperm Another article [24] pro-posed a new method of sperm morphological classificationusing the elliptic shape parameterized by the discrete Fouriertransform and reconstructed with dyadic data points Theenclosed area of boundaries as a classification feature wascalculated and transformed by wavelet transform

Although some researchers focused on sperm classifica-tion none has used our approach for sperm imaging Thispaper presents a system and a novel approach which uses aone-dimensional contour and gray level features to diagnosedifferent sperms according to their characteristics

3 Proposed Approaches

31 Overall Procedure In this section we present the algo-rithm of our work First of all we provide the flow ofour approach as Figure 3 After image segmentation thesperm head contour is extracted we transform the edge of

the sperm image into separated coordinate points and dividethe coordinate into two vectors

32 Proposed Algorithms For the classification we appliedtwo methods In the first the bilateral symmetrical functionof (119899) for the continuous situation is defined as follows

119878 (119899) = int1198621

1003817100381710038171003817100381711987511989911987510038171003817100381710038171003817

2

119889119904 minus int1198622

1003817100381710038171003817100381711987511989911987510038171003817100381710038171003817

2

119889119904 (1)

where1198621 denotes the arc on the left of the axis while1198622 is theright arc

The symmetry function is proposed as follows

Percentsym (119860119899 (119896)) = 1 minussum abs (119860 (119894) minus 119860 (2119896 minus 119894))

sum 2 lowast sqrt (119860 (119894) lowast 119860 (2119896 minus 119894))

(2)

119860119899(119896) denotes the corner point array while the 119896th point is

the smallest value of 119878(119899)In our secondmethod the geometric centre is calculated

We used the following equation to compute the horizontalcoordinate

119909mid =sum119899

119894=1119909119894

119899 (3)

where 119909mid represents the horizontal coordinate of the geo-metric centre 119909

119894represents the horizontal coordinate of each

point on the contour and 119899 represents the number of points

BioMed Research International 5

on the edgeThe vertical coordinate could be calculated in thesame way

To obtain the distance from edge to the centre we takeadvantage of the following format

119889 (119894) = radic(119909119894minus 119909mid)

2+ (119910119894minus 119910mid)

2 (4)

Considering the difficulty in choosing the starting pointwe had to avoid the problem Therefore we took rank algo-rithm [13] which ignores where to start into consideration inorder to classify the results The rank algorithm transformsa sequence of numbers 119909

1 1199092 1199093 119909

119899 into a sequence

composed of 1 and 0 The transform format is as follows

119868119899=

0 if 119909119899le 119909119899minus1

1 if 119909119899 gt 119909119899minus1

(5)

Therefore we achieved a binary system with 1 and 0We transformed 5 consecutive numbers in 119868

119899into decimal

numbers By computing the times each decimal numberappeared we ascertained the rank and the probability of theoriginal sequence Taking advantage of rank and probabilitywe calculated the similarity between two sequences Theformat is as follows

119863119898 (1198781 1198782) =sum2119898minus1

119896=0

10038161003816100381610038161198771 (119908119896) minus 1198772(119908119896)1003816100381610038161003816 1199011 (119908119896) 1199012 (119908119896)

(2119898 minus 1)sum2119898minus1

119896=01199011(119908119896) 1199012(119908119896)

(6)

By calculating the 119863119898 we distinguished the ten mostrepresentative sequences and took them as the criteria

Using the rank algorithm [13] we achieved anotherbinary system with 1 and 0 Then we transformed 8 con-secutive numbers in 119868119899 into decimal numbers and finallywe obtained the rank and the probability of the sequenceWith format (6) we obtained the similarity among thesequences Similar to the previous method we achievedanother criterion

To avoid the problem of a single criterion being toolopsided we took the gray level value of the sperm intoconsideration First we calculated the gray level value of allpoints within the sperm using the following format

119866 =

119899

sum1

119892119894 (7)

where 119866 represents the summation of the gray level value ofeach point119892119894 represents the gray level value of each point and119899 represents the total number of points within the sperm

Then we computed the gray percentage using the follow-ing format

119875 = 1 minus119866

119899 lowast 255 (8)

where 119875 represents the level of darkness

321 Joint Rank Difference and Gray Level Method By cal-culating the average rank difference of normal and abnormal

sperms we found the dividing line between them Thus therank difference itself can work as a judgment as to whetheror not the test sperm is normal We combined the two rankdifferences originating from the distance from centre to thecontour and the grey level value of the pixels in a sperm bycalculating the sum of the test spermrsquos distance average rankdifference and the average grey level value rank differencewhile each of them takes a certain weight The format is asfollows

119862 = 120572 lowast dARD + (1 minus 120572) lowast 255 lowast 119875 (9)

119862 represents the value that is the combination of the twostandards 120572 represents the weight of average rank differencedARD represents the average rank difference of distance fromthe centre to the contour

With the combination value we enhanced the originaljudgment by considering more elements and the importanceof each To find the dividing line we collected all thecombination values of the sperms and chose one of theiraverage values to provide the best accuracy for the dividingline

33 Fused SVM Method To achieve a better result we fusedthe SVM method as an advanced classifier SVM Methodsare supervised learning models with associated learningalgorithms that analyse data and recognize patterns and areused for classification and regression analysisThe basic SVMtakes a set of input sperm features and predicts for each giveninput the possible class form normal or abnormal making ita nonprobabilistic binary linear classifier

Given a training dataset (1199091 1199101) (1199092 1199102) (119909119899 119910119899)where 119909

119894isin R and 119910

119894is either 1 or minus1 indicating the class

to which the point 119909119894belongs let x = [119909

1 1199092 119909

119899]119879

The construction of the hyperplane for a linearly separableproblem is w119879x + 119887 = 0 where w is the normal vectorto the hyperplane and the parameter 119887w determines theoffset of the hyperplane from the origin along the normalvector w Thus the margin between the hyperplane and thenearest point is maximized and can be posed as the followingproblem

minw119887120585

1

2w119879w + 119862

119899

sum119894=1

120585119894

subject to 119910119894(119908119879119909119894+ 119887) ge 1 minus 120585

119894 119894 = 1 2 119899

120585119894ge 0

(10)

where 119862 is a user-defined constant as the penalty parameterof the error term

The SVM requires the optimal solution We use theLIBSVM[25] to solve this optimization problemwith the userguide given in [26]

As for the input of the SVM approach we fused the greylevel value and dARD as well First of all it extracts grey levelvalue of the pixels within the sperm head It starts from thecentre of the sperm and then it computes the following greylevel values in a square roundabout path The path map isshown in Table 1 as the sequence of numbers

6 BioMed Research International

Table 1 Grey level value extraction path

43 42 41 40 39 38 37 6444 21 20 19 18 17 36 6345 22 7 6 5 16 35 6246 23 8 1 4 15 34 6147 24 9 2 3 14 33 6048 25 10 11 12 13 32 5949 26 27 28 29 30 31 5850 51 52 53 54 55 56 57

Original sperm Sperm head

Figure 4 Extract sperm head from normal sperm

When the path collides with the contour the sequenceof grey level values takes advantage of the Rank algorithmin transforming 8 consecutive numbers 119868119899 into the decimalnumbers during the ends of extraction Then we achievedrank sequences and 119863119898 between every pair of images whichcan represent their similarities As a result we can tell whichimage represents the rest of the images the best by calculatingthe sum of the 119863

119898value from one image to all the other

images Thus we can put the rank sequences in order as wellWith the rank sequences we added the dARD sequences

at the end of corresponding rank sequence The newsequences consist of the training samples and test samplesThe ten training samples include five most representativesequences of normal sperms and five ones of abnormalsperms The rest sperms are the test samples

4 Fundamental Evaluation

In this fundamental evaluation we have undertaken variousexperiments on 80 normal and 80 abnormal sperms Theyare from a hospital the images are provided on httpscodegooglecompsupport-vector-machine-for-sperm-morphol-ogy-diagnosis as Supplementary material See Supplemen-tary Material available online at httpdxdoiorg1011552013687607 Before the classification of sperms thesegmentation is done by two steps (1) We pick the areawhere sperms do not overlay manually (2) We segment thesperms by reading the gray scale value of the picked areaWith knowing the location of the sperms we copy them fromthe area and build a new image to store each one of themWith a single sperm in an image it can cut off its tail easily

Figures 4 and 5 present an example of our segmentationresult

In this way we segmented all the 80 normal and 80abnormal sperm heads from 80 pieces of sperm images

41 Results of Our Waveform Extraction Methods In thissection we present part of the results of the steps in our

Original sperm Sperm head

Figure 5 Extract sperm head from abnormal sperm

experiments which explain how we made a choice andtheir performances In Table 2 the column Number lists thenumber of each imageThe columns Symmetry andMid showthe result of each transformation from image into waveform

Table 3 presents the waveform obtained by the firstmethod using four different starting points for the sameimage With the differences among four waves the problemarises from choosing one for the classification The resultof the second approach avoids the problem of picking thestarting point Thus we take the second result to classifysperms

42 Average Rank Difference In Figure 6 the horizontalaxis represents the serial number of each sperm image Thevertical axis represents the average difference of rank amongthe 80 normal or 80 abnormal sperms and the ten modelsperms As the result did not provide a clear dividing linebetween normal and abnormal sperms we adopted anothermethod to examine their differences We calculated thedifference in rank between the ten normal sperms and theother sperms Taking a look at Figure 6 if we take horizontalvalue 181 as the dividing line between normal and abnormalmeaning when the value of the average rank difference of asperm is below 181 we take the sperm as normal otherwisewe take it as abnormal In this case the accuracy could reach55

43 Grey Level Feature Then we focused on the influence ofthe grey level value Parts of the results are as shown in Figures7 and 8

In Figures 7 and 8 the horizontal axis represents the serialnumber of each sperm image In Figures 7(a) and 8(a) thevertical axis represents the average grey level value of thepoints of the sperms in the images In Figures 7(b) and 8(b)the vertical axis represents the grey level percentage of thesegmentation of the sperms

According to the percentage results we can ascertainthe percentage dividing line between normal and abnormalsperms to be 0268228785 With the division we can achieve725 accuracy

44 Joint Average Result We then tried to combine thetwo aspects as a judgment First of all we found the newparameter through format (7) By setting different 120572 wedetermined the different weight of each aspect The newparameter is as in Figure 9 It contains the new crite-ria produced with rank and gray level values of normaland abnormal sperms 120572 is the argument that represents

BioMed Research International 7

Table 2 Waveform extraction from both methods

Num Normal Symmetry Mid

1

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

25

20

15

10

5

0

2

0100200300400500600700800

1 9 17 25 33 41 49 57 65 73 81 89 970

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

3

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

4

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

5

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8502468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

6

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

7

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

8

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

9

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

8 BioMed Research International

Table 2 Continued

10

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

Num Abnormal Symmetry Mid

81

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

30

1 11 21 31 41 51 61 71 81 91 101

111

121

82

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

83

0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8505

101520253035

1 13 25 37 49 61 73 85 97 109

121

133

84

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 790

10203040506070

1 7 13 19 25 31 37 43 49 55 61 67 73 79

85

050

100150200250300350400

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

86

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

1012141618

1 7 13 19 25 31 37 43 49 55 61 67 73 79

87

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

88

0200400600800

10001200140016001800

1 7 13 19 25 31 37 43 49 55 61 67 73 7905

101520253035

1 7 13 19 25 31 37 43 49 55 61 67 73 79

BioMed Research International 9

Table 2 Continued

89

0200400600800

100012001400

1 7 13 19 25 31 37 43 49 55 61 67 73 790

5

10

15

20

25

30

1 7 13 19 25 31 37 43 49 55 61 67 73 79

90

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

Table 3 The waveform of different starting points from the first method

Startpoint Result

Start 1

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

Start 2

0

200

400

600

800

1000

1200

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800900

1000

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97

Table 4 Dividing line of C

120572 01 02 03 04 05 06 07 08 09DL 177 165 160 149 134 123 114 101 95

Table 5 Accuracy of different 120572

120572 01 02 03 04 05 06 07 08 09Acc 061 062 064 066 069 072 074 077 073

the weighted average rank difference taken in the criteriaWith the new parameter we can calculate its dividing line(DL) as a threshold

In Table 4 row DL represents the dividing line of each120572 value With the dividing line we can classify the spermsAssuming we have already known the exact classification ofeach sperm we can tell how many of them are correctlyclassified and thus we can come up with the accuracy Theaccuracy (Acc) of the different 120572 is as shown in Table 5

In Table 5 the row Acc represents the accuracy of each 120572value According to the result we chose the value 08 for 120572 Inthis case the accuracy can reach 76875

45 Fused SVM Result With all the data provided abovewe took advantage of the SVM method to undertake the

classification and used LIBSVM Software from httpwwwcsientuedutwsimcjlinlibsvm As for the parametersthe default values are used in the implementation and theSVM type is C-SVC the kernel function type is RBF function119890minus120574(119906minusV)2 = 1119896 cost = 1 cache size = 40 eps = 0001shrinking = 1 weight = 1

We input the training samples and test samples as men-tioned in Section 33 With ten training samples and testsamples of 160 sperms which consist of 80 normal and 80abnormal ones we obtain accuracy with 889

5 Comparison Evaluation

In this section we introduce the three methods we usedas comparisons to our method K-nearest neighbor methodworks quite effective as well the other two fail to satisfy us

51 Compared K-Nearest Neighbour Approach TheK-nearestneighbour method showed the following differences

We took the 160 images of sperms as the original clas-sifications and then we input the test image transformedall the images into gray and then calculated the differencebetween the test image and the original images Then wepicked the ten images with the smaller differences Aftercounting the classifications of the ten images we specified theone occurring most often to be the classification of the test

10 BioMed Research International

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a) Normal sperms

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(b) Abnormal sperms

Figure 6 Normal and abnormal sperms and their corresponding data

1 7 13 19 25 31 37 43 49 55 61 67 73 790

102030405060708090

100

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

(b)

Figure 7 Grey level value and percentage of normal sperms

0

20

40

60

80

100

120

140

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

07

(b)

Figure 8 Grey level value and percentage of abnormal sperms

image The classification of the top ten nearest neighbours isas in Figure 10

Taking the sperms whose sum value is larger than 3 asnormal and the sperms whose sum value is smaller than 4as abnormal we can tell that only 39 sperms are mistakenlyclassified providing accuracy of 75625

52 Compared SIFTApproach Lowe summed up the existingfeature detection method based on invariants technologyin 2004 and formally proposed an image scaling rotationand even affine transformation for invariant image with localfeature description operator based on scale space SIFT [22]

The SIFT algorithm first undertakes feature detection in scalespace and defines the key pointsrsquo positions and the scale ofthe key points and then it uses the main direction of theneighbourhood gradient of the key points as the directionfeatures of the points in order to achieve the operatorindependence of scale and the direction The format whichproduces the scale space is as follows

119871 (119909 119910 119905) = intinfin

120585=minusinfin

intinfin

120578

1

2120587119905119890minus(1205852+1205782)2119905

119891 (119909 minus 120585 119910 minus 120578) 119889120585 119889120578

(11)

BioMed Research International 11

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(a) 120572 = 01

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(b) 120572 = 02

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(c) 120572 = 03

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(d) 120572 = 04

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(e) 120572 = 05

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(f) 120572 = 06

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(g) 120572 = 07

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160

(h) 120572 = 08

020406080

100120140

1 13 25 37 49 61 73 85 97 109 121 133 145157

(i) 120572 = 09

Figure 9 The result of joint rank and gray level features of normal and abnormal sperms

12 BioMed Research International

1 7 13 19 25 31 37 43 49 55 61 67 73 790

2

4

6

8

10

12

(a) Normal sperms

1 7 13 19 25 31 37 43 49 55 61 67 73 790

051

152

253

354

45

(b) Abnormal sperms

Figure 10 The classification of the top ten neighbours of (a) normal and (b) abnormal sperms

Rectangularsegment

Segmentation afterrotation

Ellipse

Figure 11 The ellipse model to estimate the contour of the sperm

In format (11) t represents the scaling parameter Byundertaking convolution in the whole domain with a two-dimensionalGaussian kernel and input image we can achievescaling correspondent to t

The SIFT feature vector has the following features (a) it isthe local feature of an image which maintains invariance notonly on rotation scale and brightness variation but also onthe viewing angle the affine transformation and the noise(b) it is distinctive and informative and suitable for fastaccurate matching in a mass signature database (c) it canproduce a large number of SIFT feature vectors with fewobjects (d) it is of high speed

Matlab source code of SIFT is fromhttpwwwvlfeatorgindexhtml [27] We used the SIFT method to extract featurepoints of all the sperm images and achieved the ten normaland ten abnormal sperm images as in the training modelThen we calculated the difference between the trainingmodels and the other 70 normal and 70 abnormal spermimages With the dividing line we came up with accuracyof 50 The result turned out to be bad because the inputdata are simple sperm images and the SIFT algorithm cannotextract enough feature descriptors for classification

53 Compared Ellipse Model Approach With the image ofa single sperm we can use the ellipse model to estimatethe contour of the sperm First of all we used a rectangularsegment to cut out the sperm In order to make the rectangleclose to the sperm edge we needed to choose the rectanglewith the smallest area we rotated the sperm so that thesegmentation would be easier We calculated the left-mostupper-most downward-most and right-most points on theedges and built the rectangular segment based on them asdepicted in Figure 11

76888890

7563

5000

6625

000100020003000400050006000700080009000

10000

Jointaverage

K-neighbor SIFT Ellipticmodel

FusedSVM

()

Figure 12 Accuracy comparison of all methods

With the rectangle we achieved an ellipse to estimatethe sperm whose long axis radius is half the length of therectanglersquos long side and whose short axis radius is half thelength of the rectanglersquos short side As a result we ascertainedthe ovality which is the ratio of the length of the long axisto the short axis Through our experiment on 80 sperms wecould tell whether the sperm was too close to a circle whoseratio is close to 1 and whether the sperm was too slim whoseratio is close to 2 or even larger than 2 We took those whoseratio was larger than 12 and smaller than 18 as normal andthus we achieved accuracy of 6625

6 Conclusion

As shown in Figure 12 we have presented an approachfor the classification of sperm images with one-dimensionalwaveform and gray level features In the comprehensiveevaluation the joint average approach method was applied

BioMed Research International 13

to 160 sperm image samples and provided 76875 accuracyof judgment We then applied the fused SVM method tothe images and reached accuracy of 875 The result provesthat the proposed approach is superior to the previousapproaches

References

[1] N R Pal and S K Pal ldquoA review on image segmentationtechniquesrdquo Pattern Recognition vol 26 no 9 pp 1277ndash12941993

[2] A Yezzi L Zollei and T Kapur ldquoA variational frameworkfor integrating segmentation and registration through activecontoursrdquoMedical ImageAnalysis vol 7 no 2 pp 171ndash185 2003

[3] R Chrastek M Wolf K Donath et al ldquoAutomated segmenta-tion of the optic nerve head for diagnosis of glaucomardquoMedicalImage Analysis vol 9 no 4 pp 297ndash314 2005

[4] H W Kang and S Y Shin ldquoEnhanced lane interactive imagesegmentation by incremental path map constructionrdquo Graphi-cal Models vol 64 no 5 pp 282ndash303 2002

[5] J Peters O Ecabert C Meyer R Kneser and J Weese ldquoOpti-mizing boundary detection via Simulated Search with appli-cations to multi-modal heart segmentationrdquo Medical ImageAnalysis vol 14 no 1 pp 70ndash84 2010

[6] G Behiels F Maes D Vandermeulen and P Suetens ldquoEvalu-ation of image features and search strategies for segmentationof bone structures in radiographs using Active Shape ModelsrdquoMedical Image Analysis vol 6 no 1 pp 47ndash62 2002

[7] M Blanchard K Haguenoer A Apert et al ldquoSperm morphol-ogy assessment using Davidrsquos classiffication time to switch tostrict criteria Prospective comparative analysis in a selectedIVF populationrdquo International Journal of Andrology vol 34 no2 pp 145ndash152 2010

[8] MKass AWitkin andD Terzopoulos ldquoSnakes active contourmodelsrdquo International Journal of Computer Vision vol 1 no 4pp 321ndash331 1988

[9] TMcInerney andD Terzopoulos ldquoT-snakes topology adaptivesnakesrdquoMedical Image Analysis vol 4 no 2 pp 73ndash91 2000

[10] J Liang T McInerney and D Terzopoulos ldquoUnited SnakesrdquoMedical Image Analysis vol 10 no 2 pp 215ndash233 2006

[11] M Zhao Z Meng K K Tseng J S Pan and C Y Hsu ldquoSym-metry auto-detection based on contour and corner modelsrdquo inProceedings of the 5th International Conference on Genetic andEvolutionary Computing (ICGEC rsquo11) pp 345ndash349 September2011

[12] Y Mingqiang K Kidiyo and R Joseph ldquoA survey of shapefeature extraction techniquesrdquo Pattern Recognition vol 15 no7 pp 43ndash90 2008

[13] Y Wang and H Huang ldquoAnalysis of Human Heartbeat withEKG Signalsrdquo 2010

[14] Q Hua A Ji and Q He ldquoMultiple real-valued K nearestneighbor classifiers system by feature groupingrdquo in Proceedingsof the IEEE International Conference on Systems Man andCybernetics (SMC rsquo10) pp 3922ndash3925 October 2010

[15] P Viswanath and T Hitendra Sarma ldquoAn improvement to k-nearest neighbor classifierrdquo in Proceedings of the IEEE RecentAdvances in Intelligent Computational Systems (RAICS rsquo11) pp227ndash231 September 2011

[16] P Wu and Q Chen ldquoA novel SVM-based edge detectionmethodrdquo Physics Procedia vol 24 pp 2075ndash2082 2012

[17] D N Sotiropoulos and G A Tsihrintzis ldquoArtificial immunesystem-based classification in class-imbalanced image prob-lemsrdquo in Proceedings of the 8th International Conference onIntelligent InformationHiding andMultimedia Signal Processingpp 138ndash141 July 2012

[18] M Ramon FMartınez-Pastor O Garcıa-Alvarez et al ldquoTakingadvantage of the use of supervised learning methods forcharacterization of sperm population structure related withfreezability in the Iberian red deerrdquoTheriogenology vol 77 no8 pp 1661ndash1672 2012

[19] S G Goodson Z Zhang J K Tsuruta W Wang and DA OrsquoBrien ldquoClassification of mouse sperm motility patternsusing an automatedmulticlass support vectormachinesmodelrdquoBiology of Reproduction vol 84 no 6 pp 1207ndash1215 2011

[20] D T Lin and D C Pan ldquoIntegrating a mixed-feature modeland multiclass support vector machine for facial expressionrecognitionrdquo Integrated Computer-Aided Engineering vol 16no 1 pp 61ndash74 2009

[21] C G Cheng Y M Tian and W Y Jin ldquoA study on the earlydetection of colon cancer using the methods of wavelet featureextraction and SVM classifications of FTIRrdquo Spectroscopy vol22 no 5 pp 397ndash404 2008

[22] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004

[23] V R Nafisi M H Moradi and M H Nasr-Esfahani ldquoSpermidentification using elliptic model and tail detectionrdquo WorldAcademy of Science Engineering and Technology vol 6 pp 205ndash208 2005

[24] W J Yi K S Park and J S Paick ldquoParameterized charac-terization of elliptic sperm heads using Fourier representationand wavelet transformrdquo in Proceedings of the 20th AnnualInternational Conference of the IEEE Engineering in Medicineand Biology Society vol 20 pp 974ndash977 1998

[25] C C Chang andC J Lin LIBSVMmdasha library for support vectormachines Version 312 httpwwwcsientuedutwsimcjlinlibsvm

[26] CW Hsu C C Chang and C J Lin ldquoA practical guide to sup-port vector classificationrdquo 2010 httpwwwcsientuedutwsimcjlin

[27] A Vedaldi and B Fulkerson ldquoVLFeat an open and portablelibrary of computer vision algorithmsrdquo 2008 httpwwwvlfeatorg

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

2 BioMed Research International

Figure 1 The sperm morphology diagnosis system with a micro-scope

According to our comprehensive survey this has notyet been used in sperm classification

(3) We also conducted a complete comparison We com-pared our approaches with the K-nearest neighbourScale-Invariant Feature Transform (SIFT) and theelliptic model The experiment results show betterperformance than previous methods

In our evaluation we applied our approach to a spermdatabase The results show that our idea is feasible and givesbetter performance than the existing approaches

The rest of this paper comprises four parts Section 2introduces some other research studied to help ourwork Sec-tion 3 proposes the architecture and algorithm introducingthe details of our algorithm which leads to a more detailedunderstanding of our approach Section 4 provides the resultsand discussion proving that our work is feasible and finallythe conclusion is given in Section 5

2 Related Works

For this research on sperm morphology we reviewed therelated works on segmentation extraction shape descriptorand the classification algorithm as shown in Figure 2

21 Segmentation First segmentation takes place so we haveto look at several pieces of the literature [2ndash6] related toimage segmentation The first article [3] presents a methodfor optic nerve head segmentation and its validation Themethod is based on theHough transform and anchored activecontour model The results were validated by comparingthe performance of different classifiers and showed that thisapproach is suitable for automated diagnosis of and screeningfor glaucoma

Considering that there is no guarantee that the sperms weobserved will appear with the posture and position we needit is absolutely necessary for us to investigate how to deal withactive contours In research [2] they introduce a geometricalvariation frame that uses active contours to segment andobtain features from images at the same time

To obtain a better result of image segmentation paper[4] enhanced the lane for interactive image segmentation byincremental path map construction a modified version of

Related work

Segmentation

Nerve headsegmentation [6]

Image segmentation[5 9]

Heart segmentation[10]

Bonesegmentation [11]

Extraction

Sperm morphologyassessment [3]Snakes active

contour model [4]T-snakes topologyadaptive snakes [7]

United snakes [8]

Shapedescriptor

Symmetry auto-detection [2]

Classification

Analysis of humanheartbeat [13]

K-nearest neighbor

SIFT [22]

Elliptic model[23 24]

Shape featureextraction [12]

SVM methods[14 15]

[16 17 18 19 2021]

Figure 2 Related work

the live lane that can extract objects from an image interac-tively with efficiency and repeatability It guarantees a strictlybounded response time and follows the target boundary withlittle digression

Furthermore in paper [5] a new method was proposedfor the local assessment of boundary detection by a simu-lated search Its boundary detection can be optimized perlandmark during model training The success of the methodwas demonstrated for cardiac image segmentation and it wasshown to improve the capture range and accuracy of theboundary detection Another paper [6] evaluated variousimage features and different search strategies for fitting activeshape models (ASMs) to bone object boundaries in digitizedradiographs It proposed an improved search procedure thatis more robust against outlier configurations in the boundarytarget points

22 Extraction The next topic is extraction Because theresults of the sperm-head contour extraction have an essentialinfluence on the classification we studied some issues whichprovide further information As it is of great importanceto obtain the sperm-head contour precisely we studiedarticles on how to abstract contours The first was ldquoSperm

BioMed Research International 3

morphology assessment using Davidrsquos classificationrdquo [7]Thispaper aimed to compare assessment of spermmorphology byusing Davidrsquos classification (DC) based on the strict criteria(computer-assisted sperm analysis (CASA) SC) for theirability to predict fertilization in a selected in vitro fertilization(IVF) population Their results showed that the DC spermmorphology analysis was less indicative of fertilization thanCASA SC

After the DC analysis we reviewed the paper on thetechnology of extracting objectsrsquo edges The first article [8]ldquoSnakesrdquo is an energy-minimizing spline guided by externalconstraint forces and influenced by image forces that pull ittowards features such as lines and edges Snakes are activecontour models they lock onto nearby edges localizingthem accurately Scale-space continuation can be used toenlarge the capture region surrounding a feature Snakesprovide a unified account of a number of visual problemsincluding detection of edges lines and subjective contoursmotion tracking and stereo matching We have used snakessuccessfully for interactive interpretation in which user-imposed constraint forces guide the snake near features ofinterest

As the snake may not do its job well enough for currentresearch we reviewed some improved algorithms such asT-snakes [9] topology adaptive snakes In this paper theypresent a new class of deformable contours (snakes) and areapplied to the segmentation of medical images They enabletopological flexibility among other features The resultingtopology adaptive snakes or ldquoT-snakesrdquo can be used to seg-ment some of the most complex-shaped biological structuresfrom medical images in an efficient and highly automatedmanner

Moreover other authors [10] present a framework calledunited snakes which has two key features First it unifies themost popular snake variants expanding the range of objectmodelling capabilities Second it embodies the idea of thetechnique known as live wire or intelligent scissors The twotechniques can be combined advantageously by introducingan effective hard constraint mechanism They apply unitedsnakes to several different medical image analysis tasksdemonstrating the generality accuracy and robustness of thetool

23 Shape Descriptor The third topic the most related workis called shape descriptor We focused on how to transformit into a one-dimensional feature To achieve the goal westudied further related articles The first paper [11] presentsa new symmetry autodetection approach The symmetry canbe detected automatically by using corner detection Duringthe process the contour can be transferred to a waveform

For additional study of the shape descriptor paper [12]focused on presenting the existing approaches of shape-basedfeature extraction Paper [13] introduced the extraction ofwaveform features by reduced binary features used to reducecomplexity and storage

24 Classification Another common image matchingapproach the K-nearest neighbour method as a compared

target proposes a method to fuse real-value K-nearestneighbour classifiers by feature grouping [14] The real-valueK-nearest neighbour classifier can approximate continuous-valued target functions In addition it is sensitive to featureperturbation Therefore when the multiple real-value K-nearest neighbour classifier is fused by feature groupingthe performance of the fusion will be better than thesingle classifier Another K-nearest neighbour method [15]presents a novel improvement to the K-nearest neighbourmean classifier (K-NNMC) K-NNMC finds the K-nearestneighbours for each class of training patterns separately andfinds the means for each of these K-neighbours (class-wise)Classification is undertaken according to the nearest meanpattern In experiments using several standard datasets ithas been shown that the proposed classifier provides betterclassification accuracy over the conventional K-nearestneighbour method and thus it is a suitable method to beused in data mining applications

As we have been using the SVM as an advanced methodto improve the performance of our approach in order toachieve a deep understanding of SVM we paid attention torelated SVM research Research work [16] presents a newvalid edge detection algorithm based on an SVM to avoid thedisadvantages of traditional image edge detection methodsIn another SVM related work [17] the authors compared theperformance of artificial-immune-system- (AIS) based algo-rithms to a Gaussian kernel-based SVMTheir experimenta-tion indicates that the AIS-based classification paradigm hasthe intrinsic property of dealing more efficiently with highlyskewed datasets In addition research [18] takes advantage ofSVM to characterize the sperm population structure relatedto freezability The SVM was generated using sperm motilityinformation captured by CASA from thawed semen ThisSVM method was used to characterize the motile spermsubpopulations for Iberian red deer

Research [19] provided more information on the effect ofSVM on sperm research An automated quantitative methodthat objectively classifies five distinct motility patterns ofmouse sperm using the SVM method was developed Itsparameters are associated with the classified tracks and wereincorporated into established SVM algorithms to generate aseries of equations These equations were integrated into abinary decision tree that sequentially sorts uncharacterizedtracks into distinct categories

Once we had finished reading about the sperm relatedSVM wemoved on to research [20] which reveals the advan-tages of the proposed mixed-feature model and presentsthe capability of identifying human facial expressions fromstatic images The subsequent framework is a multistagediscrimination model based on global appearance featuresextracted from two-dimensional principal component anal-ysis (2DPCA) and local texture represented by a local binarypattern (LBP) The experimental results indicate that theproposed mixed-feature model is feasible and outperformsthe single-feature model

We then tried to look for research on feature extractionand the SVM classifier [21] This paper introduces a newmethod for the early detection of colon cancer using a com-bination of feature extraction based on wavelets for Fourier

4 BioMed Research International

Original image

Segment sperm head

Extract contour

Build wave

End

SVM classification

Calculate graylevel value

Calculate gray()

Calculatecombined value

GLV GLV

0

10

12

14

16

18

20

0 10 20 30 40 50 60 70 80 90

35 40 45 50 55 60 65 70

7580859095

100105110115120

() ()

Number

Number Number

Number

1

2

97

152

11

12

114

113

1

2

038088547

040800834

11

12

044727911

044294079

Figure 3 The flow chart of the sperm classification system

transform infrared spectroscopy (FTIR) and classificationwith SVM

One popular robust image-matching approach is SIFT Inpaper [22] it performs a reliable matching between differentviews of an object or scene Its features are invariant to imagescale and rotation In that sense its images can be matchedwith high reliability against a large database of features frommany images

With regard to the elliptic model it tried to estimate thecontour of a sperm by an ellipse shape Research work [23]used an ellipse to classify sperm Another article [24] pro-posed a new method of sperm morphological classificationusing the elliptic shape parameterized by the discrete Fouriertransform and reconstructed with dyadic data points Theenclosed area of boundaries as a classification feature wascalculated and transformed by wavelet transform

Although some researchers focused on sperm classifica-tion none has used our approach for sperm imaging Thispaper presents a system and a novel approach which uses aone-dimensional contour and gray level features to diagnosedifferent sperms according to their characteristics

3 Proposed Approaches

31 Overall Procedure In this section we present the algo-rithm of our work First of all we provide the flow ofour approach as Figure 3 After image segmentation thesperm head contour is extracted we transform the edge of

the sperm image into separated coordinate points and dividethe coordinate into two vectors

32 Proposed Algorithms For the classification we appliedtwo methods In the first the bilateral symmetrical functionof (119899) for the continuous situation is defined as follows

119878 (119899) = int1198621

1003817100381710038171003817100381711987511989911987510038171003817100381710038171003817

2

119889119904 minus int1198622

1003817100381710038171003817100381711987511989911987510038171003817100381710038171003817

2

119889119904 (1)

where1198621 denotes the arc on the left of the axis while1198622 is theright arc

The symmetry function is proposed as follows

Percentsym (119860119899 (119896)) = 1 minussum abs (119860 (119894) minus 119860 (2119896 minus 119894))

sum 2 lowast sqrt (119860 (119894) lowast 119860 (2119896 minus 119894))

(2)

119860119899(119896) denotes the corner point array while the 119896th point is

the smallest value of 119878(119899)In our secondmethod the geometric centre is calculated

We used the following equation to compute the horizontalcoordinate

119909mid =sum119899

119894=1119909119894

119899 (3)

where 119909mid represents the horizontal coordinate of the geo-metric centre 119909

119894represents the horizontal coordinate of each

point on the contour and 119899 represents the number of points

BioMed Research International 5

on the edgeThe vertical coordinate could be calculated in thesame way

To obtain the distance from edge to the centre we takeadvantage of the following format

119889 (119894) = radic(119909119894minus 119909mid)

2+ (119910119894minus 119910mid)

2 (4)

Considering the difficulty in choosing the starting pointwe had to avoid the problem Therefore we took rank algo-rithm [13] which ignores where to start into consideration inorder to classify the results The rank algorithm transformsa sequence of numbers 119909

1 1199092 1199093 119909

119899 into a sequence

composed of 1 and 0 The transform format is as follows

119868119899=

0 if 119909119899le 119909119899minus1

1 if 119909119899 gt 119909119899minus1

(5)

Therefore we achieved a binary system with 1 and 0We transformed 5 consecutive numbers in 119868

119899into decimal

numbers By computing the times each decimal numberappeared we ascertained the rank and the probability of theoriginal sequence Taking advantage of rank and probabilitywe calculated the similarity between two sequences Theformat is as follows

119863119898 (1198781 1198782) =sum2119898minus1

119896=0

10038161003816100381610038161198771 (119908119896) minus 1198772(119908119896)1003816100381610038161003816 1199011 (119908119896) 1199012 (119908119896)

(2119898 minus 1)sum2119898minus1

119896=01199011(119908119896) 1199012(119908119896)

(6)

By calculating the 119863119898 we distinguished the ten mostrepresentative sequences and took them as the criteria

Using the rank algorithm [13] we achieved anotherbinary system with 1 and 0 Then we transformed 8 con-secutive numbers in 119868119899 into decimal numbers and finallywe obtained the rank and the probability of the sequenceWith format (6) we obtained the similarity among thesequences Similar to the previous method we achievedanother criterion

To avoid the problem of a single criterion being toolopsided we took the gray level value of the sperm intoconsideration First we calculated the gray level value of allpoints within the sperm using the following format

119866 =

119899

sum1

119892119894 (7)

where 119866 represents the summation of the gray level value ofeach point119892119894 represents the gray level value of each point and119899 represents the total number of points within the sperm

Then we computed the gray percentage using the follow-ing format

119875 = 1 minus119866

119899 lowast 255 (8)

where 119875 represents the level of darkness

321 Joint Rank Difference and Gray Level Method By cal-culating the average rank difference of normal and abnormal

sperms we found the dividing line between them Thus therank difference itself can work as a judgment as to whetheror not the test sperm is normal We combined the two rankdifferences originating from the distance from centre to thecontour and the grey level value of the pixels in a sperm bycalculating the sum of the test spermrsquos distance average rankdifference and the average grey level value rank differencewhile each of them takes a certain weight The format is asfollows

119862 = 120572 lowast dARD + (1 minus 120572) lowast 255 lowast 119875 (9)

119862 represents the value that is the combination of the twostandards 120572 represents the weight of average rank differencedARD represents the average rank difference of distance fromthe centre to the contour

With the combination value we enhanced the originaljudgment by considering more elements and the importanceof each To find the dividing line we collected all thecombination values of the sperms and chose one of theiraverage values to provide the best accuracy for the dividingline

33 Fused SVM Method To achieve a better result we fusedthe SVM method as an advanced classifier SVM Methodsare supervised learning models with associated learningalgorithms that analyse data and recognize patterns and areused for classification and regression analysisThe basic SVMtakes a set of input sperm features and predicts for each giveninput the possible class form normal or abnormal making ita nonprobabilistic binary linear classifier

Given a training dataset (1199091 1199101) (1199092 1199102) (119909119899 119910119899)where 119909

119894isin R and 119910

119894is either 1 or minus1 indicating the class

to which the point 119909119894belongs let x = [119909

1 1199092 119909

119899]119879

The construction of the hyperplane for a linearly separableproblem is w119879x + 119887 = 0 where w is the normal vectorto the hyperplane and the parameter 119887w determines theoffset of the hyperplane from the origin along the normalvector w Thus the margin between the hyperplane and thenearest point is maximized and can be posed as the followingproblem

minw119887120585

1

2w119879w + 119862

119899

sum119894=1

120585119894

subject to 119910119894(119908119879119909119894+ 119887) ge 1 minus 120585

119894 119894 = 1 2 119899

120585119894ge 0

(10)

where 119862 is a user-defined constant as the penalty parameterof the error term

The SVM requires the optimal solution We use theLIBSVM[25] to solve this optimization problemwith the userguide given in [26]

As for the input of the SVM approach we fused the greylevel value and dARD as well First of all it extracts grey levelvalue of the pixels within the sperm head It starts from thecentre of the sperm and then it computes the following greylevel values in a square roundabout path The path map isshown in Table 1 as the sequence of numbers

6 BioMed Research International

Table 1 Grey level value extraction path

43 42 41 40 39 38 37 6444 21 20 19 18 17 36 6345 22 7 6 5 16 35 6246 23 8 1 4 15 34 6147 24 9 2 3 14 33 6048 25 10 11 12 13 32 5949 26 27 28 29 30 31 5850 51 52 53 54 55 56 57

Original sperm Sperm head

Figure 4 Extract sperm head from normal sperm

When the path collides with the contour the sequenceof grey level values takes advantage of the Rank algorithmin transforming 8 consecutive numbers 119868119899 into the decimalnumbers during the ends of extraction Then we achievedrank sequences and 119863119898 between every pair of images whichcan represent their similarities As a result we can tell whichimage represents the rest of the images the best by calculatingthe sum of the 119863

119898value from one image to all the other

images Thus we can put the rank sequences in order as wellWith the rank sequences we added the dARD sequences

at the end of corresponding rank sequence The newsequences consist of the training samples and test samplesThe ten training samples include five most representativesequences of normal sperms and five ones of abnormalsperms The rest sperms are the test samples

4 Fundamental Evaluation

In this fundamental evaluation we have undertaken variousexperiments on 80 normal and 80 abnormal sperms Theyare from a hospital the images are provided on httpscodegooglecompsupport-vector-machine-for-sperm-morphol-ogy-diagnosis as Supplementary material See Supplemen-tary Material available online at httpdxdoiorg1011552013687607 Before the classification of sperms thesegmentation is done by two steps (1) We pick the areawhere sperms do not overlay manually (2) We segment thesperms by reading the gray scale value of the picked areaWith knowing the location of the sperms we copy them fromthe area and build a new image to store each one of themWith a single sperm in an image it can cut off its tail easily

Figures 4 and 5 present an example of our segmentationresult

In this way we segmented all the 80 normal and 80abnormal sperm heads from 80 pieces of sperm images

41 Results of Our Waveform Extraction Methods In thissection we present part of the results of the steps in our

Original sperm Sperm head

Figure 5 Extract sperm head from abnormal sperm

experiments which explain how we made a choice andtheir performances In Table 2 the column Number lists thenumber of each imageThe columns Symmetry andMid showthe result of each transformation from image into waveform

Table 3 presents the waveform obtained by the firstmethod using four different starting points for the sameimage With the differences among four waves the problemarises from choosing one for the classification The resultof the second approach avoids the problem of picking thestarting point Thus we take the second result to classifysperms

42 Average Rank Difference In Figure 6 the horizontalaxis represents the serial number of each sperm image Thevertical axis represents the average difference of rank amongthe 80 normal or 80 abnormal sperms and the ten modelsperms As the result did not provide a clear dividing linebetween normal and abnormal sperms we adopted anothermethod to examine their differences We calculated thedifference in rank between the ten normal sperms and theother sperms Taking a look at Figure 6 if we take horizontalvalue 181 as the dividing line between normal and abnormalmeaning when the value of the average rank difference of asperm is below 181 we take the sperm as normal otherwisewe take it as abnormal In this case the accuracy could reach55

43 Grey Level Feature Then we focused on the influence ofthe grey level value Parts of the results are as shown in Figures7 and 8

In Figures 7 and 8 the horizontal axis represents the serialnumber of each sperm image In Figures 7(a) and 8(a) thevertical axis represents the average grey level value of thepoints of the sperms in the images In Figures 7(b) and 8(b)the vertical axis represents the grey level percentage of thesegmentation of the sperms

According to the percentage results we can ascertainthe percentage dividing line between normal and abnormalsperms to be 0268228785 With the division we can achieve725 accuracy

44 Joint Average Result We then tried to combine thetwo aspects as a judgment First of all we found the newparameter through format (7) By setting different 120572 wedetermined the different weight of each aspect The newparameter is as in Figure 9 It contains the new crite-ria produced with rank and gray level values of normaland abnormal sperms 120572 is the argument that represents

BioMed Research International 7

Table 2 Waveform extraction from both methods

Num Normal Symmetry Mid

1

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

25

20

15

10

5

0

2

0100200300400500600700800

1 9 17 25 33 41 49 57 65 73 81 89 970

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

3

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

4

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

5

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8502468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

6

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

7

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

8

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

9

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

8 BioMed Research International

Table 2 Continued

10

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

Num Abnormal Symmetry Mid

81

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

30

1 11 21 31 41 51 61 71 81 91 101

111

121

82

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

83

0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8505

101520253035

1 13 25 37 49 61 73 85 97 109

121

133

84

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 790

10203040506070

1 7 13 19 25 31 37 43 49 55 61 67 73 79

85

050

100150200250300350400

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

86

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

1012141618

1 7 13 19 25 31 37 43 49 55 61 67 73 79

87

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

88

0200400600800

10001200140016001800

1 7 13 19 25 31 37 43 49 55 61 67 73 7905

101520253035

1 7 13 19 25 31 37 43 49 55 61 67 73 79

BioMed Research International 9

Table 2 Continued

89

0200400600800

100012001400

1 7 13 19 25 31 37 43 49 55 61 67 73 790

5

10

15

20

25

30

1 7 13 19 25 31 37 43 49 55 61 67 73 79

90

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

Table 3 The waveform of different starting points from the first method

Startpoint Result

Start 1

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

Start 2

0

200

400

600

800

1000

1200

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800900

1000

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97

Table 4 Dividing line of C

120572 01 02 03 04 05 06 07 08 09DL 177 165 160 149 134 123 114 101 95

Table 5 Accuracy of different 120572

120572 01 02 03 04 05 06 07 08 09Acc 061 062 064 066 069 072 074 077 073

the weighted average rank difference taken in the criteriaWith the new parameter we can calculate its dividing line(DL) as a threshold

In Table 4 row DL represents the dividing line of each120572 value With the dividing line we can classify the spermsAssuming we have already known the exact classification ofeach sperm we can tell how many of them are correctlyclassified and thus we can come up with the accuracy Theaccuracy (Acc) of the different 120572 is as shown in Table 5

In Table 5 the row Acc represents the accuracy of each 120572value According to the result we chose the value 08 for 120572 Inthis case the accuracy can reach 76875

45 Fused SVM Result With all the data provided abovewe took advantage of the SVM method to undertake the

classification and used LIBSVM Software from httpwwwcsientuedutwsimcjlinlibsvm As for the parametersthe default values are used in the implementation and theSVM type is C-SVC the kernel function type is RBF function119890minus120574(119906minusV)2 = 1119896 cost = 1 cache size = 40 eps = 0001shrinking = 1 weight = 1

We input the training samples and test samples as men-tioned in Section 33 With ten training samples and testsamples of 160 sperms which consist of 80 normal and 80abnormal ones we obtain accuracy with 889

5 Comparison Evaluation

In this section we introduce the three methods we usedas comparisons to our method K-nearest neighbor methodworks quite effective as well the other two fail to satisfy us

51 Compared K-Nearest Neighbour Approach TheK-nearestneighbour method showed the following differences

We took the 160 images of sperms as the original clas-sifications and then we input the test image transformedall the images into gray and then calculated the differencebetween the test image and the original images Then wepicked the ten images with the smaller differences Aftercounting the classifications of the ten images we specified theone occurring most often to be the classification of the test

10 BioMed Research International

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a) Normal sperms

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(b) Abnormal sperms

Figure 6 Normal and abnormal sperms and their corresponding data

1 7 13 19 25 31 37 43 49 55 61 67 73 790

102030405060708090

100

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

(b)

Figure 7 Grey level value and percentage of normal sperms

0

20

40

60

80

100

120

140

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

07

(b)

Figure 8 Grey level value and percentage of abnormal sperms

image The classification of the top ten nearest neighbours isas in Figure 10

Taking the sperms whose sum value is larger than 3 asnormal and the sperms whose sum value is smaller than 4as abnormal we can tell that only 39 sperms are mistakenlyclassified providing accuracy of 75625

52 Compared SIFTApproach Lowe summed up the existingfeature detection method based on invariants technologyin 2004 and formally proposed an image scaling rotationand even affine transformation for invariant image with localfeature description operator based on scale space SIFT [22]

The SIFT algorithm first undertakes feature detection in scalespace and defines the key pointsrsquo positions and the scale ofthe key points and then it uses the main direction of theneighbourhood gradient of the key points as the directionfeatures of the points in order to achieve the operatorindependence of scale and the direction The format whichproduces the scale space is as follows

119871 (119909 119910 119905) = intinfin

120585=minusinfin

intinfin

120578

1

2120587119905119890minus(1205852+1205782)2119905

119891 (119909 minus 120585 119910 minus 120578) 119889120585 119889120578

(11)

BioMed Research International 11

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(a) 120572 = 01

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(b) 120572 = 02

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(c) 120572 = 03

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(d) 120572 = 04

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(e) 120572 = 05

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(f) 120572 = 06

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(g) 120572 = 07

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160

(h) 120572 = 08

020406080

100120140

1 13 25 37 49 61 73 85 97 109 121 133 145157

(i) 120572 = 09

Figure 9 The result of joint rank and gray level features of normal and abnormal sperms

12 BioMed Research International

1 7 13 19 25 31 37 43 49 55 61 67 73 790

2

4

6

8

10

12

(a) Normal sperms

1 7 13 19 25 31 37 43 49 55 61 67 73 790

051

152

253

354

45

(b) Abnormal sperms

Figure 10 The classification of the top ten neighbours of (a) normal and (b) abnormal sperms

Rectangularsegment

Segmentation afterrotation

Ellipse

Figure 11 The ellipse model to estimate the contour of the sperm

In format (11) t represents the scaling parameter Byundertaking convolution in the whole domain with a two-dimensionalGaussian kernel and input image we can achievescaling correspondent to t

The SIFT feature vector has the following features (a) it isthe local feature of an image which maintains invariance notonly on rotation scale and brightness variation but also onthe viewing angle the affine transformation and the noise(b) it is distinctive and informative and suitable for fastaccurate matching in a mass signature database (c) it canproduce a large number of SIFT feature vectors with fewobjects (d) it is of high speed

Matlab source code of SIFT is fromhttpwwwvlfeatorgindexhtml [27] We used the SIFT method to extract featurepoints of all the sperm images and achieved the ten normaland ten abnormal sperm images as in the training modelThen we calculated the difference between the trainingmodels and the other 70 normal and 70 abnormal spermimages With the dividing line we came up with accuracyof 50 The result turned out to be bad because the inputdata are simple sperm images and the SIFT algorithm cannotextract enough feature descriptors for classification

53 Compared Ellipse Model Approach With the image ofa single sperm we can use the ellipse model to estimatethe contour of the sperm First of all we used a rectangularsegment to cut out the sperm In order to make the rectangleclose to the sperm edge we needed to choose the rectanglewith the smallest area we rotated the sperm so that thesegmentation would be easier We calculated the left-mostupper-most downward-most and right-most points on theedges and built the rectangular segment based on them asdepicted in Figure 11

76888890

7563

5000

6625

000100020003000400050006000700080009000

10000

Jointaverage

K-neighbor SIFT Ellipticmodel

FusedSVM

()

Figure 12 Accuracy comparison of all methods

With the rectangle we achieved an ellipse to estimatethe sperm whose long axis radius is half the length of therectanglersquos long side and whose short axis radius is half thelength of the rectanglersquos short side As a result we ascertainedthe ovality which is the ratio of the length of the long axisto the short axis Through our experiment on 80 sperms wecould tell whether the sperm was too close to a circle whoseratio is close to 1 and whether the sperm was too slim whoseratio is close to 2 or even larger than 2 We took those whoseratio was larger than 12 and smaller than 18 as normal andthus we achieved accuracy of 6625

6 Conclusion

As shown in Figure 12 we have presented an approachfor the classification of sperm images with one-dimensionalwaveform and gray level features In the comprehensiveevaluation the joint average approach method was applied

BioMed Research International 13

to 160 sperm image samples and provided 76875 accuracyof judgment We then applied the fused SVM method tothe images and reached accuracy of 875 The result provesthat the proposed approach is superior to the previousapproaches

References

[1] N R Pal and S K Pal ldquoA review on image segmentationtechniquesrdquo Pattern Recognition vol 26 no 9 pp 1277ndash12941993

[2] A Yezzi L Zollei and T Kapur ldquoA variational frameworkfor integrating segmentation and registration through activecontoursrdquoMedical ImageAnalysis vol 7 no 2 pp 171ndash185 2003

[3] R Chrastek M Wolf K Donath et al ldquoAutomated segmenta-tion of the optic nerve head for diagnosis of glaucomardquoMedicalImage Analysis vol 9 no 4 pp 297ndash314 2005

[4] H W Kang and S Y Shin ldquoEnhanced lane interactive imagesegmentation by incremental path map constructionrdquo Graphi-cal Models vol 64 no 5 pp 282ndash303 2002

[5] J Peters O Ecabert C Meyer R Kneser and J Weese ldquoOpti-mizing boundary detection via Simulated Search with appli-cations to multi-modal heart segmentationrdquo Medical ImageAnalysis vol 14 no 1 pp 70ndash84 2010

[6] G Behiels F Maes D Vandermeulen and P Suetens ldquoEvalu-ation of image features and search strategies for segmentationof bone structures in radiographs using Active Shape ModelsrdquoMedical Image Analysis vol 6 no 1 pp 47ndash62 2002

[7] M Blanchard K Haguenoer A Apert et al ldquoSperm morphol-ogy assessment using Davidrsquos classiffication time to switch tostrict criteria Prospective comparative analysis in a selectedIVF populationrdquo International Journal of Andrology vol 34 no2 pp 145ndash152 2010

[8] MKass AWitkin andD Terzopoulos ldquoSnakes active contourmodelsrdquo International Journal of Computer Vision vol 1 no 4pp 321ndash331 1988

[9] TMcInerney andD Terzopoulos ldquoT-snakes topology adaptivesnakesrdquoMedical Image Analysis vol 4 no 2 pp 73ndash91 2000

[10] J Liang T McInerney and D Terzopoulos ldquoUnited SnakesrdquoMedical Image Analysis vol 10 no 2 pp 215ndash233 2006

[11] M Zhao Z Meng K K Tseng J S Pan and C Y Hsu ldquoSym-metry auto-detection based on contour and corner modelsrdquo inProceedings of the 5th International Conference on Genetic andEvolutionary Computing (ICGEC rsquo11) pp 345ndash349 September2011

[12] Y Mingqiang K Kidiyo and R Joseph ldquoA survey of shapefeature extraction techniquesrdquo Pattern Recognition vol 15 no7 pp 43ndash90 2008

[13] Y Wang and H Huang ldquoAnalysis of Human Heartbeat withEKG Signalsrdquo 2010

[14] Q Hua A Ji and Q He ldquoMultiple real-valued K nearestneighbor classifiers system by feature groupingrdquo in Proceedingsof the IEEE International Conference on Systems Man andCybernetics (SMC rsquo10) pp 3922ndash3925 October 2010

[15] P Viswanath and T Hitendra Sarma ldquoAn improvement to k-nearest neighbor classifierrdquo in Proceedings of the IEEE RecentAdvances in Intelligent Computational Systems (RAICS rsquo11) pp227ndash231 September 2011

[16] P Wu and Q Chen ldquoA novel SVM-based edge detectionmethodrdquo Physics Procedia vol 24 pp 2075ndash2082 2012

[17] D N Sotiropoulos and G A Tsihrintzis ldquoArtificial immunesystem-based classification in class-imbalanced image prob-lemsrdquo in Proceedings of the 8th International Conference onIntelligent InformationHiding andMultimedia Signal Processingpp 138ndash141 July 2012

[18] M Ramon FMartınez-Pastor O Garcıa-Alvarez et al ldquoTakingadvantage of the use of supervised learning methods forcharacterization of sperm population structure related withfreezability in the Iberian red deerrdquoTheriogenology vol 77 no8 pp 1661ndash1672 2012

[19] S G Goodson Z Zhang J K Tsuruta W Wang and DA OrsquoBrien ldquoClassification of mouse sperm motility patternsusing an automatedmulticlass support vectormachinesmodelrdquoBiology of Reproduction vol 84 no 6 pp 1207ndash1215 2011

[20] D T Lin and D C Pan ldquoIntegrating a mixed-feature modeland multiclass support vector machine for facial expressionrecognitionrdquo Integrated Computer-Aided Engineering vol 16no 1 pp 61ndash74 2009

[21] C G Cheng Y M Tian and W Y Jin ldquoA study on the earlydetection of colon cancer using the methods of wavelet featureextraction and SVM classifications of FTIRrdquo Spectroscopy vol22 no 5 pp 397ndash404 2008

[22] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004

[23] V R Nafisi M H Moradi and M H Nasr-Esfahani ldquoSpermidentification using elliptic model and tail detectionrdquo WorldAcademy of Science Engineering and Technology vol 6 pp 205ndash208 2005

[24] W J Yi K S Park and J S Paick ldquoParameterized charac-terization of elliptic sperm heads using Fourier representationand wavelet transformrdquo in Proceedings of the 20th AnnualInternational Conference of the IEEE Engineering in Medicineand Biology Society vol 20 pp 974ndash977 1998

[25] C C Chang andC J Lin LIBSVMmdasha library for support vectormachines Version 312 httpwwwcsientuedutwsimcjlinlibsvm

[26] CW Hsu C C Chang and C J Lin ldquoA practical guide to sup-port vector classificationrdquo 2010 httpwwwcsientuedutwsimcjlin

[27] A Vedaldi and B Fulkerson ldquoVLFeat an open and portablelibrary of computer vision algorithmsrdquo 2008 httpwwwvlfeatorg

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

BioMed Research International 3

morphology assessment using Davidrsquos classificationrdquo [7]Thispaper aimed to compare assessment of spermmorphology byusing Davidrsquos classification (DC) based on the strict criteria(computer-assisted sperm analysis (CASA) SC) for theirability to predict fertilization in a selected in vitro fertilization(IVF) population Their results showed that the DC spermmorphology analysis was less indicative of fertilization thanCASA SC

After the DC analysis we reviewed the paper on thetechnology of extracting objectsrsquo edges The first article [8]ldquoSnakesrdquo is an energy-minimizing spline guided by externalconstraint forces and influenced by image forces that pull ittowards features such as lines and edges Snakes are activecontour models they lock onto nearby edges localizingthem accurately Scale-space continuation can be used toenlarge the capture region surrounding a feature Snakesprovide a unified account of a number of visual problemsincluding detection of edges lines and subjective contoursmotion tracking and stereo matching We have used snakessuccessfully for interactive interpretation in which user-imposed constraint forces guide the snake near features ofinterest

As the snake may not do its job well enough for currentresearch we reviewed some improved algorithms such asT-snakes [9] topology adaptive snakes In this paper theypresent a new class of deformable contours (snakes) and areapplied to the segmentation of medical images They enabletopological flexibility among other features The resultingtopology adaptive snakes or ldquoT-snakesrdquo can be used to seg-ment some of the most complex-shaped biological structuresfrom medical images in an efficient and highly automatedmanner

Moreover other authors [10] present a framework calledunited snakes which has two key features First it unifies themost popular snake variants expanding the range of objectmodelling capabilities Second it embodies the idea of thetechnique known as live wire or intelligent scissors The twotechniques can be combined advantageously by introducingan effective hard constraint mechanism They apply unitedsnakes to several different medical image analysis tasksdemonstrating the generality accuracy and robustness of thetool

23 Shape Descriptor The third topic the most related workis called shape descriptor We focused on how to transformit into a one-dimensional feature To achieve the goal westudied further related articles The first paper [11] presentsa new symmetry autodetection approach The symmetry canbe detected automatically by using corner detection Duringthe process the contour can be transferred to a waveform

For additional study of the shape descriptor paper [12]focused on presenting the existing approaches of shape-basedfeature extraction Paper [13] introduced the extraction ofwaveform features by reduced binary features used to reducecomplexity and storage

24 Classification Another common image matchingapproach the K-nearest neighbour method as a compared

target proposes a method to fuse real-value K-nearestneighbour classifiers by feature grouping [14] The real-valueK-nearest neighbour classifier can approximate continuous-valued target functions In addition it is sensitive to featureperturbation Therefore when the multiple real-value K-nearest neighbour classifier is fused by feature groupingthe performance of the fusion will be better than thesingle classifier Another K-nearest neighbour method [15]presents a novel improvement to the K-nearest neighbourmean classifier (K-NNMC) K-NNMC finds the K-nearestneighbours for each class of training patterns separately andfinds the means for each of these K-neighbours (class-wise)Classification is undertaken according to the nearest meanpattern In experiments using several standard datasets ithas been shown that the proposed classifier provides betterclassification accuracy over the conventional K-nearestneighbour method and thus it is a suitable method to beused in data mining applications

As we have been using the SVM as an advanced methodto improve the performance of our approach in order toachieve a deep understanding of SVM we paid attention torelated SVM research Research work [16] presents a newvalid edge detection algorithm based on an SVM to avoid thedisadvantages of traditional image edge detection methodsIn another SVM related work [17] the authors compared theperformance of artificial-immune-system- (AIS) based algo-rithms to a Gaussian kernel-based SVMTheir experimenta-tion indicates that the AIS-based classification paradigm hasthe intrinsic property of dealing more efficiently with highlyskewed datasets In addition research [18] takes advantage ofSVM to characterize the sperm population structure relatedto freezability The SVM was generated using sperm motilityinformation captured by CASA from thawed semen ThisSVM method was used to characterize the motile spermsubpopulations for Iberian red deer

Research [19] provided more information on the effect ofSVM on sperm research An automated quantitative methodthat objectively classifies five distinct motility patterns ofmouse sperm using the SVM method was developed Itsparameters are associated with the classified tracks and wereincorporated into established SVM algorithms to generate aseries of equations These equations were integrated into abinary decision tree that sequentially sorts uncharacterizedtracks into distinct categories

Once we had finished reading about the sperm relatedSVM wemoved on to research [20] which reveals the advan-tages of the proposed mixed-feature model and presentsthe capability of identifying human facial expressions fromstatic images The subsequent framework is a multistagediscrimination model based on global appearance featuresextracted from two-dimensional principal component anal-ysis (2DPCA) and local texture represented by a local binarypattern (LBP) The experimental results indicate that theproposed mixed-feature model is feasible and outperformsthe single-feature model

We then tried to look for research on feature extractionand the SVM classifier [21] This paper introduces a newmethod for the early detection of colon cancer using a com-bination of feature extraction based on wavelets for Fourier

4 BioMed Research International

Original image

Segment sperm head

Extract contour

Build wave

End

SVM classification

Calculate graylevel value

Calculate gray()

Calculatecombined value

GLV GLV

0

10

12

14

16

18

20

0 10 20 30 40 50 60 70 80 90

35 40 45 50 55 60 65 70

7580859095

100105110115120

() ()

Number

Number Number

Number

1

2

97

152

11

12

114

113

1

2

038088547

040800834

11

12

044727911

044294079

Figure 3 The flow chart of the sperm classification system

transform infrared spectroscopy (FTIR) and classificationwith SVM

One popular robust image-matching approach is SIFT Inpaper [22] it performs a reliable matching between differentviews of an object or scene Its features are invariant to imagescale and rotation In that sense its images can be matchedwith high reliability against a large database of features frommany images

With regard to the elliptic model it tried to estimate thecontour of a sperm by an ellipse shape Research work [23]used an ellipse to classify sperm Another article [24] pro-posed a new method of sperm morphological classificationusing the elliptic shape parameterized by the discrete Fouriertransform and reconstructed with dyadic data points Theenclosed area of boundaries as a classification feature wascalculated and transformed by wavelet transform

Although some researchers focused on sperm classifica-tion none has used our approach for sperm imaging Thispaper presents a system and a novel approach which uses aone-dimensional contour and gray level features to diagnosedifferent sperms according to their characteristics

3 Proposed Approaches

31 Overall Procedure In this section we present the algo-rithm of our work First of all we provide the flow ofour approach as Figure 3 After image segmentation thesperm head contour is extracted we transform the edge of

the sperm image into separated coordinate points and dividethe coordinate into two vectors

32 Proposed Algorithms For the classification we appliedtwo methods In the first the bilateral symmetrical functionof (119899) for the continuous situation is defined as follows

119878 (119899) = int1198621

1003817100381710038171003817100381711987511989911987510038171003817100381710038171003817

2

119889119904 minus int1198622

1003817100381710038171003817100381711987511989911987510038171003817100381710038171003817

2

119889119904 (1)

where1198621 denotes the arc on the left of the axis while1198622 is theright arc

The symmetry function is proposed as follows

Percentsym (119860119899 (119896)) = 1 minussum abs (119860 (119894) minus 119860 (2119896 minus 119894))

sum 2 lowast sqrt (119860 (119894) lowast 119860 (2119896 minus 119894))

(2)

119860119899(119896) denotes the corner point array while the 119896th point is

the smallest value of 119878(119899)In our secondmethod the geometric centre is calculated

We used the following equation to compute the horizontalcoordinate

119909mid =sum119899

119894=1119909119894

119899 (3)

where 119909mid represents the horizontal coordinate of the geo-metric centre 119909

119894represents the horizontal coordinate of each

point on the contour and 119899 represents the number of points

BioMed Research International 5

on the edgeThe vertical coordinate could be calculated in thesame way

To obtain the distance from edge to the centre we takeadvantage of the following format

119889 (119894) = radic(119909119894minus 119909mid)

2+ (119910119894minus 119910mid)

2 (4)

Considering the difficulty in choosing the starting pointwe had to avoid the problem Therefore we took rank algo-rithm [13] which ignores where to start into consideration inorder to classify the results The rank algorithm transformsa sequence of numbers 119909

1 1199092 1199093 119909

119899 into a sequence

composed of 1 and 0 The transform format is as follows

119868119899=

0 if 119909119899le 119909119899minus1

1 if 119909119899 gt 119909119899minus1

(5)

Therefore we achieved a binary system with 1 and 0We transformed 5 consecutive numbers in 119868

119899into decimal

numbers By computing the times each decimal numberappeared we ascertained the rank and the probability of theoriginal sequence Taking advantage of rank and probabilitywe calculated the similarity between two sequences Theformat is as follows

119863119898 (1198781 1198782) =sum2119898minus1

119896=0

10038161003816100381610038161198771 (119908119896) minus 1198772(119908119896)1003816100381610038161003816 1199011 (119908119896) 1199012 (119908119896)

(2119898 minus 1)sum2119898minus1

119896=01199011(119908119896) 1199012(119908119896)

(6)

By calculating the 119863119898 we distinguished the ten mostrepresentative sequences and took them as the criteria

Using the rank algorithm [13] we achieved anotherbinary system with 1 and 0 Then we transformed 8 con-secutive numbers in 119868119899 into decimal numbers and finallywe obtained the rank and the probability of the sequenceWith format (6) we obtained the similarity among thesequences Similar to the previous method we achievedanother criterion

To avoid the problem of a single criterion being toolopsided we took the gray level value of the sperm intoconsideration First we calculated the gray level value of allpoints within the sperm using the following format

119866 =

119899

sum1

119892119894 (7)

where 119866 represents the summation of the gray level value ofeach point119892119894 represents the gray level value of each point and119899 represents the total number of points within the sperm

Then we computed the gray percentage using the follow-ing format

119875 = 1 minus119866

119899 lowast 255 (8)

where 119875 represents the level of darkness

321 Joint Rank Difference and Gray Level Method By cal-culating the average rank difference of normal and abnormal

sperms we found the dividing line between them Thus therank difference itself can work as a judgment as to whetheror not the test sperm is normal We combined the two rankdifferences originating from the distance from centre to thecontour and the grey level value of the pixels in a sperm bycalculating the sum of the test spermrsquos distance average rankdifference and the average grey level value rank differencewhile each of them takes a certain weight The format is asfollows

119862 = 120572 lowast dARD + (1 minus 120572) lowast 255 lowast 119875 (9)

119862 represents the value that is the combination of the twostandards 120572 represents the weight of average rank differencedARD represents the average rank difference of distance fromthe centre to the contour

With the combination value we enhanced the originaljudgment by considering more elements and the importanceof each To find the dividing line we collected all thecombination values of the sperms and chose one of theiraverage values to provide the best accuracy for the dividingline

33 Fused SVM Method To achieve a better result we fusedthe SVM method as an advanced classifier SVM Methodsare supervised learning models with associated learningalgorithms that analyse data and recognize patterns and areused for classification and regression analysisThe basic SVMtakes a set of input sperm features and predicts for each giveninput the possible class form normal or abnormal making ita nonprobabilistic binary linear classifier

Given a training dataset (1199091 1199101) (1199092 1199102) (119909119899 119910119899)where 119909

119894isin R and 119910

119894is either 1 or minus1 indicating the class

to which the point 119909119894belongs let x = [119909

1 1199092 119909

119899]119879

The construction of the hyperplane for a linearly separableproblem is w119879x + 119887 = 0 where w is the normal vectorto the hyperplane and the parameter 119887w determines theoffset of the hyperplane from the origin along the normalvector w Thus the margin between the hyperplane and thenearest point is maximized and can be posed as the followingproblem

minw119887120585

1

2w119879w + 119862

119899

sum119894=1

120585119894

subject to 119910119894(119908119879119909119894+ 119887) ge 1 minus 120585

119894 119894 = 1 2 119899

120585119894ge 0

(10)

where 119862 is a user-defined constant as the penalty parameterof the error term

The SVM requires the optimal solution We use theLIBSVM[25] to solve this optimization problemwith the userguide given in [26]

As for the input of the SVM approach we fused the greylevel value and dARD as well First of all it extracts grey levelvalue of the pixels within the sperm head It starts from thecentre of the sperm and then it computes the following greylevel values in a square roundabout path The path map isshown in Table 1 as the sequence of numbers

6 BioMed Research International

Table 1 Grey level value extraction path

43 42 41 40 39 38 37 6444 21 20 19 18 17 36 6345 22 7 6 5 16 35 6246 23 8 1 4 15 34 6147 24 9 2 3 14 33 6048 25 10 11 12 13 32 5949 26 27 28 29 30 31 5850 51 52 53 54 55 56 57

Original sperm Sperm head

Figure 4 Extract sperm head from normal sperm

When the path collides with the contour the sequenceof grey level values takes advantage of the Rank algorithmin transforming 8 consecutive numbers 119868119899 into the decimalnumbers during the ends of extraction Then we achievedrank sequences and 119863119898 between every pair of images whichcan represent their similarities As a result we can tell whichimage represents the rest of the images the best by calculatingthe sum of the 119863

119898value from one image to all the other

images Thus we can put the rank sequences in order as wellWith the rank sequences we added the dARD sequences

at the end of corresponding rank sequence The newsequences consist of the training samples and test samplesThe ten training samples include five most representativesequences of normal sperms and five ones of abnormalsperms The rest sperms are the test samples

4 Fundamental Evaluation

In this fundamental evaluation we have undertaken variousexperiments on 80 normal and 80 abnormal sperms Theyare from a hospital the images are provided on httpscodegooglecompsupport-vector-machine-for-sperm-morphol-ogy-diagnosis as Supplementary material See Supplemen-tary Material available online at httpdxdoiorg1011552013687607 Before the classification of sperms thesegmentation is done by two steps (1) We pick the areawhere sperms do not overlay manually (2) We segment thesperms by reading the gray scale value of the picked areaWith knowing the location of the sperms we copy them fromthe area and build a new image to store each one of themWith a single sperm in an image it can cut off its tail easily

Figures 4 and 5 present an example of our segmentationresult

In this way we segmented all the 80 normal and 80abnormal sperm heads from 80 pieces of sperm images

41 Results of Our Waveform Extraction Methods In thissection we present part of the results of the steps in our

Original sperm Sperm head

Figure 5 Extract sperm head from abnormal sperm

experiments which explain how we made a choice andtheir performances In Table 2 the column Number lists thenumber of each imageThe columns Symmetry andMid showthe result of each transformation from image into waveform

Table 3 presents the waveform obtained by the firstmethod using four different starting points for the sameimage With the differences among four waves the problemarises from choosing one for the classification The resultof the second approach avoids the problem of picking thestarting point Thus we take the second result to classifysperms

42 Average Rank Difference In Figure 6 the horizontalaxis represents the serial number of each sperm image Thevertical axis represents the average difference of rank amongthe 80 normal or 80 abnormal sperms and the ten modelsperms As the result did not provide a clear dividing linebetween normal and abnormal sperms we adopted anothermethod to examine their differences We calculated thedifference in rank between the ten normal sperms and theother sperms Taking a look at Figure 6 if we take horizontalvalue 181 as the dividing line between normal and abnormalmeaning when the value of the average rank difference of asperm is below 181 we take the sperm as normal otherwisewe take it as abnormal In this case the accuracy could reach55

43 Grey Level Feature Then we focused on the influence ofthe grey level value Parts of the results are as shown in Figures7 and 8

In Figures 7 and 8 the horizontal axis represents the serialnumber of each sperm image In Figures 7(a) and 8(a) thevertical axis represents the average grey level value of thepoints of the sperms in the images In Figures 7(b) and 8(b)the vertical axis represents the grey level percentage of thesegmentation of the sperms

According to the percentage results we can ascertainthe percentage dividing line between normal and abnormalsperms to be 0268228785 With the division we can achieve725 accuracy

44 Joint Average Result We then tried to combine thetwo aspects as a judgment First of all we found the newparameter through format (7) By setting different 120572 wedetermined the different weight of each aspect The newparameter is as in Figure 9 It contains the new crite-ria produced with rank and gray level values of normaland abnormal sperms 120572 is the argument that represents

BioMed Research International 7

Table 2 Waveform extraction from both methods

Num Normal Symmetry Mid

1

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

25

20

15

10

5

0

2

0100200300400500600700800

1 9 17 25 33 41 49 57 65 73 81 89 970

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

3

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

4

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

5

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8502468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

6

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

7

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

8

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

9

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

8 BioMed Research International

Table 2 Continued

10

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

Num Abnormal Symmetry Mid

81

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

30

1 11 21 31 41 51 61 71 81 91 101

111

121

82

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

83

0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8505

101520253035

1 13 25 37 49 61 73 85 97 109

121

133

84

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 790

10203040506070

1 7 13 19 25 31 37 43 49 55 61 67 73 79

85

050

100150200250300350400

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

86

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

1012141618

1 7 13 19 25 31 37 43 49 55 61 67 73 79

87

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

88

0200400600800

10001200140016001800

1 7 13 19 25 31 37 43 49 55 61 67 73 7905

101520253035

1 7 13 19 25 31 37 43 49 55 61 67 73 79

BioMed Research International 9

Table 2 Continued

89

0200400600800

100012001400

1 7 13 19 25 31 37 43 49 55 61 67 73 790

5

10

15

20

25

30

1 7 13 19 25 31 37 43 49 55 61 67 73 79

90

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

Table 3 The waveform of different starting points from the first method

Startpoint Result

Start 1

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

Start 2

0

200

400

600

800

1000

1200

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800900

1000

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97

Table 4 Dividing line of C

120572 01 02 03 04 05 06 07 08 09DL 177 165 160 149 134 123 114 101 95

Table 5 Accuracy of different 120572

120572 01 02 03 04 05 06 07 08 09Acc 061 062 064 066 069 072 074 077 073

the weighted average rank difference taken in the criteriaWith the new parameter we can calculate its dividing line(DL) as a threshold

In Table 4 row DL represents the dividing line of each120572 value With the dividing line we can classify the spermsAssuming we have already known the exact classification ofeach sperm we can tell how many of them are correctlyclassified and thus we can come up with the accuracy Theaccuracy (Acc) of the different 120572 is as shown in Table 5

In Table 5 the row Acc represents the accuracy of each 120572value According to the result we chose the value 08 for 120572 Inthis case the accuracy can reach 76875

45 Fused SVM Result With all the data provided abovewe took advantage of the SVM method to undertake the

classification and used LIBSVM Software from httpwwwcsientuedutwsimcjlinlibsvm As for the parametersthe default values are used in the implementation and theSVM type is C-SVC the kernel function type is RBF function119890minus120574(119906minusV)2 = 1119896 cost = 1 cache size = 40 eps = 0001shrinking = 1 weight = 1

We input the training samples and test samples as men-tioned in Section 33 With ten training samples and testsamples of 160 sperms which consist of 80 normal and 80abnormal ones we obtain accuracy with 889

5 Comparison Evaluation

In this section we introduce the three methods we usedas comparisons to our method K-nearest neighbor methodworks quite effective as well the other two fail to satisfy us

51 Compared K-Nearest Neighbour Approach TheK-nearestneighbour method showed the following differences

We took the 160 images of sperms as the original clas-sifications and then we input the test image transformedall the images into gray and then calculated the differencebetween the test image and the original images Then wepicked the ten images with the smaller differences Aftercounting the classifications of the ten images we specified theone occurring most often to be the classification of the test

10 BioMed Research International

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a) Normal sperms

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(b) Abnormal sperms

Figure 6 Normal and abnormal sperms and their corresponding data

1 7 13 19 25 31 37 43 49 55 61 67 73 790

102030405060708090

100

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

(b)

Figure 7 Grey level value and percentage of normal sperms

0

20

40

60

80

100

120

140

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

07

(b)

Figure 8 Grey level value and percentage of abnormal sperms

image The classification of the top ten nearest neighbours isas in Figure 10

Taking the sperms whose sum value is larger than 3 asnormal and the sperms whose sum value is smaller than 4as abnormal we can tell that only 39 sperms are mistakenlyclassified providing accuracy of 75625

52 Compared SIFTApproach Lowe summed up the existingfeature detection method based on invariants technologyin 2004 and formally proposed an image scaling rotationand even affine transformation for invariant image with localfeature description operator based on scale space SIFT [22]

The SIFT algorithm first undertakes feature detection in scalespace and defines the key pointsrsquo positions and the scale ofthe key points and then it uses the main direction of theneighbourhood gradient of the key points as the directionfeatures of the points in order to achieve the operatorindependence of scale and the direction The format whichproduces the scale space is as follows

119871 (119909 119910 119905) = intinfin

120585=minusinfin

intinfin

120578

1

2120587119905119890minus(1205852+1205782)2119905

119891 (119909 minus 120585 119910 minus 120578) 119889120585 119889120578

(11)

BioMed Research International 11

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(a) 120572 = 01

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(b) 120572 = 02

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(c) 120572 = 03

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(d) 120572 = 04

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(e) 120572 = 05

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(f) 120572 = 06

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(g) 120572 = 07

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160

(h) 120572 = 08

020406080

100120140

1 13 25 37 49 61 73 85 97 109 121 133 145157

(i) 120572 = 09

Figure 9 The result of joint rank and gray level features of normal and abnormal sperms

12 BioMed Research International

1 7 13 19 25 31 37 43 49 55 61 67 73 790

2

4

6

8

10

12

(a) Normal sperms

1 7 13 19 25 31 37 43 49 55 61 67 73 790

051

152

253

354

45

(b) Abnormal sperms

Figure 10 The classification of the top ten neighbours of (a) normal and (b) abnormal sperms

Rectangularsegment

Segmentation afterrotation

Ellipse

Figure 11 The ellipse model to estimate the contour of the sperm

In format (11) t represents the scaling parameter Byundertaking convolution in the whole domain with a two-dimensionalGaussian kernel and input image we can achievescaling correspondent to t

The SIFT feature vector has the following features (a) it isthe local feature of an image which maintains invariance notonly on rotation scale and brightness variation but also onthe viewing angle the affine transformation and the noise(b) it is distinctive and informative and suitable for fastaccurate matching in a mass signature database (c) it canproduce a large number of SIFT feature vectors with fewobjects (d) it is of high speed

Matlab source code of SIFT is fromhttpwwwvlfeatorgindexhtml [27] We used the SIFT method to extract featurepoints of all the sperm images and achieved the ten normaland ten abnormal sperm images as in the training modelThen we calculated the difference between the trainingmodels and the other 70 normal and 70 abnormal spermimages With the dividing line we came up with accuracyof 50 The result turned out to be bad because the inputdata are simple sperm images and the SIFT algorithm cannotextract enough feature descriptors for classification

53 Compared Ellipse Model Approach With the image ofa single sperm we can use the ellipse model to estimatethe contour of the sperm First of all we used a rectangularsegment to cut out the sperm In order to make the rectangleclose to the sperm edge we needed to choose the rectanglewith the smallest area we rotated the sperm so that thesegmentation would be easier We calculated the left-mostupper-most downward-most and right-most points on theedges and built the rectangular segment based on them asdepicted in Figure 11

76888890

7563

5000

6625

000100020003000400050006000700080009000

10000

Jointaverage

K-neighbor SIFT Ellipticmodel

FusedSVM

()

Figure 12 Accuracy comparison of all methods

With the rectangle we achieved an ellipse to estimatethe sperm whose long axis radius is half the length of therectanglersquos long side and whose short axis radius is half thelength of the rectanglersquos short side As a result we ascertainedthe ovality which is the ratio of the length of the long axisto the short axis Through our experiment on 80 sperms wecould tell whether the sperm was too close to a circle whoseratio is close to 1 and whether the sperm was too slim whoseratio is close to 2 or even larger than 2 We took those whoseratio was larger than 12 and smaller than 18 as normal andthus we achieved accuracy of 6625

6 Conclusion

As shown in Figure 12 we have presented an approachfor the classification of sperm images with one-dimensionalwaveform and gray level features In the comprehensiveevaluation the joint average approach method was applied

BioMed Research International 13

to 160 sperm image samples and provided 76875 accuracyof judgment We then applied the fused SVM method tothe images and reached accuracy of 875 The result provesthat the proposed approach is superior to the previousapproaches

References

[1] N R Pal and S K Pal ldquoA review on image segmentationtechniquesrdquo Pattern Recognition vol 26 no 9 pp 1277ndash12941993

[2] A Yezzi L Zollei and T Kapur ldquoA variational frameworkfor integrating segmentation and registration through activecontoursrdquoMedical ImageAnalysis vol 7 no 2 pp 171ndash185 2003

[3] R Chrastek M Wolf K Donath et al ldquoAutomated segmenta-tion of the optic nerve head for diagnosis of glaucomardquoMedicalImage Analysis vol 9 no 4 pp 297ndash314 2005

[4] H W Kang and S Y Shin ldquoEnhanced lane interactive imagesegmentation by incremental path map constructionrdquo Graphi-cal Models vol 64 no 5 pp 282ndash303 2002

[5] J Peters O Ecabert C Meyer R Kneser and J Weese ldquoOpti-mizing boundary detection via Simulated Search with appli-cations to multi-modal heart segmentationrdquo Medical ImageAnalysis vol 14 no 1 pp 70ndash84 2010

[6] G Behiels F Maes D Vandermeulen and P Suetens ldquoEvalu-ation of image features and search strategies for segmentationof bone structures in radiographs using Active Shape ModelsrdquoMedical Image Analysis vol 6 no 1 pp 47ndash62 2002

[7] M Blanchard K Haguenoer A Apert et al ldquoSperm morphol-ogy assessment using Davidrsquos classiffication time to switch tostrict criteria Prospective comparative analysis in a selectedIVF populationrdquo International Journal of Andrology vol 34 no2 pp 145ndash152 2010

[8] MKass AWitkin andD Terzopoulos ldquoSnakes active contourmodelsrdquo International Journal of Computer Vision vol 1 no 4pp 321ndash331 1988

[9] TMcInerney andD Terzopoulos ldquoT-snakes topology adaptivesnakesrdquoMedical Image Analysis vol 4 no 2 pp 73ndash91 2000

[10] J Liang T McInerney and D Terzopoulos ldquoUnited SnakesrdquoMedical Image Analysis vol 10 no 2 pp 215ndash233 2006

[11] M Zhao Z Meng K K Tseng J S Pan and C Y Hsu ldquoSym-metry auto-detection based on contour and corner modelsrdquo inProceedings of the 5th International Conference on Genetic andEvolutionary Computing (ICGEC rsquo11) pp 345ndash349 September2011

[12] Y Mingqiang K Kidiyo and R Joseph ldquoA survey of shapefeature extraction techniquesrdquo Pattern Recognition vol 15 no7 pp 43ndash90 2008

[13] Y Wang and H Huang ldquoAnalysis of Human Heartbeat withEKG Signalsrdquo 2010

[14] Q Hua A Ji and Q He ldquoMultiple real-valued K nearestneighbor classifiers system by feature groupingrdquo in Proceedingsof the IEEE International Conference on Systems Man andCybernetics (SMC rsquo10) pp 3922ndash3925 October 2010

[15] P Viswanath and T Hitendra Sarma ldquoAn improvement to k-nearest neighbor classifierrdquo in Proceedings of the IEEE RecentAdvances in Intelligent Computational Systems (RAICS rsquo11) pp227ndash231 September 2011

[16] P Wu and Q Chen ldquoA novel SVM-based edge detectionmethodrdquo Physics Procedia vol 24 pp 2075ndash2082 2012

[17] D N Sotiropoulos and G A Tsihrintzis ldquoArtificial immunesystem-based classification in class-imbalanced image prob-lemsrdquo in Proceedings of the 8th International Conference onIntelligent InformationHiding andMultimedia Signal Processingpp 138ndash141 July 2012

[18] M Ramon FMartınez-Pastor O Garcıa-Alvarez et al ldquoTakingadvantage of the use of supervised learning methods forcharacterization of sperm population structure related withfreezability in the Iberian red deerrdquoTheriogenology vol 77 no8 pp 1661ndash1672 2012

[19] S G Goodson Z Zhang J K Tsuruta W Wang and DA OrsquoBrien ldquoClassification of mouse sperm motility patternsusing an automatedmulticlass support vectormachinesmodelrdquoBiology of Reproduction vol 84 no 6 pp 1207ndash1215 2011

[20] D T Lin and D C Pan ldquoIntegrating a mixed-feature modeland multiclass support vector machine for facial expressionrecognitionrdquo Integrated Computer-Aided Engineering vol 16no 1 pp 61ndash74 2009

[21] C G Cheng Y M Tian and W Y Jin ldquoA study on the earlydetection of colon cancer using the methods of wavelet featureextraction and SVM classifications of FTIRrdquo Spectroscopy vol22 no 5 pp 397ndash404 2008

[22] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004

[23] V R Nafisi M H Moradi and M H Nasr-Esfahani ldquoSpermidentification using elliptic model and tail detectionrdquo WorldAcademy of Science Engineering and Technology vol 6 pp 205ndash208 2005

[24] W J Yi K S Park and J S Paick ldquoParameterized charac-terization of elliptic sperm heads using Fourier representationand wavelet transformrdquo in Proceedings of the 20th AnnualInternational Conference of the IEEE Engineering in Medicineand Biology Society vol 20 pp 974ndash977 1998

[25] C C Chang andC J Lin LIBSVMmdasha library for support vectormachines Version 312 httpwwwcsientuedutwsimcjlinlibsvm

[26] CW Hsu C C Chang and C J Lin ldquoA practical guide to sup-port vector classificationrdquo 2010 httpwwwcsientuedutwsimcjlin

[27] A Vedaldi and B Fulkerson ldquoVLFeat an open and portablelibrary of computer vision algorithmsrdquo 2008 httpwwwvlfeatorg

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

4 BioMed Research International

Original image

Segment sperm head

Extract contour

Build wave

End

SVM classification

Calculate graylevel value

Calculate gray()

Calculatecombined value

GLV GLV

0

10

12

14

16

18

20

0 10 20 30 40 50 60 70 80 90

35 40 45 50 55 60 65 70

7580859095

100105110115120

() ()

Number

Number Number

Number

1

2

97

152

11

12

114

113

1

2

038088547

040800834

11

12

044727911

044294079

Figure 3 The flow chart of the sperm classification system

transform infrared spectroscopy (FTIR) and classificationwith SVM

One popular robust image-matching approach is SIFT Inpaper [22] it performs a reliable matching between differentviews of an object or scene Its features are invariant to imagescale and rotation In that sense its images can be matchedwith high reliability against a large database of features frommany images

With regard to the elliptic model it tried to estimate thecontour of a sperm by an ellipse shape Research work [23]used an ellipse to classify sperm Another article [24] pro-posed a new method of sperm morphological classificationusing the elliptic shape parameterized by the discrete Fouriertransform and reconstructed with dyadic data points Theenclosed area of boundaries as a classification feature wascalculated and transformed by wavelet transform

Although some researchers focused on sperm classifica-tion none has used our approach for sperm imaging Thispaper presents a system and a novel approach which uses aone-dimensional contour and gray level features to diagnosedifferent sperms according to their characteristics

3 Proposed Approaches

31 Overall Procedure In this section we present the algo-rithm of our work First of all we provide the flow ofour approach as Figure 3 After image segmentation thesperm head contour is extracted we transform the edge of

the sperm image into separated coordinate points and dividethe coordinate into two vectors

32 Proposed Algorithms For the classification we appliedtwo methods In the first the bilateral symmetrical functionof (119899) for the continuous situation is defined as follows

119878 (119899) = int1198621

1003817100381710038171003817100381711987511989911987510038171003817100381710038171003817

2

119889119904 minus int1198622

1003817100381710038171003817100381711987511989911987510038171003817100381710038171003817

2

119889119904 (1)

where1198621 denotes the arc on the left of the axis while1198622 is theright arc

The symmetry function is proposed as follows

Percentsym (119860119899 (119896)) = 1 minussum abs (119860 (119894) minus 119860 (2119896 minus 119894))

sum 2 lowast sqrt (119860 (119894) lowast 119860 (2119896 minus 119894))

(2)

119860119899(119896) denotes the corner point array while the 119896th point is

the smallest value of 119878(119899)In our secondmethod the geometric centre is calculated

We used the following equation to compute the horizontalcoordinate

119909mid =sum119899

119894=1119909119894

119899 (3)

where 119909mid represents the horizontal coordinate of the geo-metric centre 119909

119894represents the horizontal coordinate of each

point on the contour and 119899 represents the number of points

BioMed Research International 5

on the edgeThe vertical coordinate could be calculated in thesame way

To obtain the distance from edge to the centre we takeadvantage of the following format

119889 (119894) = radic(119909119894minus 119909mid)

2+ (119910119894minus 119910mid)

2 (4)

Considering the difficulty in choosing the starting pointwe had to avoid the problem Therefore we took rank algo-rithm [13] which ignores where to start into consideration inorder to classify the results The rank algorithm transformsa sequence of numbers 119909

1 1199092 1199093 119909

119899 into a sequence

composed of 1 and 0 The transform format is as follows

119868119899=

0 if 119909119899le 119909119899minus1

1 if 119909119899 gt 119909119899minus1

(5)

Therefore we achieved a binary system with 1 and 0We transformed 5 consecutive numbers in 119868

119899into decimal

numbers By computing the times each decimal numberappeared we ascertained the rank and the probability of theoriginal sequence Taking advantage of rank and probabilitywe calculated the similarity between two sequences Theformat is as follows

119863119898 (1198781 1198782) =sum2119898minus1

119896=0

10038161003816100381610038161198771 (119908119896) minus 1198772(119908119896)1003816100381610038161003816 1199011 (119908119896) 1199012 (119908119896)

(2119898 minus 1)sum2119898minus1

119896=01199011(119908119896) 1199012(119908119896)

(6)

By calculating the 119863119898 we distinguished the ten mostrepresentative sequences and took them as the criteria

Using the rank algorithm [13] we achieved anotherbinary system with 1 and 0 Then we transformed 8 con-secutive numbers in 119868119899 into decimal numbers and finallywe obtained the rank and the probability of the sequenceWith format (6) we obtained the similarity among thesequences Similar to the previous method we achievedanother criterion

To avoid the problem of a single criterion being toolopsided we took the gray level value of the sperm intoconsideration First we calculated the gray level value of allpoints within the sperm using the following format

119866 =

119899

sum1

119892119894 (7)

where 119866 represents the summation of the gray level value ofeach point119892119894 represents the gray level value of each point and119899 represents the total number of points within the sperm

Then we computed the gray percentage using the follow-ing format

119875 = 1 minus119866

119899 lowast 255 (8)

where 119875 represents the level of darkness

321 Joint Rank Difference and Gray Level Method By cal-culating the average rank difference of normal and abnormal

sperms we found the dividing line between them Thus therank difference itself can work as a judgment as to whetheror not the test sperm is normal We combined the two rankdifferences originating from the distance from centre to thecontour and the grey level value of the pixels in a sperm bycalculating the sum of the test spermrsquos distance average rankdifference and the average grey level value rank differencewhile each of them takes a certain weight The format is asfollows

119862 = 120572 lowast dARD + (1 minus 120572) lowast 255 lowast 119875 (9)

119862 represents the value that is the combination of the twostandards 120572 represents the weight of average rank differencedARD represents the average rank difference of distance fromthe centre to the contour

With the combination value we enhanced the originaljudgment by considering more elements and the importanceof each To find the dividing line we collected all thecombination values of the sperms and chose one of theiraverage values to provide the best accuracy for the dividingline

33 Fused SVM Method To achieve a better result we fusedthe SVM method as an advanced classifier SVM Methodsare supervised learning models with associated learningalgorithms that analyse data and recognize patterns and areused for classification and regression analysisThe basic SVMtakes a set of input sperm features and predicts for each giveninput the possible class form normal or abnormal making ita nonprobabilistic binary linear classifier

Given a training dataset (1199091 1199101) (1199092 1199102) (119909119899 119910119899)where 119909

119894isin R and 119910

119894is either 1 or minus1 indicating the class

to which the point 119909119894belongs let x = [119909

1 1199092 119909

119899]119879

The construction of the hyperplane for a linearly separableproblem is w119879x + 119887 = 0 where w is the normal vectorto the hyperplane and the parameter 119887w determines theoffset of the hyperplane from the origin along the normalvector w Thus the margin between the hyperplane and thenearest point is maximized and can be posed as the followingproblem

minw119887120585

1

2w119879w + 119862

119899

sum119894=1

120585119894

subject to 119910119894(119908119879119909119894+ 119887) ge 1 minus 120585

119894 119894 = 1 2 119899

120585119894ge 0

(10)

where 119862 is a user-defined constant as the penalty parameterof the error term

The SVM requires the optimal solution We use theLIBSVM[25] to solve this optimization problemwith the userguide given in [26]

As for the input of the SVM approach we fused the greylevel value and dARD as well First of all it extracts grey levelvalue of the pixels within the sperm head It starts from thecentre of the sperm and then it computes the following greylevel values in a square roundabout path The path map isshown in Table 1 as the sequence of numbers

6 BioMed Research International

Table 1 Grey level value extraction path

43 42 41 40 39 38 37 6444 21 20 19 18 17 36 6345 22 7 6 5 16 35 6246 23 8 1 4 15 34 6147 24 9 2 3 14 33 6048 25 10 11 12 13 32 5949 26 27 28 29 30 31 5850 51 52 53 54 55 56 57

Original sperm Sperm head

Figure 4 Extract sperm head from normal sperm

When the path collides with the contour the sequenceof grey level values takes advantage of the Rank algorithmin transforming 8 consecutive numbers 119868119899 into the decimalnumbers during the ends of extraction Then we achievedrank sequences and 119863119898 between every pair of images whichcan represent their similarities As a result we can tell whichimage represents the rest of the images the best by calculatingthe sum of the 119863

119898value from one image to all the other

images Thus we can put the rank sequences in order as wellWith the rank sequences we added the dARD sequences

at the end of corresponding rank sequence The newsequences consist of the training samples and test samplesThe ten training samples include five most representativesequences of normal sperms and five ones of abnormalsperms The rest sperms are the test samples

4 Fundamental Evaluation

In this fundamental evaluation we have undertaken variousexperiments on 80 normal and 80 abnormal sperms Theyare from a hospital the images are provided on httpscodegooglecompsupport-vector-machine-for-sperm-morphol-ogy-diagnosis as Supplementary material See Supplemen-tary Material available online at httpdxdoiorg1011552013687607 Before the classification of sperms thesegmentation is done by two steps (1) We pick the areawhere sperms do not overlay manually (2) We segment thesperms by reading the gray scale value of the picked areaWith knowing the location of the sperms we copy them fromthe area and build a new image to store each one of themWith a single sperm in an image it can cut off its tail easily

Figures 4 and 5 present an example of our segmentationresult

In this way we segmented all the 80 normal and 80abnormal sperm heads from 80 pieces of sperm images

41 Results of Our Waveform Extraction Methods In thissection we present part of the results of the steps in our

Original sperm Sperm head

Figure 5 Extract sperm head from abnormal sperm

experiments which explain how we made a choice andtheir performances In Table 2 the column Number lists thenumber of each imageThe columns Symmetry andMid showthe result of each transformation from image into waveform

Table 3 presents the waveform obtained by the firstmethod using four different starting points for the sameimage With the differences among four waves the problemarises from choosing one for the classification The resultof the second approach avoids the problem of picking thestarting point Thus we take the second result to classifysperms

42 Average Rank Difference In Figure 6 the horizontalaxis represents the serial number of each sperm image Thevertical axis represents the average difference of rank amongthe 80 normal or 80 abnormal sperms and the ten modelsperms As the result did not provide a clear dividing linebetween normal and abnormal sperms we adopted anothermethod to examine their differences We calculated thedifference in rank between the ten normal sperms and theother sperms Taking a look at Figure 6 if we take horizontalvalue 181 as the dividing line between normal and abnormalmeaning when the value of the average rank difference of asperm is below 181 we take the sperm as normal otherwisewe take it as abnormal In this case the accuracy could reach55

43 Grey Level Feature Then we focused on the influence ofthe grey level value Parts of the results are as shown in Figures7 and 8

In Figures 7 and 8 the horizontal axis represents the serialnumber of each sperm image In Figures 7(a) and 8(a) thevertical axis represents the average grey level value of thepoints of the sperms in the images In Figures 7(b) and 8(b)the vertical axis represents the grey level percentage of thesegmentation of the sperms

According to the percentage results we can ascertainthe percentage dividing line between normal and abnormalsperms to be 0268228785 With the division we can achieve725 accuracy

44 Joint Average Result We then tried to combine thetwo aspects as a judgment First of all we found the newparameter through format (7) By setting different 120572 wedetermined the different weight of each aspect The newparameter is as in Figure 9 It contains the new crite-ria produced with rank and gray level values of normaland abnormal sperms 120572 is the argument that represents

BioMed Research International 7

Table 2 Waveform extraction from both methods

Num Normal Symmetry Mid

1

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

25

20

15

10

5

0

2

0100200300400500600700800

1 9 17 25 33 41 49 57 65 73 81 89 970

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

3

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

4

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

5

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8502468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

6

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

7

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

8

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

9

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

8 BioMed Research International

Table 2 Continued

10

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

Num Abnormal Symmetry Mid

81

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

30

1 11 21 31 41 51 61 71 81 91 101

111

121

82

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

83

0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8505

101520253035

1 13 25 37 49 61 73 85 97 109

121

133

84

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 790

10203040506070

1 7 13 19 25 31 37 43 49 55 61 67 73 79

85

050

100150200250300350400

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

86

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

1012141618

1 7 13 19 25 31 37 43 49 55 61 67 73 79

87

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

88

0200400600800

10001200140016001800

1 7 13 19 25 31 37 43 49 55 61 67 73 7905

101520253035

1 7 13 19 25 31 37 43 49 55 61 67 73 79

BioMed Research International 9

Table 2 Continued

89

0200400600800

100012001400

1 7 13 19 25 31 37 43 49 55 61 67 73 790

5

10

15

20

25

30

1 7 13 19 25 31 37 43 49 55 61 67 73 79

90

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

Table 3 The waveform of different starting points from the first method

Startpoint Result

Start 1

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

Start 2

0

200

400

600

800

1000

1200

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800900

1000

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97

Table 4 Dividing line of C

120572 01 02 03 04 05 06 07 08 09DL 177 165 160 149 134 123 114 101 95

Table 5 Accuracy of different 120572

120572 01 02 03 04 05 06 07 08 09Acc 061 062 064 066 069 072 074 077 073

the weighted average rank difference taken in the criteriaWith the new parameter we can calculate its dividing line(DL) as a threshold

In Table 4 row DL represents the dividing line of each120572 value With the dividing line we can classify the spermsAssuming we have already known the exact classification ofeach sperm we can tell how many of them are correctlyclassified and thus we can come up with the accuracy Theaccuracy (Acc) of the different 120572 is as shown in Table 5

In Table 5 the row Acc represents the accuracy of each 120572value According to the result we chose the value 08 for 120572 Inthis case the accuracy can reach 76875

45 Fused SVM Result With all the data provided abovewe took advantage of the SVM method to undertake the

classification and used LIBSVM Software from httpwwwcsientuedutwsimcjlinlibsvm As for the parametersthe default values are used in the implementation and theSVM type is C-SVC the kernel function type is RBF function119890minus120574(119906minusV)2 = 1119896 cost = 1 cache size = 40 eps = 0001shrinking = 1 weight = 1

We input the training samples and test samples as men-tioned in Section 33 With ten training samples and testsamples of 160 sperms which consist of 80 normal and 80abnormal ones we obtain accuracy with 889

5 Comparison Evaluation

In this section we introduce the three methods we usedas comparisons to our method K-nearest neighbor methodworks quite effective as well the other two fail to satisfy us

51 Compared K-Nearest Neighbour Approach TheK-nearestneighbour method showed the following differences

We took the 160 images of sperms as the original clas-sifications and then we input the test image transformedall the images into gray and then calculated the differencebetween the test image and the original images Then wepicked the ten images with the smaller differences Aftercounting the classifications of the ten images we specified theone occurring most often to be the classification of the test

10 BioMed Research International

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a) Normal sperms

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(b) Abnormal sperms

Figure 6 Normal and abnormal sperms and their corresponding data

1 7 13 19 25 31 37 43 49 55 61 67 73 790

102030405060708090

100

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

(b)

Figure 7 Grey level value and percentage of normal sperms

0

20

40

60

80

100

120

140

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

07

(b)

Figure 8 Grey level value and percentage of abnormal sperms

image The classification of the top ten nearest neighbours isas in Figure 10

Taking the sperms whose sum value is larger than 3 asnormal and the sperms whose sum value is smaller than 4as abnormal we can tell that only 39 sperms are mistakenlyclassified providing accuracy of 75625

52 Compared SIFTApproach Lowe summed up the existingfeature detection method based on invariants technologyin 2004 and formally proposed an image scaling rotationand even affine transformation for invariant image with localfeature description operator based on scale space SIFT [22]

The SIFT algorithm first undertakes feature detection in scalespace and defines the key pointsrsquo positions and the scale ofthe key points and then it uses the main direction of theneighbourhood gradient of the key points as the directionfeatures of the points in order to achieve the operatorindependence of scale and the direction The format whichproduces the scale space is as follows

119871 (119909 119910 119905) = intinfin

120585=minusinfin

intinfin

120578

1

2120587119905119890minus(1205852+1205782)2119905

119891 (119909 minus 120585 119910 minus 120578) 119889120585 119889120578

(11)

BioMed Research International 11

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(a) 120572 = 01

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(b) 120572 = 02

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(c) 120572 = 03

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(d) 120572 = 04

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(e) 120572 = 05

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(f) 120572 = 06

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(g) 120572 = 07

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160

(h) 120572 = 08

020406080

100120140

1 13 25 37 49 61 73 85 97 109 121 133 145157

(i) 120572 = 09

Figure 9 The result of joint rank and gray level features of normal and abnormal sperms

12 BioMed Research International

1 7 13 19 25 31 37 43 49 55 61 67 73 790

2

4

6

8

10

12

(a) Normal sperms

1 7 13 19 25 31 37 43 49 55 61 67 73 790

051

152

253

354

45

(b) Abnormal sperms

Figure 10 The classification of the top ten neighbours of (a) normal and (b) abnormal sperms

Rectangularsegment

Segmentation afterrotation

Ellipse

Figure 11 The ellipse model to estimate the contour of the sperm

In format (11) t represents the scaling parameter Byundertaking convolution in the whole domain with a two-dimensionalGaussian kernel and input image we can achievescaling correspondent to t

The SIFT feature vector has the following features (a) it isthe local feature of an image which maintains invariance notonly on rotation scale and brightness variation but also onthe viewing angle the affine transformation and the noise(b) it is distinctive and informative and suitable for fastaccurate matching in a mass signature database (c) it canproduce a large number of SIFT feature vectors with fewobjects (d) it is of high speed

Matlab source code of SIFT is fromhttpwwwvlfeatorgindexhtml [27] We used the SIFT method to extract featurepoints of all the sperm images and achieved the ten normaland ten abnormal sperm images as in the training modelThen we calculated the difference between the trainingmodels and the other 70 normal and 70 abnormal spermimages With the dividing line we came up with accuracyof 50 The result turned out to be bad because the inputdata are simple sperm images and the SIFT algorithm cannotextract enough feature descriptors for classification

53 Compared Ellipse Model Approach With the image ofa single sperm we can use the ellipse model to estimatethe contour of the sperm First of all we used a rectangularsegment to cut out the sperm In order to make the rectangleclose to the sperm edge we needed to choose the rectanglewith the smallest area we rotated the sperm so that thesegmentation would be easier We calculated the left-mostupper-most downward-most and right-most points on theedges and built the rectangular segment based on them asdepicted in Figure 11

76888890

7563

5000

6625

000100020003000400050006000700080009000

10000

Jointaverage

K-neighbor SIFT Ellipticmodel

FusedSVM

()

Figure 12 Accuracy comparison of all methods

With the rectangle we achieved an ellipse to estimatethe sperm whose long axis radius is half the length of therectanglersquos long side and whose short axis radius is half thelength of the rectanglersquos short side As a result we ascertainedthe ovality which is the ratio of the length of the long axisto the short axis Through our experiment on 80 sperms wecould tell whether the sperm was too close to a circle whoseratio is close to 1 and whether the sperm was too slim whoseratio is close to 2 or even larger than 2 We took those whoseratio was larger than 12 and smaller than 18 as normal andthus we achieved accuracy of 6625

6 Conclusion

As shown in Figure 12 we have presented an approachfor the classification of sperm images with one-dimensionalwaveform and gray level features In the comprehensiveevaluation the joint average approach method was applied

BioMed Research International 13

to 160 sperm image samples and provided 76875 accuracyof judgment We then applied the fused SVM method tothe images and reached accuracy of 875 The result provesthat the proposed approach is superior to the previousapproaches

References

[1] N R Pal and S K Pal ldquoA review on image segmentationtechniquesrdquo Pattern Recognition vol 26 no 9 pp 1277ndash12941993

[2] A Yezzi L Zollei and T Kapur ldquoA variational frameworkfor integrating segmentation and registration through activecontoursrdquoMedical ImageAnalysis vol 7 no 2 pp 171ndash185 2003

[3] R Chrastek M Wolf K Donath et al ldquoAutomated segmenta-tion of the optic nerve head for diagnosis of glaucomardquoMedicalImage Analysis vol 9 no 4 pp 297ndash314 2005

[4] H W Kang and S Y Shin ldquoEnhanced lane interactive imagesegmentation by incremental path map constructionrdquo Graphi-cal Models vol 64 no 5 pp 282ndash303 2002

[5] J Peters O Ecabert C Meyer R Kneser and J Weese ldquoOpti-mizing boundary detection via Simulated Search with appli-cations to multi-modal heart segmentationrdquo Medical ImageAnalysis vol 14 no 1 pp 70ndash84 2010

[6] G Behiels F Maes D Vandermeulen and P Suetens ldquoEvalu-ation of image features and search strategies for segmentationof bone structures in radiographs using Active Shape ModelsrdquoMedical Image Analysis vol 6 no 1 pp 47ndash62 2002

[7] M Blanchard K Haguenoer A Apert et al ldquoSperm morphol-ogy assessment using Davidrsquos classiffication time to switch tostrict criteria Prospective comparative analysis in a selectedIVF populationrdquo International Journal of Andrology vol 34 no2 pp 145ndash152 2010

[8] MKass AWitkin andD Terzopoulos ldquoSnakes active contourmodelsrdquo International Journal of Computer Vision vol 1 no 4pp 321ndash331 1988

[9] TMcInerney andD Terzopoulos ldquoT-snakes topology adaptivesnakesrdquoMedical Image Analysis vol 4 no 2 pp 73ndash91 2000

[10] J Liang T McInerney and D Terzopoulos ldquoUnited SnakesrdquoMedical Image Analysis vol 10 no 2 pp 215ndash233 2006

[11] M Zhao Z Meng K K Tseng J S Pan and C Y Hsu ldquoSym-metry auto-detection based on contour and corner modelsrdquo inProceedings of the 5th International Conference on Genetic andEvolutionary Computing (ICGEC rsquo11) pp 345ndash349 September2011

[12] Y Mingqiang K Kidiyo and R Joseph ldquoA survey of shapefeature extraction techniquesrdquo Pattern Recognition vol 15 no7 pp 43ndash90 2008

[13] Y Wang and H Huang ldquoAnalysis of Human Heartbeat withEKG Signalsrdquo 2010

[14] Q Hua A Ji and Q He ldquoMultiple real-valued K nearestneighbor classifiers system by feature groupingrdquo in Proceedingsof the IEEE International Conference on Systems Man andCybernetics (SMC rsquo10) pp 3922ndash3925 October 2010

[15] P Viswanath and T Hitendra Sarma ldquoAn improvement to k-nearest neighbor classifierrdquo in Proceedings of the IEEE RecentAdvances in Intelligent Computational Systems (RAICS rsquo11) pp227ndash231 September 2011

[16] P Wu and Q Chen ldquoA novel SVM-based edge detectionmethodrdquo Physics Procedia vol 24 pp 2075ndash2082 2012

[17] D N Sotiropoulos and G A Tsihrintzis ldquoArtificial immunesystem-based classification in class-imbalanced image prob-lemsrdquo in Proceedings of the 8th International Conference onIntelligent InformationHiding andMultimedia Signal Processingpp 138ndash141 July 2012

[18] M Ramon FMartınez-Pastor O Garcıa-Alvarez et al ldquoTakingadvantage of the use of supervised learning methods forcharacterization of sperm population structure related withfreezability in the Iberian red deerrdquoTheriogenology vol 77 no8 pp 1661ndash1672 2012

[19] S G Goodson Z Zhang J K Tsuruta W Wang and DA OrsquoBrien ldquoClassification of mouse sperm motility patternsusing an automatedmulticlass support vectormachinesmodelrdquoBiology of Reproduction vol 84 no 6 pp 1207ndash1215 2011

[20] D T Lin and D C Pan ldquoIntegrating a mixed-feature modeland multiclass support vector machine for facial expressionrecognitionrdquo Integrated Computer-Aided Engineering vol 16no 1 pp 61ndash74 2009

[21] C G Cheng Y M Tian and W Y Jin ldquoA study on the earlydetection of colon cancer using the methods of wavelet featureextraction and SVM classifications of FTIRrdquo Spectroscopy vol22 no 5 pp 397ndash404 2008

[22] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004

[23] V R Nafisi M H Moradi and M H Nasr-Esfahani ldquoSpermidentification using elliptic model and tail detectionrdquo WorldAcademy of Science Engineering and Technology vol 6 pp 205ndash208 2005

[24] W J Yi K S Park and J S Paick ldquoParameterized charac-terization of elliptic sperm heads using Fourier representationand wavelet transformrdquo in Proceedings of the 20th AnnualInternational Conference of the IEEE Engineering in Medicineand Biology Society vol 20 pp 974ndash977 1998

[25] C C Chang andC J Lin LIBSVMmdasha library for support vectormachines Version 312 httpwwwcsientuedutwsimcjlinlibsvm

[26] CW Hsu C C Chang and C J Lin ldquoA practical guide to sup-port vector classificationrdquo 2010 httpwwwcsientuedutwsimcjlin

[27] A Vedaldi and B Fulkerson ldquoVLFeat an open and portablelibrary of computer vision algorithmsrdquo 2008 httpwwwvlfeatorg

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

BioMed Research International 5

on the edgeThe vertical coordinate could be calculated in thesame way

To obtain the distance from edge to the centre we takeadvantage of the following format

119889 (119894) = radic(119909119894minus 119909mid)

2+ (119910119894minus 119910mid)

2 (4)

Considering the difficulty in choosing the starting pointwe had to avoid the problem Therefore we took rank algo-rithm [13] which ignores where to start into consideration inorder to classify the results The rank algorithm transformsa sequence of numbers 119909

1 1199092 1199093 119909

119899 into a sequence

composed of 1 and 0 The transform format is as follows

119868119899=

0 if 119909119899le 119909119899minus1

1 if 119909119899 gt 119909119899minus1

(5)

Therefore we achieved a binary system with 1 and 0We transformed 5 consecutive numbers in 119868

119899into decimal

numbers By computing the times each decimal numberappeared we ascertained the rank and the probability of theoriginal sequence Taking advantage of rank and probabilitywe calculated the similarity between two sequences Theformat is as follows

119863119898 (1198781 1198782) =sum2119898minus1

119896=0

10038161003816100381610038161198771 (119908119896) minus 1198772(119908119896)1003816100381610038161003816 1199011 (119908119896) 1199012 (119908119896)

(2119898 minus 1)sum2119898minus1

119896=01199011(119908119896) 1199012(119908119896)

(6)

By calculating the 119863119898 we distinguished the ten mostrepresentative sequences and took them as the criteria

Using the rank algorithm [13] we achieved anotherbinary system with 1 and 0 Then we transformed 8 con-secutive numbers in 119868119899 into decimal numbers and finallywe obtained the rank and the probability of the sequenceWith format (6) we obtained the similarity among thesequences Similar to the previous method we achievedanother criterion

To avoid the problem of a single criterion being toolopsided we took the gray level value of the sperm intoconsideration First we calculated the gray level value of allpoints within the sperm using the following format

119866 =

119899

sum1

119892119894 (7)

where 119866 represents the summation of the gray level value ofeach point119892119894 represents the gray level value of each point and119899 represents the total number of points within the sperm

Then we computed the gray percentage using the follow-ing format

119875 = 1 minus119866

119899 lowast 255 (8)

where 119875 represents the level of darkness

321 Joint Rank Difference and Gray Level Method By cal-culating the average rank difference of normal and abnormal

sperms we found the dividing line between them Thus therank difference itself can work as a judgment as to whetheror not the test sperm is normal We combined the two rankdifferences originating from the distance from centre to thecontour and the grey level value of the pixels in a sperm bycalculating the sum of the test spermrsquos distance average rankdifference and the average grey level value rank differencewhile each of them takes a certain weight The format is asfollows

119862 = 120572 lowast dARD + (1 minus 120572) lowast 255 lowast 119875 (9)

119862 represents the value that is the combination of the twostandards 120572 represents the weight of average rank differencedARD represents the average rank difference of distance fromthe centre to the contour

With the combination value we enhanced the originaljudgment by considering more elements and the importanceof each To find the dividing line we collected all thecombination values of the sperms and chose one of theiraverage values to provide the best accuracy for the dividingline

33 Fused SVM Method To achieve a better result we fusedthe SVM method as an advanced classifier SVM Methodsare supervised learning models with associated learningalgorithms that analyse data and recognize patterns and areused for classification and regression analysisThe basic SVMtakes a set of input sperm features and predicts for each giveninput the possible class form normal or abnormal making ita nonprobabilistic binary linear classifier

Given a training dataset (1199091 1199101) (1199092 1199102) (119909119899 119910119899)where 119909

119894isin R and 119910

119894is either 1 or minus1 indicating the class

to which the point 119909119894belongs let x = [119909

1 1199092 119909

119899]119879

The construction of the hyperplane for a linearly separableproblem is w119879x + 119887 = 0 where w is the normal vectorto the hyperplane and the parameter 119887w determines theoffset of the hyperplane from the origin along the normalvector w Thus the margin between the hyperplane and thenearest point is maximized and can be posed as the followingproblem

minw119887120585

1

2w119879w + 119862

119899

sum119894=1

120585119894

subject to 119910119894(119908119879119909119894+ 119887) ge 1 minus 120585

119894 119894 = 1 2 119899

120585119894ge 0

(10)

where 119862 is a user-defined constant as the penalty parameterof the error term

The SVM requires the optimal solution We use theLIBSVM[25] to solve this optimization problemwith the userguide given in [26]

As for the input of the SVM approach we fused the greylevel value and dARD as well First of all it extracts grey levelvalue of the pixels within the sperm head It starts from thecentre of the sperm and then it computes the following greylevel values in a square roundabout path The path map isshown in Table 1 as the sequence of numbers

6 BioMed Research International

Table 1 Grey level value extraction path

43 42 41 40 39 38 37 6444 21 20 19 18 17 36 6345 22 7 6 5 16 35 6246 23 8 1 4 15 34 6147 24 9 2 3 14 33 6048 25 10 11 12 13 32 5949 26 27 28 29 30 31 5850 51 52 53 54 55 56 57

Original sperm Sperm head

Figure 4 Extract sperm head from normal sperm

When the path collides with the contour the sequenceof grey level values takes advantage of the Rank algorithmin transforming 8 consecutive numbers 119868119899 into the decimalnumbers during the ends of extraction Then we achievedrank sequences and 119863119898 between every pair of images whichcan represent their similarities As a result we can tell whichimage represents the rest of the images the best by calculatingthe sum of the 119863

119898value from one image to all the other

images Thus we can put the rank sequences in order as wellWith the rank sequences we added the dARD sequences

at the end of corresponding rank sequence The newsequences consist of the training samples and test samplesThe ten training samples include five most representativesequences of normal sperms and five ones of abnormalsperms The rest sperms are the test samples

4 Fundamental Evaluation

In this fundamental evaluation we have undertaken variousexperiments on 80 normal and 80 abnormal sperms Theyare from a hospital the images are provided on httpscodegooglecompsupport-vector-machine-for-sperm-morphol-ogy-diagnosis as Supplementary material See Supplemen-tary Material available online at httpdxdoiorg1011552013687607 Before the classification of sperms thesegmentation is done by two steps (1) We pick the areawhere sperms do not overlay manually (2) We segment thesperms by reading the gray scale value of the picked areaWith knowing the location of the sperms we copy them fromthe area and build a new image to store each one of themWith a single sperm in an image it can cut off its tail easily

Figures 4 and 5 present an example of our segmentationresult

In this way we segmented all the 80 normal and 80abnormal sperm heads from 80 pieces of sperm images

41 Results of Our Waveform Extraction Methods In thissection we present part of the results of the steps in our

Original sperm Sperm head

Figure 5 Extract sperm head from abnormal sperm

experiments which explain how we made a choice andtheir performances In Table 2 the column Number lists thenumber of each imageThe columns Symmetry andMid showthe result of each transformation from image into waveform

Table 3 presents the waveform obtained by the firstmethod using four different starting points for the sameimage With the differences among four waves the problemarises from choosing one for the classification The resultof the second approach avoids the problem of picking thestarting point Thus we take the second result to classifysperms

42 Average Rank Difference In Figure 6 the horizontalaxis represents the serial number of each sperm image Thevertical axis represents the average difference of rank amongthe 80 normal or 80 abnormal sperms and the ten modelsperms As the result did not provide a clear dividing linebetween normal and abnormal sperms we adopted anothermethod to examine their differences We calculated thedifference in rank between the ten normal sperms and theother sperms Taking a look at Figure 6 if we take horizontalvalue 181 as the dividing line between normal and abnormalmeaning when the value of the average rank difference of asperm is below 181 we take the sperm as normal otherwisewe take it as abnormal In this case the accuracy could reach55

43 Grey Level Feature Then we focused on the influence ofthe grey level value Parts of the results are as shown in Figures7 and 8

In Figures 7 and 8 the horizontal axis represents the serialnumber of each sperm image In Figures 7(a) and 8(a) thevertical axis represents the average grey level value of thepoints of the sperms in the images In Figures 7(b) and 8(b)the vertical axis represents the grey level percentage of thesegmentation of the sperms

According to the percentage results we can ascertainthe percentage dividing line between normal and abnormalsperms to be 0268228785 With the division we can achieve725 accuracy

44 Joint Average Result We then tried to combine thetwo aspects as a judgment First of all we found the newparameter through format (7) By setting different 120572 wedetermined the different weight of each aspect The newparameter is as in Figure 9 It contains the new crite-ria produced with rank and gray level values of normaland abnormal sperms 120572 is the argument that represents

BioMed Research International 7

Table 2 Waveform extraction from both methods

Num Normal Symmetry Mid

1

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

25

20

15

10

5

0

2

0100200300400500600700800

1 9 17 25 33 41 49 57 65 73 81 89 970

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

3

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

4

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

5

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8502468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

6

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

7

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

8

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

9

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

8 BioMed Research International

Table 2 Continued

10

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

Num Abnormal Symmetry Mid

81

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

30

1 11 21 31 41 51 61 71 81 91 101

111

121

82

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

83

0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8505

101520253035

1 13 25 37 49 61 73 85 97 109

121

133

84

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 790

10203040506070

1 7 13 19 25 31 37 43 49 55 61 67 73 79

85

050

100150200250300350400

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

86

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

1012141618

1 7 13 19 25 31 37 43 49 55 61 67 73 79

87

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

88

0200400600800

10001200140016001800

1 7 13 19 25 31 37 43 49 55 61 67 73 7905

101520253035

1 7 13 19 25 31 37 43 49 55 61 67 73 79

BioMed Research International 9

Table 2 Continued

89

0200400600800

100012001400

1 7 13 19 25 31 37 43 49 55 61 67 73 790

5

10

15

20

25

30

1 7 13 19 25 31 37 43 49 55 61 67 73 79

90

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

Table 3 The waveform of different starting points from the first method

Startpoint Result

Start 1

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

Start 2

0

200

400

600

800

1000

1200

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800900

1000

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97

Table 4 Dividing line of C

120572 01 02 03 04 05 06 07 08 09DL 177 165 160 149 134 123 114 101 95

Table 5 Accuracy of different 120572

120572 01 02 03 04 05 06 07 08 09Acc 061 062 064 066 069 072 074 077 073

the weighted average rank difference taken in the criteriaWith the new parameter we can calculate its dividing line(DL) as a threshold

In Table 4 row DL represents the dividing line of each120572 value With the dividing line we can classify the spermsAssuming we have already known the exact classification ofeach sperm we can tell how many of them are correctlyclassified and thus we can come up with the accuracy Theaccuracy (Acc) of the different 120572 is as shown in Table 5

In Table 5 the row Acc represents the accuracy of each 120572value According to the result we chose the value 08 for 120572 Inthis case the accuracy can reach 76875

45 Fused SVM Result With all the data provided abovewe took advantage of the SVM method to undertake the

classification and used LIBSVM Software from httpwwwcsientuedutwsimcjlinlibsvm As for the parametersthe default values are used in the implementation and theSVM type is C-SVC the kernel function type is RBF function119890minus120574(119906minusV)2 = 1119896 cost = 1 cache size = 40 eps = 0001shrinking = 1 weight = 1

We input the training samples and test samples as men-tioned in Section 33 With ten training samples and testsamples of 160 sperms which consist of 80 normal and 80abnormal ones we obtain accuracy with 889

5 Comparison Evaluation

In this section we introduce the three methods we usedas comparisons to our method K-nearest neighbor methodworks quite effective as well the other two fail to satisfy us

51 Compared K-Nearest Neighbour Approach TheK-nearestneighbour method showed the following differences

We took the 160 images of sperms as the original clas-sifications and then we input the test image transformedall the images into gray and then calculated the differencebetween the test image and the original images Then wepicked the ten images with the smaller differences Aftercounting the classifications of the ten images we specified theone occurring most often to be the classification of the test

10 BioMed Research International

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a) Normal sperms

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(b) Abnormal sperms

Figure 6 Normal and abnormal sperms and their corresponding data

1 7 13 19 25 31 37 43 49 55 61 67 73 790

102030405060708090

100

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

(b)

Figure 7 Grey level value and percentage of normal sperms

0

20

40

60

80

100

120

140

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

07

(b)

Figure 8 Grey level value and percentage of abnormal sperms

image The classification of the top ten nearest neighbours isas in Figure 10

Taking the sperms whose sum value is larger than 3 asnormal and the sperms whose sum value is smaller than 4as abnormal we can tell that only 39 sperms are mistakenlyclassified providing accuracy of 75625

52 Compared SIFTApproach Lowe summed up the existingfeature detection method based on invariants technologyin 2004 and formally proposed an image scaling rotationand even affine transformation for invariant image with localfeature description operator based on scale space SIFT [22]

The SIFT algorithm first undertakes feature detection in scalespace and defines the key pointsrsquo positions and the scale ofthe key points and then it uses the main direction of theneighbourhood gradient of the key points as the directionfeatures of the points in order to achieve the operatorindependence of scale and the direction The format whichproduces the scale space is as follows

119871 (119909 119910 119905) = intinfin

120585=minusinfin

intinfin

120578

1

2120587119905119890minus(1205852+1205782)2119905

119891 (119909 minus 120585 119910 minus 120578) 119889120585 119889120578

(11)

BioMed Research International 11

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(a) 120572 = 01

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(b) 120572 = 02

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(c) 120572 = 03

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(d) 120572 = 04

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(e) 120572 = 05

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(f) 120572 = 06

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(g) 120572 = 07

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160

(h) 120572 = 08

020406080

100120140

1 13 25 37 49 61 73 85 97 109 121 133 145157

(i) 120572 = 09

Figure 9 The result of joint rank and gray level features of normal and abnormal sperms

12 BioMed Research International

1 7 13 19 25 31 37 43 49 55 61 67 73 790

2

4

6

8

10

12

(a) Normal sperms

1 7 13 19 25 31 37 43 49 55 61 67 73 790

051

152

253

354

45

(b) Abnormal sperms

Figure 10 The classification of the top ten neighbours of (a) normal and (b) abnormal sperms

Rectangularsegment

Segmentation afterrotation

Ellipse

Figure 11 The ellipse model to estimate the contour of the sperm

In format (11) t represents the scaling parameter Byundertaking convolution in the whole domain with a two-dimensionalGaussian kernel and input image we can achievescaling correspondent to t

The SIFT feature vector has the following features (a) it isthe local feature of an image which maintains invariance notonly on rotation scale and brightness variation but also onthe viewing angle the affine transformation and the noise(b) it is distinctive and informative and suitable for fastaccurate matching in a mass signature database (c) it canproduce a large number of SIFT feature vectors with fewobjects (d) it is of high speed

Matlab source code of SIFT is fromhttpwwwvlfeatorgindexhtml [27] We used the SIFT method to extract featurepoints of all the sperm images and achieved the ten normaland ten abnormal sperm images as in the training modelThen we calculated the difference between the trainingmodels and the other 70 normal and 70 abnormal spermimages With the dividing line we came up with accuracyof 50 The result turned out to be bad because the inputdata are simple sperm images and the SIFT algorithm cannotextract enough feature descriptors for classification

53 Compared Ellipse Model Approach With the image ofa single sperm we can use the ellipse model to estimatethe contour of the sperm First of all we used a rectangularsegment to cut out the sperm In order to make the rectangleclose to the sperm edge we needed to choose the rectanglewith the smallest area we rotated the sperm so that thesegmentation would be easier We calculated the left-mostupper-most downward-most and right-most points on theedges and built the rectangular segment based on them asdepicted in Figure 11

76888890

7563

5000

6625

000100020003000400050006000700080009000

10000

Jointaverage

K-neighbor SIFT Ellipticmodel

FusedSVM

()

Figure 12 Accuracy comparison of all methods

With the rectangle we achieved an ellipse to estimatethe sperm whose long axis radius is half the length of therectanglersquos long side and whose short axis radius is half thelength of the rectanglersquos short side As a result we ascertainedthe ovality which is the ratio of the length of the long axisto the short axis Through our experiment on 80 sperms wecould tell whether the sperm was too close to a circle whoseratio is close to 1 and whether the sperm was too slim whoseratio is close to 2 or even larger than 2 We took those whoseratio was larger than 12 and smaller than 18 as normal andthus we achieved accuracy of 6625

6 Conclusion

As shown in Figure 12 we have presented an approachfor the classification of sperm images with one-dimensionalwaveform and gray level features In the comprehensiveevaluation the joint average approach method was applied

BioMed Research International 13

to 160 sperm image samples and provided 76875 accuracyof judgment We then applied the fused SVM method tothe images and reached accuracy of 875 The result provesthat the proposed approach is superior to the previousapproaches

References

[1] N R Pal and S K Pal ldquoA review on image segmentationtechniquesrdquo Pattern Recognition vol 26 no 9 pp 1277ndash12941993

[2] A Yezzi L Zollei and T Kapur ldquoA variational frameworkfor integrating segmentation and registration through activecontoursrdquoMedical ImageAnalysis vol 7 no 2 pp 171ndash185 2003

[3] R Chrastek M Wolf K Donath et al ldquoAutomated segmenta-tion of the optic nerve head for diagnosis of glaucomardquoMedicalImage Analysis vol 9 no 4 pp 297ndash314 2005

[4] H W Kang and S Y Shin ldquoEnhanced lane interactive imagesegmentation by incremental path map constructionrdquo Graphi-cal Models vol 64 no 5 pp 282ndash303 2002

[5] J Peters O Ecabert C Meyer R Kneser and J Weese ldquoOpti-mizing boundary detection via Simulated Search with appli-cations to multi-modal heart segmentationrdquo Medical ImageAnalysis vol 14 no 1 pp 70ndash84 2010

[6] G Behiels F Maes D Vandermeulen and P Suetens ldquoEvalu-ation of image features and search strategies for segmentationof bone structures in radiographs using Active Shape ModelsrdquoMedical Image Analysis vol 6 no 1 pp 47ndash62 2002

[7] M Blanchard K Haguenoer A Apert et al ldquoSperm morphol-ogy assessment using Davidrsquos classiffication time to switch tostrict criteria Prospective comparative analysis in a selectedIVF populationrdquo International Journal of Andrology vol 34 no2 pp 145ndash152 2010

[8] MKass AWitkin andD Terzopoulos ldquoSnakes active contourmodelsrdquo International Journal of Computer Vision vol 1 no 4pp 321ndash331 1988

[9] TMcInerney andD Terzopoulos ldquoT-snakes topology adaptivesnakesrdquoMedical Image Analysis vol 4 no 2 pp 73ndash91 2000

[10] J Liang T McInerney and D Terzopoulos ldquoUnited SnakesrdquoMedical Image Analysis vol 10 no 2 pp 215ndash233 2006

[11] M Zhao Z Meng K K Tseng J S Pan and C Y Hsu ldquoSym-metry auto-detection based on contour and corner modelsrdquo inProceedings of the 5th International Conference on Genetic andEvolutionary Computing (ICGEC rsquo11) pp 345ndash349 September2011

[12] Y Mingqiang K Kidiyo and R Joseph ldquoA survey of shapefeature extraction techniquesrdquo Pattern Recognition vol 15 no7 pp 43ndash90 2008

[13] Y Wang and H Huang ldquoAnalysis of Human Heartbeat withEKG Signalsrdquo 2010

[14] Q Hua A Ji and Q He ldquoMultiple real-valued K nearestneighbor classifiers system by feature groupingrdquo in Proceedingsof the IEEE International Conference on Systems Man andCybernetics (SMC rsquo10) pp 3922ndash3925 October 2010

[15] P Viswanath and T Hitendra Sarma ldquoAn improvement to k-nearest neighbor classifierrdquo in Proceedings of the IEEE RecentAdvances in Intelligent Computational Systems (RAICS rsquo11) pp227ndash231 September 2011

[16] P Wu and Q Chen ldquoA novel SVM-based edge detectionmethodrdquo Physics Procedia vol 24 pp 2075ndash2082 2012

[17] D N Sotiropoulos and G A Tsihrintzis ldquoArtificial immunesystem-based classification in class-imbalanced image prob-lemsrdquo in Proceedings of the 8th International Conference onIntelligent InformationHiding andMultimedia Signal Processingpp 138ndash141 July 2012

[18] M Ramon FMartınez-Pastor O Garcıa-Alvarez et al ldquoTakingadvantage of the use of supervised learning methods forcharacterization of sperm population structure related withfreezability in the Iberian red deerrdquoTheriogenology vol 77 no8 pp 1661ndash1672 2012

[19] S G Goodson Z Zhang J K Tsuruta W Wang and DA OrsquoBrien ldquoClassification of mouse sperm motility patternsusing an automatedmulticlass support vectormachinesmodelrdquoBiology of Reproduction vol 84 no 6 pp 1207ndash1215 2011

[20] D T Lin and D C Pan ldquoIntegrating a mixed-feature modeland multiclass support vector machine for facial expressionrecognitionrdquo Integrated Computer-Aided Engineering vol 16no 1 pp 61ndash74 2009

[21] C G Cheng Y M Tian and W Y Jin ldquoA study on the earlydetection of colon cancer using the methods of wavelet featureextraction and SVM classifications of FTIRrdquo Spectroscopy vol22 no 5 pp 397ndash404 2008

[22] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004

[23] V R Nafisi M H Moradi and M H Nasr-Esfahani ldquoSpermidentification using elliptic model and tail detectionrdquo WorldAcademy of Science Engineering and Technology vol 6 pp 205ndash208 2005

[24] W J Yi K S Park and J S Paick ldquoParameterized charac-terization of elliptic sperm heads using Fourier representationand wavelet transformrdquo in Proceedings of the 20th AnnualInternational Conference of the IEEE Engineering in Medicineand Biology Society vol 20 pp 974ndash977 1998

[25] C C Chang andC J Lin LIBSVMmdasha library for support vectormachines Version 312 httpwwwcsientuedutwsimcjlinlibsvm

[26] CW Hsu C C Chang and C J Lin ldquoA practical guide to sup-port vector classificationrdquo 2010 httpwwwcsientuedutwsimcjlin

[27] A Vedaldi and B Fulkerson ldquoVLFeat an open and portablelibrary of computer vision algorithmsrdquo 2008 httpwwwvlfeatorg

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

6 BioMed Research International

Table 1 Grey level value extraction path

43 42 41 40 39 38 37 6444 21 20 19 18 17 36 6345 22 7 6 5 16 35 6246 23 8 1 4 15 34 6147 24 9 2 3 14 33 6048 25 10 11 12 13 32 5949 26 27 28 29 30 31 5850 51 52 53 54 55 56 57

Original sperm Sperm head

Figure 4 Extract sperm head from normal sperm

When the path collides with the contour the sequenceof grey level values takes advantage of the Rank algorithmin transforming 8 consecutive numbers 119868119899 into the decimalnumbers during the ends of extraction Then we achievedrank sequences and 119863119898 between every pair of images whichcan represent their similarities As a result we can tell whichimage represents the rest of the images the best by calculatingthe sum of the 119863

119898value from one image to all the other

images Thus we can put the rank sequences in order as wellWith the rank sequences we added the dARD sequences

at the end of corresponding rank sequence The newsequences consist of the training samples and test samplesThe ten training samples include five most representativesequences of normal sperms and five ones of abnormalsperms The rest sperms are the test samples

4 Fundamental Evaluation

In this fundamental evaluation we have undertaken variousexperiments on 80 normal and 80 abnormal sperms Theyare from a hospital the images are provided on httpscodegooglecompsupport-vector-machine-for-sperm-morphol-ogy-diagnosis as Supplementary material See Supplemen-tary Material available online at httpdxdoiorg1011552013687607 Before the classification of sperms thesegmentation is done by two steps (1) We pick the areawhere sperms do not overlay manually (2) We segment thesperms by reading the gray scale value of the picked areaWith knowing the location of the sperms we copy them fromthe area and build a new image to store each one of themWith a single sperm in an image it can cut off its tail easily

Figures 4 and 5 present an example of our segmentationresult

In this way we segmented all the 80 normal and 80abnormal sperm heads from 80 pieces of sperm images

41 Results of Our Waveform Extraction Methods In thissection we present part of the results of the steps in our

Original sperm Sperm head

Figure 5 Extract sperm head from abnormal sperm

experiments which explain how we made a choice andtheir performances In Table 2 the column Number lists thenumber of each imageThe columns Symmetry andMid showthe result of each transformation from image into waveform

Table 3 presents the waveform obtained by the firstmethod using four different starting points for the sameimage With the differences among four waves the problemarises from choosing one for the classification The resultof the second approach avoids the problem of picking thestarting point Thus we take the second result to classifysperms

42 Average Rank Difference In Figure 6 the horizontalaxis represents the serial number of each sperm image Thevertical axis represents the average difference of rank amongthe 80 normal or 80 abnormal sperms and the ten modelsperms As the result did not provide a clear dividing linebetween normal and abnormal sperms we adopted anothermethod to examine their differences We calculated thedifference in rank between the ten normal sperms and theother sperms Taking a look at Figure 6 if we take horizontalvalue 181 as the dividing line between normal and abnormalmeaning when the value of the average rank difference of asperm is below 181 we take the sperm as normal otherwisewe take it as abnormal In this case the accuracy could reach55

43 Grey Level Feature Then we focused on the influence ofthe grey level value Parts of the results are as shown in Figures7 and 8

In Figures 7 and 8 the horizontal axis represents the serialnumber of each sperm image In Figures 7(a) and 8(a) thevertical axis represents the average grey level value of thepoints of the sperms in the images In Figures 7(b) and 8(b)the vertical axis represents the grey level percentage of thesegmentation of the sperms

According to the percentage results we can ascertainthe percentage dividing line between normal and abnormalsperms to be 0268228785 With the division we can achieve725 accuracy

44 Joint Average Result We then tried to combine thetwo aspects as a judgment First of all we found the newparameter through format (7) By setting different 120572 wedetermined the different weight of each aspect The newparameter is as in Figure 9 It contains the new crite-ria produced with rank and gray level values of normaland abnormal sperms 120572 is the argument that represents

BioMed Research International 7

Table 2 Waveform extraction from both methods

Num Normal Symmetry Mid

1

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

25

20

15

10

5

0

2

0100200300400500600700800

1 9 17 25 33 41 49 57 65 73 81 89 970

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

3

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

4

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

5

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8502468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

6

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

7

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

8

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

9

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

8 BioMed Research International

Table 2 Continued

10

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

Num Abnormal Symmetry Mid

81

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

30

1 11 21 31 41 51 61 71 81 91 101

111

121

82

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

83

0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8505

101520253035

1 13 25 37 49 61 73 85 97 109

121

133

84

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 790

10203040506070

1 7 13 19 25 31 37 43 49 55 61 67 73 79

85

050

100150200250300350400

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

86

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

1012141618

1 7 13 19 25 31 37 43 49 55 61 67 73 79

87

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

88

0200400600800

10001200140016001800

1 7 13 19 25 31 37 43 49 55 61 67 73 7905

101520253035

1 7 13 19 25 31 37 43 49 55 61 67 73 79

BioMed Research International 9

Table 2 Continued

89

0200400600800

100012001400

1 7 13 19 25 31 37 43 49 55 61 67 73 790

5

10

15

20

25

30

1 7 13 19 25 31 37 43 49 55 61 67 73 79

90

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

Table 3 The waveform of different starting points from the first method

Startpoint Result

Start 1

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

Start 2

0

200

400

600

800

1000

1200

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800900

1000

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97

Table 4 Dividing line of C

120572 01 02 03 04 05 06 07 08 09DL 177 165 160 149 134 123 114 101 95

Table 5 Accuracy of different 120572

120572 01 02 03 04 05 06 07 08 09Acc 061 062 064 066 069 072 074 077 073

the weighted average rank difference taken in the criteriaWith the new parameter we can calculate its dividing line(DL) as a threshold

In Table 4 row DL represents the dividing line of each120572 value With the dividing line we can classify the spermsAssuming we have already known the exact classification ofeach sperm we can tell how many of them are correctlyclassified and thus we can come up with the accuracy Theaccuracy (Acc) of the different 120572 is as shown in Table 5

In Table 5 the row Acc represents the accuracy of each 120572value According to the result we chose the value 08 for 120572 Inthis case the accuracy can reach 76875

45 Fused SVM Result With all the data provided abovewe took advantage of the SVM method to undertake the

classification and used LIBSVM Software from httpwwwcsientuedutwsimcjlinlibsvm As for the parametersthe default values are used in the implementation and theSVM type is C-SVC the kernel function type is RBF function119890minus120574(119906minusV)2 = 1119896 cost = 1 cache size = 40 eps = 0001shrinking = 1 weight = 1

We input the training samples and test samples as men-tioned in Section 33 With ten training samples and testsamples of 160 sperms which consist of 80 normal and 80abnormal ones we obtain accuracy with 889

5 Comparison Evaluation

In this section we introduce the three methods we usedas comparisons to our method K-nearest neighbor methodworks quite effective as well the other two fail to satisfy us

51 Compared K-Nearest Neighbour Approach TheK-nearestneighbour method showed the following differences

We took the 160 images of sperms as the original clas-sifications and then we input the test image transformedall the images into gray and then calculated the differencebetween the test image and the original images Then wepicked the ten images with the smaller differences Aftercounting the classifications of the ten images we specified theone occurring most often to be the classification of the test

10 BioMed Research International

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a) Normal sperms

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(b) Abnormal sperms

Figure 6 Normal and abnormal sperms and their corresponding data

1 7 13 19 25 31 37 43 49 55 61 67 73 790

102030405060708090

100

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

(b)

Figure 7 Grey level value and percentage of normal sperms

0

20

40

60

80

100

120

140

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

07

(b)

Figure 8 Grey level value and percentage of abnormal sperms

image The classification of the top ten nearest neighbours isas in Figure 10

Taking the sperms whose sum value is larger than 3 asnormal and the sperms whose sum value is smaller than 4as abnormal we can tell that only 39 sperms are mistakenlyclassified providing accuracy of 75625

52 Compared SIFTApproach Lowe summed up the existingfeature detection method based on invariants technologyin 2004 and formally proposed an image scaling rotationand even affine transformation for invariant image with localfeature description operator based on scale space SIFT [22]

The SIFT algorithm first undertakes feature detection in scalespace and defines the key pointsrsquo positions and the scale ofthe key points and then it uses the main direction of theneighbourhood gradient of the key points as the directionfeatures of the points in order to achieve the operatorindependence of scale and the direction The format whichproduces the scale space is as follows

119871 (119909 119910 119905) = intinfin

120585=minusinfin

intinfin

120578

1

2120587119905119890minus(1205852+1205782)2119905

119891 (119909 minus 120585 119910 minus 120578) 119889120585 119889120578

(11)

BioMed Research International 11

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(a) 120572 = 01

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(b) 120572 = 02

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(c) 120572 = 03

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(d) 120572 = 04

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(e) 120572 = 05

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(f) 120572 = 06

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(g) 120572 = 07

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160

(h) 120572 = 08

020406080

100120140

1 13 25 37 49 61 73 85 97 109 121 133 145157

(i) 120572 = 09

Figure 9 The result of joint rank and gray level features of normal and abnormal sperms

12 BioMed Research International

1 7 13 19 25 31 37 43 49 55 61 67 73 790

2

4

6

8

10

12

(a) Normal sperms

1 7 13 19 25 31 37 43 49 55 61 67 73 790

051

152

253

354

45

(b) Abnormal sperms

Figure 10 The classification of the top ten neighbours of (a) normal and (b) abnormal sperms

Rectangularsegment

Segmentation afterrotation

Ellipse

Figure 11 The ellipse model to estimate the contour of the sperm

In format (11) t represents the scaling parameter Byundertaking convolution in the whole domain with a two-dimensionalGaussian kernel and input image we can achievescaling correspondent to t

The SIFT feature vector has the following features (a) it isthe local feature of an image which maintains invariance notonly on rotation scale and brightness variation but also onthe viewing angle the affine transformation and the noise(b) it is distinctive and informative and suitable for fastaccurate matching in a mass signature database (c) it canproduce a large number of SIFT feature vectors with fewobjects (d) it is of high speed

Matlab source code of SIFT is fromhttpwwwvlfeatorgindexhtml [27] We used the SIFT method to extract featurepoints of all the sperm images and achieved the ten normaland ten abnormal sperm images as in the training modelThen we calculated the difference between the trainingmodels and the other 70 normal and 70 abnormal spermimages With the dividing line we came up with accuracyof 50 The result turned out to be bad because the inputdata are simple sperm images and the SIFT algorithm cannotextract enough feature descriptors for classification

53 Compared Ellipse Model Approach With the image ofa single sperm we can use the ellipse model to estimatethe contour of the sperm First of all we used a rectangularsegment to cut out the sperm In order to make the rectangleclose to the sperm edge we needed to choose the rectanglewith the smallest area we rotated the sperm so that thesegmentation would be easier We calculated the left-mostupper-most downward-most and right-most points on theedges and built the rectangular segment based on them asdepicted in Figure 11

76888890

7563

5000

6625

000100020003000400050006000700080009000

10000

Jointaverage

K-neighbor SIFT Ellipticmodel

FusedSVM

()

Figure 12 Accuracy comparison of all methods

With the rectangle we achieved an ellipse to estimatethe sperm whose long axis radius is half the length of therectanglersquos long side and whose short axis radius is half thelength of the rectanglersquos short side As a result we ascertainedthe ovality which is the ratio of the length of the long axisto the short axis Through our experiment on 80 sperms wecould tell whether the sperm was too close to a circle whoseratio is close to 1 and whether the sperm was too slim whoseratio is close to 2 or even larger than 2 We took those whoseratio was larger than 12 and smaller than 18 as normal andthus we achieved accuracy of 6625

6 Conclusion

As shown in Figure 12 we have presented an approachfor the classification of sperm images with one-dimensionalwaveform and gray level features In the comprehensiveevaluation the joint average approach method was applied

BioMed Research International 13

to 160 sperm image samples and provided 76875 accuracyof judgment We then applied the fused SVM method tothe images and reached accuracy of 875 The result provesthat the proposed approach is superior to the previousapproaches

References

[1] N R Pal and S K Pal ldquoA review on image segmentationtechniquesrdquo Pattern Recognition vol 26 no 9 pp 1277ndash12941993

[2] A Yezzi L Zollei and T Kapur ldquoA variational frameworkfor integrating segmentation and registration through activecontoursrdquoMedical ImageAnalysis vol 7 no 2 pp 171ndash185 2003

[3] R Chrastek M Wolf K Donath et al ldquoAutomated segmenta-tion of the optic nerve head for diagnosis of glaucomardquoMedicalImage Analysis vol 9 no 4 pp 297ndash314 2005

[4] H W Kang and S Y Shin ldquoEnhanced lane interactive imagesegmentation by incremental path map constructionrdquo Graphi-cal Models vol 64 no 5 pp 282ndash303 2002

[5] J Peters O Ecabert C Meyer R Kneser and J Weese ldquoOpti-mizing boundary detection via Simulated Search with appli-cations to multi-modal heart segmentationrdquo Medical ImageAnalysis vol 14 no 1 pp 70ndash84 2010

[6] G Behiels F Maes D Vandermeulen and P Suetens ldquoEvalu-ation of image features and search strategies for segmentationof bone structures in radiographs using Active Shape ModelsrdquoMedical Image Analysis vol 6 no 1 pp 47ndash62 2002

[7] M Blanchard K Haguenoer A Apert et al ldquoSperm morphol-ogy assessment using Davidrsquos classiffication time to switch tostrict criteria Prospective comparative analysis in a selectedIVF populationrdquo International Journal of Andrology vol 34 no2 pp 145ndash152 2010

[8] MKass AWitkin andD Terzopoulos ldquoSnakes active contourmodelsrdquo International Journal of Computer Vision vol 1 no 4pp 321ndash331 1988

[9] TMcInerney andD Terzopoulos ldquoT-snakes topology adaptivesnakesrdquoMedical Image Analysis vol 4 no 2 pp 73ndash91 2000

[10] J Liang T McInerney and D Terzopoulos ldquoUnited SnakesrdquoMedical Image Analysis vol 10 no 2 pp 215ndash233 2006

[11] M Zhao Z Meng K K Tseng J S Pan and C Y Hsu ldquoSym-metry auto-detection based on contour and corner modelsrdquo inProceedings of the 5th International Conference on Genetic andEvolutionary Computing (ICGEC rsquo11) pp 345ndash349 September2011

[12] Y Mingqiang K Kidiyo and R Joseph ldquoA survey of shapefeature extraction techniquesrdquo Pattern Recognition vol 15 no7 pp 43ndash90 2008

[13] Y Wang and H Huang ldquoAnalysis of Human Heartbeat withEKG Signalsrdquo 2010

[14] Q Hua A Ji and Q He ldquoMultiple real-valued K nearestneighbor classifiers system by feature groupingrdquo in Proceedingsof the IEEE International Conference on Systems Man andCybernetics (SMC rsquo10) pp 3922ndash3925 October 2010

[15] P Viswanath and T Hitendra Sarma ldquoAn improvement to k-nearest neighbor classifierrdquo in Proceedings of the IEEE RecentAdvances in Intelligent Computational Systems (RAICS rsquo11) pp227ndash231 September 2011

[16] P Wu and Q Chen ldquoA novel SVM-based edge detectionmethodrdquo Physics Procedia vol 24 pp 2075ndash2082 2012

[17] D N Sotiropoulos and G A Tsihrintzis ldquoArtificial immunesystem-based classification in class-imbalanced image prob-lemsrdquo in Proceedings of the 8th International Conference onIntelligent InformationHiding andMultimedia Signal Processingpp 138ndash141 July 2012

[18] M Ramon FMartınez-Pastor O Garcıa-Alvarez et al ldquoTakingadvantage of the use of supervised learning methods forcharacterization of sperm population structure related withfreezability in the Iberian red deerrdquoTheriogenology vol 77 no8 pp 1661ndash1672 2012

[19] S G Goodson Z Zhang J K Tsuruta W Wang and DA OrsquoBrien ldquoClassification of mouse sperm motility patternsusing an automatedmulticlass support vectormachinesmodelrdquoBiology of Reproduction vol 84 no 6 pp 1207ndash1215 2011

[20] D T Lin and D C Pan ldquoIntegrating a mixed-feature modeland multiclass support vector machine for facial expressionrecognitionrdquo Integrated Computer-Aided Engineering vol 16no 1 pp 61ndash74 2009

[21] C G Cheng Y M Tian and W Y Jin ldquoA study on the earlydetection of colon cancer using the methods of wavelet featureextraction and SVM classifications of FTIRrdquo Spectroscopy vol22 no 5 pp 397ndash404 2008

[22] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004

[23] V R Nafisi M H Moradi and M H Nasr-Esfahani ldquoSpermidentification using elliptic model and tail detectionrdquo WorldAcademy of Science Engineering and Technology vol 6 pp 205ndash208 2005

[24] W J Yi K S Park and J S Paick ldquoParameterized charac-terization of elliptic sperm heads using Fourier representationand wavelet transformrdquo in Proceedings of the 20th AnnualInternational Conference of the IEEE Engineering in Medicineand Biology Society vol 20 pp 974ndash977 1998

[25] C C Chang andC J Lin LIBSVMmdasha library for support vectormachines Version 312 httpwwwcsientuedutwsimcjlinlibsvm

[26] CW Hsu C C Chang and C J Lin ldquoA practical guide to sup-port vector classificationrdquo 2010 httpwwwcsientuedutwsimcjlin

[27] A Vedaldi and B Fulkerson ldquoVLFeat an open and portablelibrary of computer vision algorithmsrdquo 2008 httpwwwvlfeatorg

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

BioMed Research International 7

Table 2 Waveform extraction from both methods

Num Normal Symmetry Mid

1

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

25

20

15

10

5

0

2

0100200300400500600700800

1 9 17 25 33 41 49 57 65 73 81 89 970

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

3

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

4

0100200300400500600700800900

1 9 17 25 33 41 49 57 65 73 81 890

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

5

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8502468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

6

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91

7

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89 97

8

0100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

9

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

8 BioMed Research International

Table 2 Continued

10

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

Num Abnormal Symmetry Mid

81

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

30

1 11 21 31 41 51 61 71 81 91 101

111

121

82

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

83

0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8505

101520253035

1 13 25 37 49 61 73 85 97 109

121

133

84

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 790

10203040506070

1 7 13 19 25 31 37 43 49 55 61 67 73 79

85

050

100150200250300350400

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

86

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

1012141618

1 7 13 19 25 31 37 43 49 55 61 67 73 79

87

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

88

0200400600800

10001200140016001800

1 7 13 19 25 31 37 43 49 55 61 67 73 7905

101520253035

1 7 13 19 25 31 37 43 49 55 61 67 73 79

BioMed Research International 9

Table 2 Continued

89

0200400600800

100012001400

1 7 13 19 25 31 37 43 49 55 61 67 73 790

5

10

15

20

25

30

1 7 13 19 25 31 37 43 49 55 61 67 73 79

90

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

Table 3 The waveform of different starting points from the first method

Startpoint Result

Start 1

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

Start 2

0

200

400

600

800

1000

1200

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800900

1000

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97

Table 4 Dividing line of C

120572 01 02 03 04 05 06 07 08 09DL 177 165 160 149 134 123 114 101 95

Table 5 Accuracy of different 120572

120572 01 02 03 04 05 06 07 08 09Acc 061 062 064 066 069 072 074 077 073

the weighted average rank difference taken in the criteriaWith the new parameter we can calculate its dividing line(DL) as a threshold

In Table 4 row DL represents the dividing line of each120572 value With the dividing line we can classify the spermsAssuming we have already known the exact classification ofeach sperm we can tell how many of them are correctlyclassified and thus we can come up with the accuracy Theaccuracy (Acc) of the different 120572 is as shown in Table 5

In Table 5 the row Acc represents the accuracy of each 120572value According to the result we chose the value 08 for 120572 Inthis case the accuracy can reach 76875

45 Fused SVM Result With all the data provided abovewe took advantage of the SVM method to undertake the

classification and used LIBSVM Software from httpwwwcsientuedutwsimcjlinlibsvm As for the parametersthe default values are used in the implementation and theSVM type is C-SVC the kernel function type is RBF function119890minus120574(119906minusV)2 = 1119896 cost = 1 cache size = 40 eps = 0001shrinking = 1 weight = 1

We input the training samples and test samples as men-tioned in Section 33 With ten training samples and testsamples of 160 sperms which consist of 80 normal and 80abnormal ones we obtain accuracy with 889

5 Comparison Evaluation

In this section we introduce the three methods we usedas comparisons to our method K-nearest neighbor methodworks quite effective as well the other two fail to satisfy us

51 Compared K-Nearest Neighbour Approach TheK-nearestneighbour method showed the following differences

We took the 160 images of sperms as the original clas-sifications and then we input the test image transformedall the images into gray and then calculated the differencebetween the test image and the original images Then wepicked the ten images with the smaller differences Aftercounting the classifications of the ten images we specified theone occurring most often to be the classification of the test

10 BioMed Research International

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a) Normal sperms

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(b) Abnormal sperms

Figure 6 Normal and abnormal sperms and their corresponding data

1 7 13 19 25 31 37 43 49 55 61 67 73 790

102030405060708090

100

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

(b)

Figure 7 Grey level value and percentage of normal sperms

0

20

40

60

80

100

120

140

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

07

(b)

Figure 8 Grey level value and percentage of abnormal sperms

image The classification of the top ten nearest neighbours isas in Figure 10

Taking the sperms whose sum value is larger than 3 asnormal and the sperms whose sum value is smaller than 4as abnormal we can tell that only 39 sperms are mistakenlyclassified providing accuracy of 75625

52 Compared SIFTApproach Lowe summed up the existingfeature detection method based on invariants technologyin 2004 and formally proposed an image scaling rotationand even affine transformation for invariant image with localfeature description operator based on scale space SIFT [22]

The SIFT algorithm first undertakes feature detection in scalespace and defines the key pointsrsquo positions and the scale ofthe key points and then it uses the main direction of theneighbourhood gradient of the key points as the directionfeatures of the points in order to achieve the operatorindependence of scale and the direction The format whichproduces the scale space is as follows

119871 (119909 119910 119905) = intinfin

120585=minusinfin

intinfin

120578

1

2120587119905119890minus(1205852+1205782)2119905

119891 (119909 minus 120585 119910 minus 120578) 119889120585 119889120578

(11)

BioMed Research International 11

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(a) 120572 = 01

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(b) 120572 = 02

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(c) 120572 = 03

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(d) 120572 = 04

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(e) 120572 = 05

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(f) 120572 = 06

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(g) 120572 = 07

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160

(h) 120572 = 08

020406080

100120140

1 13 25 37 49 61 73 85 97 109 121 133 145157

(i) 120572 = 09

Figure 9 The result of joint rank and gray level features of normal and abnormal sperms

12 BioMed Research International

1 7 13 19 25 31 37 43 49 55 61 67 73 790

2

4

6

8

10

12

(a) Normal sperms

1 7 13 19 25 31 37 43 49 55 61 67 73 790

051

152

253

354

45

(b) Abnormal sperms

Figure 10 The classification of the top ten neighbours of (a) normal and (b) abnormal sperms

Rectangularsegment

Segmentation afterrotation

Ellipse

Figure 11 The ellipse model to estimate the contour of the sperm

In format (11) t represents the scaling parameter Byundertaking convolution in the whole domain with a two-dimensionalGaussian kernel and input image we can achievescaling correspondent to t

The SIFT feature vector has the following features (a) it isthe local feature of an image which maintains invariance notonly on rotation scale and brightness variation but also onthe viewing angle the affine transformation and the noise(b) it is distinctive and informative and suitable for fastaccurate matching in a mass signature database (c) it canproduce a large number of SIFT feature vectors with fewobjects (d) it is of high speed

Matlab source code of SIFT is fromhttpwwwvlfeatorgindexhtml [27] We used the SIFT method to extract featurepoints of all the sperm images and achieved the ten normaland ten abnormal sperm images as in the training modelThen we calculated the difference between the trainingmodels and the other 70 normal and 70 abnormal spermimages With the dividing line we came up with accuracyof 50 The result turned out to be bad because the inputdata are simple sperm images and the SIFT algorithm cannotextract enough feature descriptors for classification

53 Compared Ellipse Model Approach With the image ofa single sperm we can use the ellipse model to estimatethe contour of the sperm First of all we used a rectangularsegment to cut out the sperm In order to make the rectangleclose to the sperm edge we needed to choose the rectanglewith the smallest area we rotated the sperm so that thesegmentation would be easier We calculated the left-mostupper-most downward-most and right-most points on theedges and built the rectangular segment based on them asdepicted in Figure 11

76888890

7563

5000

6625

000100020003000400050006000700080009000

10000

Jointaverage

K-neighbor SIFT Ellipticmodel

FusedSVM

()

Figure 12 Accuracy comparison of all methods

With the rectangle we achieved an ellipse to estimatethe sperm whose long axis radius is half the length of therectanglersquos long side and whose short axis radius is half thelength of the rectanglersquos short side As a result we ascertainedthe ovality which is the ratio of the length of the long axisto the short axis Through our experiment on 80 sperms wecould tell whether the sperm was too close to a circle whoseratio is close to 1 and whether the sperm was too slim whoseratio is close to 2 or even larger than 2 We took those whoseratio was larger than 12 and smaller than 18 as normal andthus we achieved accuracy of 6625

6 Conclusion

As shown in Figure 12 we have presented an approachfor the classification of sperm images with one-dimensionalwaveform and gray level features In the comprehensiveevaluation the joint average approach method was applied

BioMed Research International 13

to 160 sperm image samples and provided 76875 accuracyof judgment We then applied the fused SVM method tothe images and reached accuracy of 875 The result provesthat the proposed approach is superior to the previousapproaches

References

[1] N R Pal and S K Pal ldquoA review on image segmentationtechniquesrdquo Pattern Recognition vol 26 no 9 pp 1277ndash12941993

[2] A Yezzi L Zollei and T Kapur ldquoA variational frameworkfor integrating segmentation and registration through activecontoursrdquoMedical ImageAnalysis vol 7 no 2 pp 171ndash185 2003

[3] R Chrastek M Wolf K Donath et al ldquoAutomated segmenta-tion of the optic nerve head for diagnosis of glaucomardquoMedicalImage Analysis vol 9 no 4 pp 297ndash314 2005

[4] H W Kang and S Y Shin ldquoEnhanced lane interactive imagesegmentation by incremental path map constructionrdquo Graphi-cal Models vol 64 no 5 pp 282ndash303 2002

[5] J Peters O Ecabert C Meyer R Kneser and J Weese ldquoOpti-mizing boundary detection via Simulated Search with appli-cations to multi-modal heart segmentationrdquo Medical ImageAnalysis vol 14 no 1 pp 70ndash84 2010

[6] G Behiels F Maes D Vandermeulen and P Suetens ldquoEvalu-ation of image features and search strategies for segmentationof bone structures in radiographs using Active Shape ModelsrdquoMedical Image Analysis vol 6 no 1 pp 47ndash62 2002

[7] M Blanchard K Haguenoer A Apert et al ldquoSperm morphol-ogy assessment using Davidrsquos classiffication time to switch tostrict criteria Prospective comparative analysis in a selectedIVF populationrdquo International Journal of Andrology vol 34 no2 pp 145ndash152 2010

[8] MKass AWitkin andD Terzopoulos ldquoSnakes active contourmodelsrdquo International Journal of Computer Vision vol 1 no 4pp 321ndash331 1988

[9] TMcInerney andD Terzopoulos ldquoT-snakes topology adaptivesnakesrdquoMedical Image Analysis vol 4 no 2 pp 73ndash91 2000

[10] J Liang T McInerney and D Terzopoulos ldquoUnited SnakesrdquoMedical Image Analysis vol 10 no 2 pp 215ndash233 2006

[11] M Zhao Z Meng K K Tseng J S Pan and C Y Hsu ldquoSym-metry auto-detection based on contour and corner modelsrdquo inProceedings of the 5th International Conference on Genetic andEvolutionary Computing (ICGEC rsquo11) pp 345ndash349 September2011

[12] Y Mingqiang K Kidiyo and R Joseph ldquoA survey of shapefeature extraction techniquesrdquo Pattern Recognition vol 15 no7 pp 43ndash90 2008

[13] Y Wang and H Huang ldquoAnalysis of Human Heartbeat withEKG Signalsrdquo 2010

[14] Q Hua A Ji and Q He ldquoMultiple real-valued K nearestneighbor classifiers system by feature groupingrdquo in Proceedingsof the IEEE International Conference on Systems Man andCybernetics (SMC rsquo10) pp 3922ndash3925 October 2010

[15] P Viswanath and T Hitendra Sarma ldquoAn improvement to k-nearest neighbor classifierrdquo in Proceedings of the IEEE RecentAdvances in Intelligent Computational Systems (RAICS rsquo11) pp227ndash231 September 2011

[16] P Wu and Q Chen ldquoA novel SVM-based edge detectionmethodrdquo Physics Procedia vol 24 pp 2075ndash2082 2012

[17] D N Sotiropoulos and G A Tsihrintzis ldquoArtificial immunesystem-based classification in class-imbalanced image prob-lemsrdquo in Proceedings of the 8th International Conference onIntelligent InformationHiding andMultimedia Signal Processingpp 138ndash141 July 2012

[18] M Ramon FMartınez-Pastor O Garcıa-Alvarez et al ldquoTakingadvantage of the use of supervised learning methods forcharacterization of sperm population structure related withfreezability in the Iberian red deerrdquoTheriogenology vol 77 no8 pp 1661ndash1672 2012

[19] S G Goodson Z Zhang J K Tsuruta W Wang and DA OrsquoBrien ldquoClassification of mouse sperm motility patternsusing an automatedmulticlass support vectormachinesmodelrdquoBiology of Reproduction vol 84 no 6 pp 1207ndash1215 2011

[20] D T Lin and D C Pan ldquoIntegrating a mixed-feature modeland multiclass support vector machine for facial expressionrecognitionrdquo Integrated Computer-Aided Engineering vol 16no 1 pp 61ndash74 2009

[21] C G Cheng Y M Tian and W Y Jin ldquoA study on the earlydetection of colon cancer using the methods of wavelet featureextraction and SVM classifications of FTIRrdquo Spectroscopy vol22 no 5 pp 397ndash404 2008

[22] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004

[23] V R Nafisi M H Moradi and M H Nasr-Esfahani ldquoSpermidentification using elliptic model and tail detectionrdquo WorldAcademy of Science Engineering and Technology vol 6 pp 205ndash208 2005

[24] W J Yi K S Park and J S Paick ldquoParameterized charac-terization of elliptic sperm heads using Fourier representationand wavelet transformrdquo in Proceedings of the 20th AnnualInternational Conference of the IEEE Engineering in Medicineand Biology Society vol 20 pp 974ndash977 1998

[25] C C Chang andC J Lin LIBSVMmdasha library for support vectormachines Version 312 httpwwwcsientuedutwsimcjlinlibsvm

[26] CW Hsu C C Chang and C J Lin ldquoA practical guide to sup-port vector classificationrdquo 2010 httpwwwcsientuedutwsimcjlin

[27] A Vedaldi and B Fulkerson ldquoVLFeat an open and portablelibrary of computer vision algorithmsrdquo 2008 httpwwwvlfeatorg

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

8 BioMed Research International

Table 2 Continued

10

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

Num Abnormal Symmetry Mid

81

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

5

10

15

20

25

30

1 11 21 31 41 51 61 71 81 91 101

111

121

82

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 0

5

10

15

20

25

1 9 17 25 33 41 49 57 65 73 81 89

83

0

5

10

15

20

25

1 7 13 19 25 31 37 43 49 55 61 67 73 79 8505

101520253035

1 13 25 37 49 61 73 85 97 109

121

133

84

0

100

200

300

400

500

600

1 7 13 19 25 31 37 43 49 55 61 67 73 790

10203040506070

1 7 13 19 25 31 37 43 49 55 61 67 73 79

85

050

100150200250300350400

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

86

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

1012141618

1 7 13 19 25 31 37 43 49 55 61 67 73 79

87

050

100150200250300350400450

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

88

0200400600800

10001200140016001800

1 7 13 19 25 31 37 43 49 55 61 67 73 7905

101520253035

1 7 13 19 25 31 37 43 49 55 61 67 73 79

BioMed Research International 9

Table 2 Continued

89

0200400600800

100012001400

1 7 13 19 25 31 37 43 49 55 61 67 73 790

5

10

15

20

25

30

1 7 13 19 25 31 37 43 49 55 61 67 73 79

90

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

Table 3 The waveform of different starting points from the first method

Startpoint Result

Start 1

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

Start 2

0

200

400

600

800

1000

1200

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800900

1000

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97

Table 4 Dividing line of C

120572 01 02 03 04 05 06 07 08 09DL 177 165 160 149 134 123 114 101 95

Table 5 Accuracy of different 120572

120572 01 02 03 04 05 06 07 08 09Acc 061 062 064 066 069 072 074 077 073

the weighted average rank difference taken in the criteriaWith the new parameter we can calculate its dividing line(DL) as a threshold

In Table 4 row DL represents the dividing line of each120572 value With the dividing line we can classify the spermsAssuming we have already known the exact classification ofeach sperm we can tell how many of them are correctlyclassified and thus we can come up with the accuracy Theaccuracy (Acc) of the different 120572 is as shown in Table 5

In Table 5 the row Acc represents the accuracy of each 120572value According to the result we chose the value 08 for 120572 Inthis case the accuracy can reach 76875

45 Fused SVM Result With all the data provided abovewe took advantage of the SVM method to undertake the

classification and used LIBSVM Software from httpwwwcsientuedutwsimcjlinlibsvm As for the parametersthe default values are used in the implementation and theSVM type is C-SVC the kernel function type is RBF function119890minus120574(119906minusV)2 = 1119896 cost = 1 cache size = 40 eps = 0001shrinking = 1 weight = 1

We input the training samples and test samples as men-tioned in Section 33 With ten training samples and testsamples of 160 sperms which consist of 80 normal and 80abnormal ones we obtain accuracy with 889

5 Comparison Evaluation

In this section we introduce the three methods we usedas comparisons to our method K-nearest neighbor methodworks quite effective as well the other two fail to satisfy us

51 Compared K-Nearest Neighbour Approach TheK-nearestneighbour method showed the following differences

We took the 160 images of sperms as the original clas-sifications and then we input the test image transformedall the images into gray and then calculated the differencebetween the test image and the original images Then wepicked the ten images with the smaller differences Aftercounting the classifications of the ten images we specified theone occurring most often to be the classification of the test

10 BioMed Research International

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a) Normal sperms

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(b) Abnormal sperms

Figure 6 Normal and abnormal sperms and their corresponding data

1 7 13 19 25 31 37 43 49 55 61 67 73 790

102030405060708090

100

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

(b)

Figure 7 Grey level value and percentage of normal sperms

0

20

40

60

80

100

120

140

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

07

(b)

Figure 8 Grey level value and percentage of abnormal sperms

image The classification of the top ten nearest neighbours isas in Figure 10

Taking the sperms whose sum value is larger than 3 asnormal and the sperms whose sum value is smaller than 4as abnormal we can tell that only 39 sperms are mistakenlyclassified providing accuracy of 75625

52 Compared SIFTApproach Lowe summed up the existingfeature detection method based on invariants technologyin 2004 and formally proposed an image scaling rotationand even affine transformation for invariant image with localfeature description operator based on scale space SIFT [22]

The SIFT algorithm first undertakes feature detection in scalespace and defines the key pointsrsquo positions and the scale ofthe key points and then it uses the main direction of theneighbourhood gradient of the key points as the directionfeatures of the points in order to achieve the operatorindependence of scale and the direction The format whichproduces the scale space is as follows

119871 (119909 119910 119905) = intinfin

120585=minusinfin

intinfin

120578

1

2120587119905119890minus(1205852+1205782)2119905

119891 (119909 minus 120585 119910 minus 120578) 119889120585 119889120578

(11)

BioMed Research International 11

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(a) 120572 = 01

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(b) 120572 = 02

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(c) 120572 = 03

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(d) 120572 = 04

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(e) 120572 = 05

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(f) 120572 = 06

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(g) 120572 = 07

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160

(h) 120572 = 08

020406080

100120140

1 13 25 37 49 61 73 85 97 109 121 133 145157

(i) 120572 = 09

Figure 9 The result of joint rank and gray level features of normal and abnormal sperms

12 BioMed Research International

1 7 13 19 25 31 37 43 49 55 61 67 73 790

2

4

6

8

10

12

(a) Normal sperms

1 7 13 19 25 31 37 43 49 55 61 67 73 790

051

152

253

354

45

(b) Abnormal sperms

Figure 10 The classification of the top ten neighbours of (a) normal and (b) abnormal sperms

Rectangularsegment

Segmentation afterrotation

Ellipse

Figure 11 The ellipse model to estimate the contour of the sperm

In format (11) t represents the scaling parameter Byundertaking convolution in the whole domain with a two-dimensionalGaussian kernel and input image we can achievescaling correspondent to t

The SIFT feature vector has the following features (a) it isthe local feature of an image which maintains invariance notonly on rotation scale and brightness variation but also onthe viewing angle the affine transformation and the noise(b) it is distinctive and informative and suitable for fastaccurate matching in a mass signature database (c) it canproduce a large number of SIFT feature vectors with fewobjects (d) it is of high speed

Matlab source code of SIFT is fromhttpwwwvlfeatorgindexhtml [27] We used the SIFT method to extract featurepoints of all the sperm images and achieved the ten normaland ten abnormal sperm images as in the training modelThen we calculated the difference between the trainingmodels and the other 70 normal and 70 abnormal spermimages With the dividing line we came up with accuracyof 50 The result turned out to be bad because the inputdata are simple sperm images and the SIFT algorithm cannotextract enough feature descriptors for classification

53 Compared Ellipse Model Approach With the image ofa single sperm we can use the ellipse model to estimatethe contour of the sperm First of all we used a rectangularsegment to cut out the sperm In order to make the rectangleclose to the sperm edge we needed to choose the rectanglewith the smallest area we rotated the sperm so that thesegmentation would be easier We calculated the left-mostupper-most downward-most and right-most points on theedges and built the rectangular segment based on them asdepicted in Figure 11

76888890

7563

5000

6625

000100020003000400050006000700080009000

10000

Jointaverage

K-neighbor SIFT Ellipticmodel

FusedSVM

()

Figure 12 Accuracy comparison of all methods

With the rectangle we achieved an ellipse to estimatethe sperm whose long axis radius is half the length of therectanglersquos long side and whose short axis radius is half thelength of the rectanglersquos short side As a result we ascertainedthe ovality which is the ratio of the length of the long axisto the short axis Through our experiment on 80 sperms wecould tell whether the sperm was too close to a circle whoseratio is close to 1 and whether the sperm was too slim whoseratio is close to 2 or even larger than 2 We took those whoseratio was larger than 12 and smaller than 18 as normal andthus we achieved accuracy of 6625

6 Conclusion

As shown in Figure 12 we have presented an approachfor the classification of sperm images with one-dimensionalwaveform and gray level features In the comprehensiveevaluation the joint average approach method was applied

BioMed Research International 13

to 160 sperm image samples and provided 76875 accuracyof judgment We then applied the fused SVM method tothe images and reached accuracy of 875 The result provesthat the proposed approach is superior to the previousapproaches

References

[1] N R Pal and S K Pal ldquoA review on image segmentationtechniquesrdquo Pattern Recognition vol 26 no 9 pp 1277ndash12941993

[2] A Yezzi L Zollei and T Kapur ldquoA variational frameworkfor integrating segmentation and registration through activecontoursrdquoMedical ImageAnalysis vol 7 no 2 pp 171ndash185 2003

[3] R Chrastek M Wolf K Donath et al ldquoAutomated segmenta-tion of the optic nerve head for diagnosis of glaucomardquoMedicalImage Analysis vol 9 no 4 pp 297ndash314 2005

[4] H W Kang and S Y Shin ldquoEnhanced lane interactive imagesegmentation by incremental path map constructionrdquo Graphi-cal Models vol 64 no 5 pp 282ndash303 2002

[5] J Peters O Ecabert C Meyer R Kneser and J Weese ldquoOpti-mizing boundary detection via Simulated Search with appli-cations to multi-modal heart segmentationrdquo Medical ImageAnalysis vol 14 no 1 pp 70ndash84 2010

[6] G Behiels F Maes D Vandermeulen and P Suetens ldquoEvalu-ation of image features and search strategies for segmentationof bone structures in radiographs using Active Shape ModelsrdquoMedical Image Analysis vol 6 no 1 pp 47ndash62 2002

[7] M Blanchard K Haguenoer A Apert et al ldquoSperm morphol-ogy assessment using Davidrsquos classiffication time to switch tostrict criteria Prospective comparative analysis in a selectedIVF populationrdquo International Journal of Andrology vol 34 no2 pp 145ndash152 2010

[8] MKass AWitkin andD Terzopoulos ldquoSnakes active contourmodelsrdquo International Journal of Computer Vision vol 1 no 4pp 321ndash331 1988

[9] TMcInerney andD Terzopoulos ldquoT-snakes topology adaptivesnakesrdquoMedical Image Analysis vol 4 no 2 pp 73ndash91 2000

[10] J Liang T McInerney and D Terzopoulos ldquoUnited SnakesrdquoMedical Image Analysis vol 10 no 2 pp 215ndash233 2006

[11] M Zhao Z Meng K K Tseng J S Pan and C Y Hsu ldquoSym-metry auto-detection based on contour and corner modelsrdquo inProceedings of the 5th International Conference on Genetic andEvolutionary Computing (ICGEC rsquo11) pp 345ndash349 September2011

[12] Y Mingqiang K Kidiyo and R Joseph ldquoA survey of shapefeature extraction techniquesrdquo Pattern Recognition vol 15 no7 pp 43ndash90 2008

[13] Y Wang and H Huang ldquoAnalysis of Human Heartbeat withEKG Signalsrdquo 2010

[14] Q Hua A Ji and Q He ldquoMultiple real-valued K nearestneighbor classifiers system by feature groupingrdquo in Proceedingsof the IEEE International Conference on Systems Man andCybernetics (SMC rsquo10) pp 3922ndash3925 October 2010

[15] P Viswanath and T Hitendra Sarma ldquoAn improvement to k-nearest neighbor classifierrdquo in Proceedings of the IEEE RecentAdvances in Intelligent Computational Systems (RAICS rsquo11) pp227ndash231 September 2011

[16] P Wu and Q Chen ldquoA novel SVM-based edge detectionmethodrdquo Physics Procedia vol 24 pp 2075ndash2082 2012

[17] D N Sotiropoulos and G A Tsihrintzis ldquoArtificial immunesystem-based classification in class-imbalanced image prob-lemsrdquo in Proceedings of the 8th International Conference onIntelligent InformationHiding andMultimedia Signal Processingpp 138ndash141 July 2012

[18] M Ramon FMartınez-Pastor O Garcıa-Alvarez et al ldquoTakingadvantage of the use of supervised learning methods forcharacterization of sperm population structure related withfreezability in the Iberian red deerrdquoTheriogenology vol 77 no8 pp 1661ndash1672 2012

[19] S G Goodson Z Zhang J K Tsuruta W Wang and DA OrsquoBrien ldquoClassification of mouse sperm motility patternsusing an automatedmulticlass support vectormachinesmodelrdquoBiology of Reproduction vol 84 no 6 pp 1207ndash1215 2011

[20] D T Lin and D C Pan ldquoIntegrating a mixed-feature modeland multiclass support vector machine for facial expressionrecognitionrdquo Integrated Computer-Aided Engineering vol 16no 1 pp 61ndash74 2009

[21] C G Cheng Y M Tian and W Y Jin ldquoA study on the earlydetection of colon cancer using the methods of wavelet featureextraction and SVM classifications of FTIRrdquo Spectroscopy vol22 no 5 pp 397ndash404 2008

[22] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004

[23] V R Nafisi M H Moradi and M H Nasr-Esfahani ldquoSpermidentification using elliptic model and tail detectionrdquo WorldAcademy of Science Engineering and Technology vol 6 pp 205ndash208 2005

[24] W J Yi K S Park and J S Paick ldquoParameterized charac-terization of elliptic sperm heads using Fourier representationand wavelet transformrdquo in Proceedings of the 20th AnnualInternational Conference of the IEEE Engineering in Medicineand Biology Society vol 20 pp 974ndash977 1998

[25] C C Chang andC J Lin LIBSVMmdasha library for support vectormachines Version 312 httpwwwcsientuedutwsimcjlinlibsvm

[26] CW Hsu C C Chang and C J Lin ldquoA practical guide to sup-port vector classificationrdquo 2010 httpwwwcsientuedutwsimcjlin

[27] A Vedaldi and B Fulkerson ldquoVLFeat an open and portablelibrary of computer vision algorithmsrdquo 2008 httpwwwvlfeatorg

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

BioMed Research International 9

Table 2 Continued

89

0200400600800

100012001400

1 7 13 19 25 31 37 43 49 55 61 67 73 790

5

10

15

20

25

30

1 7 13 19 25 31 37 43 49 55 61 67 73 79

90

0100200300400500600700800900

1 7 13 19 25 31 37 43 49 55 61 67 73 7902468

101214161820

1 7 13 19 25 31 37 43 49 55 61 67 73 79

Table 3 The waveform of different starting points from the first method

Startpoint Result

Start 1

0100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 850

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85

Start 2

0

200

400

600

800

1000

1200

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800900

1000

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 970

100200300400500600700800

1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97

Table 4 Dividing line of C

120572 01 02 03 04 05 06 07 08 09DL 177 165 160 149 134 123 114 101 95

Table 5 Accuracy of different 120572

120572 01 02 03 04 05 06 07 08 09Acc 061 062 064 066 069 072 074 077 073

the weighted average rank difference taken in the criteriaWith the new parameter we can calculate its dividing line(DL) as a threshold

In Table 4 row DL represents the dividing line of each120572 value With the dividing line we can classify the spermsAssuming we have already known the exact classification ofeach sperm we can tell how many of them are correctlyclassified and thus we can come up with the accuracy Theaccuracy (Acc) of the different 120572 is as shown in Table 5

In Table 5 the row Acc represents the accuracy of each 120572value According to the result we chose the value 08 for 120572 Inthis case the accuracy can reach 76875

45 Fused SVM Result With all the data provided abovewe took advantage of the SVM method to undertake the

classification and used LIBSVM Software from httpwwwcsientuedutwsimcjlinlibsvm As for the parametersthe default values are used in the implementation and theSVM type is C-SVC the kernel function type is RBF function119890minus120574(119906minusV)2 = 1119896 cost = 1 cache size = 40 eps = 0001shrinking = 1 weight = 1

We input the training samples and test samples as men-tioned in Section 33 With ten training samples and testsamples of 160 sperms which consist of 80 normal and 80abnormal ones we obtain accuracy with 889

5 Comparison Evaluation

In this section we introduce the three methods we usedas comparisons to our method K-nearest neighbor methodworks quite effective as well the other two fail to satisfy us

51 Compared K-Nearest Neighbour Approach TheK-nearestneighbour method showed the following differences

We took the 160 images of sperms as the original clas-sifications and then we input the test image transformedall the images into gray and then calculated the differencebetween the test image and the original images Then wepicked the ten images with the smaller differences Aftercounting the classifications of the ten images we specified theone occurring most often to be the classification of the test

10 BioMed Research International

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a) Normal sperms

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(b) Abnormal sperms

Figure 6 Normal and abnormal sperms and their corresponding data

1 7 13 19 25 31 37 43 49 55 61 67 73 790

102030405060708090

100

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

(b)

Figure 7 Grey level value and percentage of normal sperms

0

20

40

60

80

100

120

140

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

07

(b)

Figure 8 Grey level value and percentage of abnormal sperms

image The classification of the top ten nearest neighbours isas in Figure 10

Taking the sperms whose sum value is larger than 3 asnormal and the sperms whose sum value is smaller than 4as abnormal we can tell that only 39 sperms are mistakenlyclassified providing accuracy of 75625

52 Compared SIFTApproach Lowe summed up the existingfeature detection method based on invariants technologyin 2004 and formally proposed an image scaling rotationand even affine transformation for invariant image with localfeature description operator based on scale space SIFT [22]

The SIFT algorithm first undertakes feature detection in scalespace and defines the key pointsrsquo positions and the scale ofthe key points and then it uses the main direction of theneighbourhood gradient of the key points as the directionfeatures of the points in order to achieve the operatorindependence of scale and the direction The format whichproduces the scale space is as follows

119871 (119909 119910 119905) = intinfin

120585=minusinfin

intinfin

120578

1

2120587119905119890minus(1205852+1205782)2119905

119891 (119909 minus 120585 119910 minus 120578) 119889120585 119889120578

(11)

BioMed Research International 11

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(a) 120572 = 01

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(b) 120572 = 02

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(c) 120572 = 03

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(d) 120572 = 04

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(e) 120572 = 05

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(f) 120572 = 06

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(g) 120572 = 07

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160

(h) 120572 = 08

020406080

100120140

1 13 25 37 49 61 73 85 97 109 121 133 145157

(i) 120572 = 09

Figure 9 The result of joint rank and gray level features of normal and abnormal sperms

12 BioMed Research International

1 7 13 19 25 31 37 43 49 55 61 67 73 790

2

4

6

8

10

12

(a) Normal sperms

1 7 13 19 25 31 37 43 49 55 61 67 73 790

051

152

253

354

45

(b) Abnormal sperms

Figure 10 The classification of the top ten neighbours of (a) normal and (b) abnormal sperms

Rectangularsegment

Segmentation afterrotation

Ellipse

Figure 11 The ellipse model to estimate the contour of the sperm

In format (11) t represents the scaling parameter Byundertaking convolution in the whole domain with a two-dimensionalGaussian kernel and input image we can achievescaling correspondent to t

The SIFT feature vector has the following features (a) it isthe local feature of an image which maintains invariance notonly on rotation scale and brightness variation but also onthe viewing angle the affine transformation and the noise(b) it is distinctive and informative and suitable for fastaccurate matching in a mass signature database (c) it canproduce a large number of SIFT feature vectors with fewobjects (d) it is of high speed

Matlab source code of SIFT is fromhttpwwwvlfeatorgindexhtml [27] We used the SIFT method to extract featurepoints of all the sperm images and achieved the ten normaland ten abnormal sperm images as in the training modelThen we calculated the difference between the trainingmodels and the other 70 normal and 70 abnormal spermimages With the dividing line we came up with accuracyof 50 The result turned out to be bad because the inputdata are simple sperm images and the SIFT algorithm cannotextract enough feature descriptors for classification

53 Compared Ellipse Model Approach With the image ofa single sperm we can use the ellipse model to estimatethe contour of the sperm First of all we used a rectangularsegment to cut out the sperm In order to make the rectangleclose to the sperm edge we needed to choose the rectanglewith the smallest area we rotated the sperm so that thesegmentation would be easier We calculated the left-mostupper-most downward-most and right-most points on theedges and built the rectangular segment based on them asdepicted in Figure 11

76888890

7563

5000

6625

000100020003000400050006000700080009000

10000

Jointaverage

K-neighbor SIFT Ellipticmodel

FusedSVM

()

Figure 12 Accuracy comparison of all methods

With the rectangle we achieved an ellipse to estimatethe sperm whose long axis radius is half the length of therectanglersquos long side and whose short axis radius is half thelength of the rectanglersquos short side As a result we ascertainedthe ovality which is the ratio of the length of the long axisto the short axis Through our experiment on 80 sperms wecould tell whether the sperm was too close to a circle whoseratio is close to 1 and whether the sperm was too slim whoseratio is close to 2 or even larger than 2 We took those whoseratio was larger than 12 and smaller than 18 as normal andthus we achieved accuracy of 6625

6 Conclusion

As shown in Figure 12 we have presented an approachfor the classification of sperm images with one-dimensionalwaveform and gray level features In the comprehensiveevaluation the joint average approach method was applied

BioMed Research International 13

to 160 sperm image samples and provided 76875 accuracyof judgment We then applied the fused SVM method tothe images and reached accuracy of 875 The result provesthat the proposed approach is superior to the previousapproaches

References

[1] N R Pal and S K Pal ldquoA review on image segmentationtechniquesrdquo Pattern Recognition vol 26 no 9 pp 1277ndash12941993

[2] A Yezzi L Zollei and T Kapur ldquoA variational frameworkfor integrating segmentation and registration through activecontoursrdquoMedical ImageAnalysis vol 7 no 2 pp 171ndash185 2003

[3] R Chrastek M Wolf K Donath et al ldquoAutomated segmenta-tion of the optic nerve head for diagnosis of glaucomardquoMedicalImage Analysis vol 9 no 4 pp 297ndash314 2005

[4] H W Kang and S Y Shin ldquoEnhanced lane interactive imagesegmentation by incremental path map constructionrdquo Graphi-cal Models vol 64 no 5 pp 282ndash303 2002

[5] J Peters O Ecabert C Meyer R Kneser and J Weese ldquoOpti-mizing boundary detection via Simulated Search with appli-cations to multi-modal heart segmentationrdquo Medical ImageAnalysis vol 14 no 1 pp 70ndash84 2010

[6] G Behiels F Maes D Vandermeulen and P Suetens ldquoEvalu-ation of image features and search strategies for segmentationof bone structures in radiographs using Active Shape ModelsrdquoMedical Image Analysis vol 6 no 1 pp 47ndash62 2002

[7] M Blanchard K Haguenoer A Apert et al ldquoSperm morphol-ogy assessment using Davidrsquos classiffication time to switch tostrict criteria Prospective comparative analysis in a selectedIVF populationrdquo International Journal of Andrology vol 34 no2 pp 145ndash152 2010

[8] MKass AWitkin andD Terzopoulos ldquoSnakes active contourmodelsrdquo International Journal of Computer Vision vol 1 no 4pp 321ndash331 1988

[9] TMcInerney andD Terzopoulos ldquoT-snakes topology adaptivesnakesrdquoMedical Image Analysis vol 4 no 2 pp 73ndash91 2000

[10] J Liang T McInerney and D Terzopoulos ldquoUnited SnakesrdquoMedical Image Analysis vol 10 no 2 pp 215ndash233 2006

[11] M Zhao Z Meng K K Tseng J S Pan and C Y Hsu ldquoSym-metry auto-detection based on contour and corner modelsrdquo inProceedings of the 5th International Conference on Genetic andEvolutionary Computing (ICGEC rsquo11) pp 345ndash349 September2011

[12] Y Mingqiang K Kidiyo and R Joseph ldquoA survey of shapefeature extraction techniquesrdquo Pattern Recognition vol 15 no7 pp 43ndash90 2008

[13] Y Wang and H Huang ldquoAnalysis of Human Heartbeat withEKG Signalsrdquo 2010

[14] Q Hua A Ji and Q He ldquoMultiple real-valued K nearestneighbor classifiers system by feature groupingrdquo in Proceedingsof the IEEE International Conference on Systems Man andCybernetics (SMC rsquo10) pp 3922ndash3925 October 2010

[15] P Viswanath and T Hitendra Sarma ldquoAn improvement to k-nearest neighbor classifierrdquo in Proceedings of the IEEE RecentAdvances in Intelligent Computational Systems (RAICS rsquo11) pp227ndash231 September 2011

[16] P Wu and Q Chen ldquoA novel SVM-based edge detectionmethodrdquo Physics Procedia vol 24 pp 2075ndash2082 2012

[17] D N Sotiropoulos and G A Tsihrintzis ldquoArtificial immunesystem-based classification in class-imbalanced image prob-lemsrdquo in Proceedings of the 8th International Conference onIntelligent InformationHiding andMultimedia Signal Processingpp 138ndash141 July 2012

[18] M Ramon FMartınez-Pastor O Garcıa-Alvarez et al ldquoTakingadvantage of the use of supervised learning methods forcharacterization of sperm population structure related withfreezability in the Iberian red deerrdquoTheriogenology vol 77 no8 pp 1661ndash1672 2012

[19] S G Goodson Z Zhang J K Tsuruta W Wang and DA OrsquoBrien ldquoClassification of mouse sperm motility patternsusing an automatedmulticlass support vectormachinesmodelrdquoBiology of Reproduction vol 84 no 6 pp 1207ndash1215 2011

[20] D T Lin and D C Pan ldquoIntegrating a mixed-feature modeland multiclass support vector machine for facial expressionrecognitionrdquo Integrated Computer-Aided Engineering vol 16no 1 pp 61ndash74 2009

[21] C G Cheng Y M Tian and W Y Jin ldquoA study on the earlydetection of colon cancer using the methods of wavelet featureextraction and SVM classifications of FTIRrdquo Spectroscopy vol22 no 5 pp 397ndash404 2008

[22] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004

[23] V R Nafisi M H Moradi and M H Nasr-Esfahani ldquoSpermidentification using elliptic model and tail detectionrdquo WorldAcademy of Science Engineering and Technology vol 6 pp 205ndash208 2005

[24] W J Yi K S Park and J S Paick ldquoParameterized charac-terization of elliptic sperm heads using Fourier representationand wavelet transformrdquo in Proceedings of the 20th AnnualInternational Conference of the IEEE Engineering in Medicineand Biology Society vol 20 pp 974ndash977 1998

[25] C C Chang andC J Lin LIBSVMmdasha library for support vectormachines Version 312 httpwwwcsientuedutwsimcjlinlibsvm

[26] CW Hsu C C Chang and C J Lin ldquoA practical guide to sup-port vector classificationrdquo 2010 httpwwwcsientuedutwsimcjlin

[27] A Vedaldi and B Fulkerson ldquoVLFeat an open and portablelibrary of computer vision algorithmsrdquo 2008 httpwwwvlfeatorg

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

10 BioMed Research International

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a) Normal sperms

0

50

100

150

200

250

300

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(b) Abnormal sperms

Figure 6 Normal and abnormal sperms and their corresponding data

1 7 13 19 25 31 37 43 49 55 61 67 73 790

102030405060708090

100

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

(b)

Figure 7 Grey level value and percentage of normal sperms

0

20

40

60

80

100

120

140

1 7 13 19 25 31 37 43 49 55 61 67 73 79

(a)

1 7 13 19 25 31 37 43 49 55 61 67 73 790

01

02

03

04

05

06

07

(b)

Figure 8 Grey level value and percentage of abnormal sperms

image The classification of the top ten nearest neighbours isas in Figure 10

Taking the sperms whose sum value is larger than 3 asnormal and the sperms whose sum value is smaller than 4as abnormal we can tell that only 39 sperms are mistakenlyclassified providing accuracy of 75625

52 Compared SIFTApproach Lowe summed up the existingfeature detection method based on invariants technologyin 2004 and formally proposed an image scaling rotationand even affine transformation for invariant image with localfeature description operator based on scale space SIFT [22]

The SIFT algorithm first undertakes feature detection in scalespace and defines the key pointsrsquo positions and the scale ofthe key points and then it uses the main direction of theneighbourhood gradient of the key points as the directionfeatures of the points in order to achieve the operatorindependence of scale and the direction The format whichproduces the scale space is as follows

119871 (119909 119910 119905) = intinfin

120585=minusinfin

intinfin

120578

1

2120587119905119890minus(1205852+1205782)2119905

119891 (119909 minus 120585 119910 minus 120578) 119889120585 119889120578

(11)

BioMed Research International 11

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(a) 120572 = 01

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(b) 120572 = 02

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(c) 120572 = 03

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(d) 120572 = 04

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(e) 120572 = 05

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(f) 120572 = 06

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(g) 120572 = 07

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160

(h) 120572 = 08

020406080

100120140

1 13 25 37 49 61 73 85 97 109 121 133 145157

(i) 120572 = 09

Figure 9 The result of joint rank and gray level features of normal and abnormal sperms

12 BioMed Research International

1 7 13 19 25 31 37 43 49 55 61 67 73 790

2

4

6

8

10

12

(a) Normal sperms

1 7 13 19 25 31 37 43 49 55 61 67 73 790

051

152

253

354

45

(b) Abnormal sperms

Figure 10 The classification of the top ten neighbours of (a) normal and (b) abnormal sperms

Rectangularsegment

Segmentation afterrotation

Ellipse

Figure 11 The ellipse model to estimate the contour of the sperm

In format (11) t represents the scaling parameter Byundertaking convolution in the whole domain with a two-dimensionalGaussian kernel and input image we can achievescaling correspondent to t

The SIFT feature vector has the following features (a) it isthe local feature of an image which maintains invariance notonly on rotation scale and brightness variation but also onthe viewing angle the affine transformation and the noise(b) it is distinctive and informative and suitable for fastaccurate matching in a mass signature database (c) it canproduce a large number of SIFT feature vectors with fewobjects (d) it is of high speed

Matlab source code of SIFT is fromhttpwwwvlfeatorgindexhtml [27] We used the SIFT method to extract featurepoints of all the sperm images and achieved the ten normaland ten abnormal sperm images as in the training modelThen we calculated the difference between the trainingmodels and the other 70 normal and 70 abnormal spermimages With the dividing line we came up with accuracyof 50 The result turned out to be bad because the inputdata are simple sperm images and the SIFT algorithm cannotextract enough feature descriptors for classification

53 Compared Ellipse Model Approach With the image ofa single sperm we can use the ellipse model to estimatethe contour of the sperm First of all we used a rectangularsegment to cut out the sperm In order to make the rectangleclose to the sperm edge we needed to choose the rectanglewith the smallest area we rotated the sperm so that thesegmentation would be easier We calculated the left-mostupper-most downward-most and right-most points on theedges and built the rectangular segment based on them asdepicted in Figure 11

76888890

7563

5000

6625

000100020003000400050006000700080009000

10000

Jointaverage

K-neighbor SIFT Ellipticmodel

FusedSVM

()

Figure 12 Accuracy comparison of all methods

With the rectangle we achieved an ellipse to estimatethe sperm whose long axis radius is half the length of therectanglersquos long side and whose short axis radius is half thelength of the rectanglersquos short side As a result we ascertainedthe ovality which is the ratio of the length of the long axisto the short axis Through our experiment on 80 sperms wecould tell whether the sperm was too close to a circle whoseratio is close to 1 and whether the sperm was too slim whoseratio is close to 2 or even larger than 2 We took those whoseratio was larger than 12 and smaller than 18 as normal andthus we achieved accuracy of 6625

6 Conclusion

As shown in Figure 12 we have presented an approachfor the classification of sperm images with one-dimensionalwaveform and gray level features In the comprehensiveevaluation the joint average approach method was applied

BioMed Research International 13

to 160 sperm image samples and provided 76875 accuracyof judgment We then applied the fused SVM method tothe images and reached accuracy of 875 The result provesthat the proposed approach is superior to the previousapproaches

References

[1] N R Pal and S K Pal ldquoA review on image segmentationtechniquesrdquo Pattern Recognition vol 26 no 9 pp 1277ndash12941993

[2] A Yezzi L Zollei and T Kapur ldquoA variational frameworkfor integrating segmentation and registration through activecontoursrdquoMedical ImageAnalysis vol 7 no 2 pp 171ndash185 2003

[3] R Chrastek M Wolf K Donath et al ldquoAutomated segmenta-tion of the optic nerve head for diagnosis of glaucomardquoMedicalImage Analysis vol 9 no 4 pp 297ndash314 2005

[4] H W Kang and S Y Shin ldquoEnhanced lane interactive imagesegmentation by incremental path map constructionrdquo Graphi-cal Models vol 64 no 5 pp 282ndash303 2002

[5] J Peters O Ecabert C Meyer R Kneser and J Weese ldquoOpti-mizing boundary detection via Simulated Search with appli-cations to multi-modal heart segmentationrdquo Medical ImageAnalysis vol 14 no 1 pp 70ndash84 2010

[6] G Behiels F Maes D Vandermeulen and P Suetens ldquoEvalu-ation of image features and search strategies for segmentationof bone structures in radiographs using Active Shape ModelsrdquoMedical Image Analysis vol 6 no 1 pp 47ndash62 2002

[7] M Blanchard K Haguenoer A Apert et al ldquoSperm morphol-ogy assessment using Davidrsquos classiffication time to switch tostrict criteria Prospective comparative analysis in a selectedIVF populationrdquo International Journal of Andrology vol 34 no2 pp 145ndash152 2010

[8] MKass AWitkin andD Terzopoulos ldquoSnakes active contourmodelsrdquo International Journal of Computer Vision vol 1 no 4pp 321ndash331 1988

[9] TMcInerney andD Terzopoulos ldquoT-snakes topology adaptivesnakesrdquoMedical Image Analysis vol 4 no 2 pp 73ndash91 2000

[10] J Liang T McInerney and D Terzopoulos ldquoUnited SnakesrdquoMedical Image Analysis vol 10 no 2 pp 215ndash233 2006

[11] M Zhao Z Meng K K Tseng J S Pan and C Y Hsu ldquoSym-metry auto-detection based on contour and corner modelsrdquo inProceedings of the 5th International Conference on Genetic andEvolutionary Computing (ICGEC rsquo11) pp 345ndash349 September2011

[12] Y Mingqiang K Kidiyo and R Joseph ldquoA survey of shapefeature extraction techniquesrdquo Pattern Recognition vol 15 no7 pp 43ndash90 2008

[13] Y Wang and H Huang ldquoAnalysis of Human Heartbeat withEKG Signalsrdquo 2010

[14] Q Hua A Ji and Q He ldquoMultiple real-valued K nearestneighbor classifiers system by feature groupingrdquo in Proceedingsof the IEEE International Conference on Systems Man andCybernetics (SMC rsquo10) pp 3922ndash3925 October 2010

[15] P Viswanath and T Hitendra Sarma ldquoAn improvement to k-nearest neighbor classifierrdquo in Proceedings of the IEEE RecentAdvances in Intelligent Computational Systems (RAICS rsquo11) pp227ndash231 September 2011

[16] P Wu and Q Chen ldquoA novel SVM-based edge detectionmethodrdquo Physics Procedia vol 24 pp 2075ndash2082 2012

[17] D N Sotiropoulos and G A Tsihrintzis ldquoArtificial immunesystem-based classification in class-imbalanced image prob-lemsrdquo in Proceedings of the 8th International Conference onIntelligent InformationHiding andMultimedia Signal Processingpp 138ndash141 July 2012

[18] M Ramon FMartınez-Pastor O Garcıa-Alvarez et al ldquoTakingadvantage of the use of supervised learning methods forcharacterization of sperm population structure related withfreezability in the Iberian red deerrdquoTheriogenology vol 77 no8 pp 1661ndash1672 2012

[19] S G Goodson Z Zhang J K Tsuruta W Wang and DA OrsquoBrien ldquoClassification of mouse sperm motility patternsusing an automatedmulticlass support vectormachinesmodelrdquoBiology of Reproduction vol 84 no 6 pp 1207ndash1215 2011

[20] D T Lin and D C Pan ldquoIntegrating a mixed-feature modeland multiclass support vector machine for facial expressionrecognitionrdquo Integrated Computer-Aided Engineering vol 16no 1 pp 61ndash74 2009

[21] C G Cheng Y M Tian and W Y Jin ldquoA study on the earlydetection of colon cancer using the methods of wavelet featureextraction and SVM classifications of FTIRrdquo Spectroscopy vol22 no 5 pp 397ndash404 2008

[22] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004

[23] V R Nafisi M H Moradi and M H Nasr-Esfahani ldquoSpermidentification using elliptic model and tail detectionrdquo WorldAcademy of Science Engineering and Technology vol 6 pp 205ndash208 2005

[24] W J Yi K S Park and J S Paick ldquoParameterized charac-terization of elliptic sperm heads using Fourier representationand wavelet transformrdquo in Proceedings of the 20th AnnualInternational Conference of the IEEE Engineering in Medicineand Biology Society vol 20 pp 974ndash977 1998

[25] C C Chang andC J Lin LIBSVMmdasha library for support vectormachines Version 312 httpwwwcsientuedutwsimcjlinlibsvm

[26] CW Hsu C C Chang and C J Lin ldquoA practical guide to sup-port vector classificationrdquo 2010 httpwwwcsientuedutwsimcjlin

[27] A Vedaldi and B Fulkerson ldquoVLFeat an open and portablelibrary of computer vision algorithmsrdquo 2008 httpwwwvlfeatorg

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

BioMed Research International 11

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(a) 120572 = 01

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(b) 120572 = 02

0

50

100

150

200

250

1 13 25 37 49 61 73 85 97 109 121 133 145 157

(c) 120572 = 03

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(d) 120572 = 04

1 13 25 37 49 61 73 85 97 109 121 133145 1570

20406080

100120140160180200

(e) 120572 = 05

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(f) 120572 = 06

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160180

(g) 120572 = 07

1 13 25 37 49 61 73 85 97 109 121 133 1451570

20406080

100120140160

(h) 120572 = 08

020406080

100120140

1 13 25 37 49 61 73 85 97 109 121 133 145157

(i) 120572 = 09

Figure 9 The result of joint rank and gray level features of normal and abnormal sperms

12 BioMed Research International

1 7 13 19 25 31 37 43 49 55 61 67 73 790

2

4

6

8

10

12

(a) Normal sperms

1 7 13 19 25 31 37 43 49 55 61 67 73 790

051

152

253

354

45

(b) Abnormal sperms

Figure 10 The classification of the top ten neighbours of (a) normal and (b) abnormal sperms

Rectangularsegment

Segmentation afterrotation

Ellipse

Figure 11 The ellipse model to estimate the contour of the sperm

In format (11) t represents the scaling parameter Byundertaking convolution in the whole domain with a two-dimensionalGaussian kernel and input image we can achievescaling correspondent to t

The SIFT feature vector has the following features (a) it isthe local feature of an image which maintains invariance notonly on rotation scale and brightness variation but also onthe viewing angle the affine transformation and the noise(b) it is distinctive and informative and suitable for fastaccurate matching in a mass signature database (c) it canproduce a large number of SIFT feature vectors with fewobjects (d) it is of high speed

Matlab source code of SIFT is fromhttpwwwvlfeatorgindexhtml [27] We used the SIFT method to extract featurepoints of all the sperm images and achieved the ten normaland ten abnormal sperm images as in the training modelThen we calculated the difference between the trainingmodels and the other 70 normal and 70 abnormal spermimages With the dividing line we came up with accuracyof 50 The result turned out to be bad because the inputdata are simple sperm images and the SIFT algorithm cannotextract enough feature descriptors for classification

53 Compared Ellipse Model Approach With the image ofa single sperm we can use the ellipse model to estimatethe contour of the sperm First of all we used a rectangularsegment to cut out the sperm In order to make the rectangleclose to the sperm edge we needed to choose the rectanglewith the smallest area we rotated the sperm so that thesegmentation would be easier We calculated the left-mostupper-most downward-most and right-most points on theedges and built the rectangular segment based on them asdepicted in Figure 11

76888890

7563

5000

6625

000100020003000400050006000700080009000

10000

Jointaverage

K-neighbor SIFT Ellipticmodel

FusedSVM

()

Figure 12 Accuracy comparison of all methods

With the rectangle we achieved an ellipse to estimatethe sperm whose long axis radius is half the length of therectanglersquos long side and whose short axis radius is half thelength of the rectanglersquos short side As a result we ascertainedthe ovality which is the ratio of the length of the long axisto the short axis Through our experiment on 80 sperms wecould tell whether the sperm was too close to a circle whoseratio is close to 1 and whether the sperm was too slim whoseratio is close to 2 or even larger than 2 We took those whoseratio was larger than 12 and smaller than 18 as normal andthus we achieved accuracy of 6625

6 Conclusion

As shown in Figure 12 we have presented an approachfor the classification of sperm images with one-dimensionalwaveform and gray level features In the comprehensiveevaluation the joint average approach method was applied

BioMed Research International 13

to 160 sperm image samples and provided 76875 accuracyof judgment We then applied the fused SVM method tothe images and reached accuracy of 875 The result provesthat the proposed approach is superior to the previousapproaches

References

[1] N R Pal and S K Pal ldquoA review on image segmentationtechniquesrdquo Pattern Recognition vol 26 no 9 pp 1277ndash12941993

[2] A Yezzi L Zollei and T Kapur ldquoA variational frameworkfor integrating segmentation and registration through activecontoursrdquoMedical ImageAnalysis vol 7 no 2 pp 171ndash185 2003

[3] R Chrastek M Wolf K Donath et al ldquoAutomated segmenta-tion of the optic nerve head for diagnosis of glaucomardquoMedicalImage Analysis vol 9 no 4 pp 297ndash314 2005

[4] H W Kang and S Y Shin ldquoEnhanced lane interactive imagesegmentation by incremental path map constructionrdquo Graphi-cal Models vol 64 no 5 pp 282ndash303 2002

[5] J Peters O Ecabert C Meyer R Kneser and J Weese ldquoOpti-mizing boundary detection via Simulated Search with appli-cations to multi-modal heart segmentationrdquo Medical ImageAnalysis vol 14 no 1 pp 70ndash84 2010

[6] G Behiels F Maes D Vandermeulen and P Suetens ldquoEvalu-ation of image features and search strategies for segmentationof bone structures in radiographs using Active Shape ModelsrdquoMedical Image Analysis vol 6 no 1 pp 47ndash62 2002

[7] M Blanchard K Haguenoer A Apert et al ldquoSperm morphol-ogy assessment using Davidrsquos classiffication time to switch tostrict criteria Prospective comparative analysis in a selectedIVF populationrdquo International Journal of Andrology vol 34 no2 pp 145ndash152 2010

[8] MKass AWitkin andD Terzopoulos ldquoSnakes active contourmodelsrdquo International Journal of Computer Vision vol 1 no 4pp 321ndash331 1988

[9] TMcInerney andD Terzopoulos ldquoT-snakes topology adaptivesnakesrdquoMedical Image Analysis vol 4 no 2 pp 73ndash91 2000

[10] J Liang T McInerney and D Terzopoulos ldquoUnited SnakesrdquoMedical Image Analysis vol 10 no 2 pp 215ndash233 2006

[11] M Zhao Z Meng K K Tseng J S Pan and C Y Hsu ldquoSym-metry auto-detection based on contour and corner modelsrdquo inProceedings of the 5th International Conference on Genetic andEvolutionary Computing (ICGEC rsquo11) pp 345ndash349 September2011

[12] Y Mingqiang K Kidiyo and R Joseph ldquoA survey of shapefeature extraction techniquesrdquo Pattern Recognition vol 15 no7 pp 43ndash90 2008

[13] Y Wang and H Huang ldquoAnalysis of Human Heartbeat withEKG Signalsrdquo 2010

[14] Q Hua A Ji and Q He ldquoMultiple real-valued K nearestneighbor classifiers system by feature groupingrdquo in Proceedingsof the IEEE International Conference on Systems Man andCybernetics (SMC rsquo10) pp 3922ndash3925 October 2010

[15] P Viswanath and T Hitendra Sarma ldquoAn improvement to k-nearest neighbor classifierrdquo in Proceedings of the IEEE RecentAdvances in Intelligent Computational Systems (RAICS rsquo11) pp227ndash231 September 2011

[16] P Wu and Q Chen ldquoA novel SVM-based edge detectionmethodrdquo Physics Procedia vol 24 pp 2075ndash2082 2012

[17] D N Sotiropoulos and G A Tsihrintzis ldquoArtificial immunesystem-based classification in class-imbalanced image prob-lemsrdquo in Proceedings of the 8th International Conference onIntelligent InformationHiding andMultimedia Signal Processingpp 138ndash141 July 2012

[18] M Ramon FMartınez-Pastor O Garcıa-Alvarez et al ldquoTakingadvantage of the use of supervised learning methods forcharacterization of sperm population structure related withfreezability in the Iberian red deerrdquoTheriogenology vol 77 no8 pp 1661ndash1672 2012

[19] S G Goodson Z Zhang J K Tsuruta W Wang and DA OrsquoBrien ldquoClassification of mouse sperm motility patternsusing an automatedmulticlass support vectormachinesmodelrdquoBiology of Reproduction vol 84 no 6 pp 1207ndash1215 2011

[20] D T Lin and D C Pan ldquoIntegrating a mixed-feature modeland multiclass support vector machine for facial expressionrecognitionrdquo Integrated Computer-Aided Engineering vol 16no 1 pp 61ndash74 2009

[21] C G Cheng Y M Tian and W Y Jin ldquoA study on the earlydetection of colon cancer using the methods of wavelet featureextraction and SVM classifications of FTIRrdquo Spectroscopy vol22 no 5 pp 397ndash404 2008

[22] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004

[23] V R Nafisi M H Moradi and M H Nasr-Esfahani ldquoSpermidentification using elliptic model and tail detectionrdquo WorldAcademy of Science Engineering and Technology vol 6 pp 205ndash208 2005

[24] W J Yi K S Park and J S Paick ldquoParameterized charac-terization of elliptic sperm heads using Fourier representationand wavelet transformrdquo in Proceedings of the 20th AnnualInternational Conference of the IEEE Engineering in Medicineand Biology Society vol 20 pp 974ndash977 1998

[25] C C Chang andC J Lin LIBSVMmdasha library for support vectormachines Version 312 httpwwwcsientuedutwsimcjlinlibsvm

[26] CW Hsu C C Chang and C J Lin ldquoA practical guide to sup-port vector classificationrdquo 2010 httpwwwcsientuedutwsimcjlin

[27] A Vedaldi and B Fulkerson ldquoVLFeat an open and portablelibrary of computer vision algorithmsrdquo 2008 httpwwwvlfeatorg

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

12 BioMed Research International

1 7 13 19 25 31 37 43 49 55 61 67 73 790

2

4

6

8

10

12

(a) Normal sperms

1 7 13 19 25 31 37 43 49 55 61 67 73 790

051

152

253

354

45

(b) Abnormal sperms

Figure 10 The classification of the top ten neighbours of (a) normal and (b) abnormal sperms

Rectangularsegment

Segmentation afterrotation

Ellipse

Figure 11 The ellipse model to estimate the contour of the sperm

In format (11) t represents the scaling parameter Byundertaking convolution in the whole domain with a two-dimensionalGaussian kernel and input image we can achievescaling correspondent to t

The SIFT feature vector has the following features (a) it isthe local feature of an image which maintains invariance notonly on rotation scale and brightness variation but also onthe viewing angle the affine transformation and the noise(b) it is distinctive and informative and suitable for fastaccurate matching in a mass signature database (c) it canproduce a large number of SIFT feature vectors with fewobjects (d) it is of high speed

Matlab source code of SIFT is fromhttpwwwvlfeatorgindexhtml [27] We used the SIFT method to extract featurepoints of all the sperm images and achieved the ten normaland ten abnormal sperm images as in the training modelThen we calculated the difference between the trainingmodels and the other 70 normal and 70 abnormal spermimages With the dividing line we came up with accuracyof 50 The result turned out to be bad because the inputdata are simple sperm images and the SIFT algorithm cannotextract enough feature descriptors for classification

53 Compared Ellipse Model Approach With the image ofa single sperm we can use the ellipse model to estimatethe contour of the sperm First of all we used a rectangularsegment to cut out the sperm In order to make the rectangleclose to the sperm edge we needed to choose the rectanglewith the smallest area we rotated the sperm so that thesegmentation would be easier We calculated the left-mostupper-most downward-most and right-most points on theedges and built the rectangular segment based on them asdepicted in Figure 11

76888890

7563

5000

6625

000100020003000400050006000700080009000

10000

Jointaverage

K-neighbor SIFT Ellipticmodel

FusedSVM

()

Figure 12 Accuracy comparison of all methods

With the rectangle we achieved an ellipse to estimatethe sperm whose long axis radius is half the length of therectanglersquos long side and whose short axis radius is half thelength of the rectanglersquos short side As a result we ascertainedthe ovality which is the ratio of the length of the long axisto the short axis Through our experiment on 80 sperms wecould tell whether the sperm was too close to a circle whoseratio is close to 1 and whether the sperm was too slim whoseratio is close to 2 or even larger than 2 We took those whoseratio was larger than 12 and smaller than 18 as normal andthus we achieved accuracy of 6625

6 Conclusion

As shown in Figure 12 we have presented an approachfor the classification of sperm images with one-dimensionalwaveform and gray level features In the comprehensiveevaluation the joint average approach method was applied

BioMed Research International 13

to 160 sperm image samples and provided 76875 accuracyof judgment We then applied the fused SVM method tothe images and reached accuracy of 875 The result provesthat the proposed approach is superior to the previousapproaches

References

[1] N R Pal and S K Pal ldquoA review on image segmentationtechniquesrdquo Pattern Recognition vol 26 no 9 pp 1277ndash12941993

[2] A Yezzi L Zollei and T Kapur ldquoA variational frameworkfor integrating segmentation and registration through activecontoursrdquoMedical ImageAnalysis vol 7 no 2 pp 171ndash185 2003

[3] R Chrastek M Wolf K Donath et al ldquoAutomated segmenta-tion of the optic nerve head for diagnosis of glaucomardquoMedicalImage Analysis vol 9 no 4 pp 297ndash314 2005

[4] H W Kang and S Y Shin ldquoEnhanced lane interactive imagesegmentation by incremental path map constructionrdquo Graphi-cal Models vol 64 no 5 pp 282ndash303 2002

[5] J Peters O Ecabert C Meyer R Kneser and J Weese ldquoOpti-mizing boundary detection via Simulated Search with appli-cations to multi-modal heart segmentationrdquo Medical ImageAnalysis vol 14 no 1 pp 70ndash84 2010

[6] G Behiels F Maes D Vandermeulen and P Suetens ldquoEvalu-ation of image features and search strategies for segmentationof bone structures in radiographs using Active Shape ModelsrdquoMedical Image Analysis vol 6 no 1 pp 47ndash62 2002

[7] M Blanchard K Haguenoer A Apert et al ldquoSperm morphol-ogy assessment using Davidrsquos classiffication time to switch tostrict criteria Prospective comparative analysis in a selectedIVF populationrdquo International Journal of Andrology vol 34 no2 pp 145ndash152 2010

[8] MKass AWitkin andD Terzopoulos ldquoSnakes active contourmodelsrdquo International Journal of Computer Vision vol 1 no 4pp 321ndash331 1988

[9] TMcInerney andD Terzopoulos ldquoT-snakes topology adaptivesnakesrdquoMedical Image Analysis vol 4 no 2 pp 73ndash91 2000

[10] J Liang T McInerney and D Terzopoulos ldquoUnited SnakesrdquoMedical Image Analysis vol 10 no 2 pp 215ndash233 2006

[11] M Zhao Z Meng K K Tseng J S Pan and C Y Hsu ldquoSym-metry auto-detection based on contour and corner modelsrdquo inProceedings of the 5th International Conference on Genetic andEvolutionary Computing (ICGEC rsquo11) pp 345ndash349 September2011

[12] Y Mingqiang K Kidiyo and R Joseph ldquoA survey of shapefeature extraction techniquesrdquo Pattern Recognition vol 15 no7 pp 43ndash90 2008

[13] Y Wang and H Huang ldquoAnalysis of Human Heartbeat withEKG Signalsrdquo 2010

[14] Q Hua A Ji and Q He ldquoMultiple real-valued K nearestneighbor classifiers system by feature groupingrdquo in Proceedingsof the IEEE International Conference on Systems Man andCybernetics (SMC rsquo10) pp 3922ndash3925 October 2010

[15] P Viswanath and T Hitendra Sarma ldquoAn improvement to k-nearest neighbor classifierrdquo in Proceedings of the IEEE RecentAdvances in Intelligent Computational Systems (RAICS rsquo11) pp227ndash231 September 2011

[16] P Wu and Q Chen ldquoA novel SVM-based edge detectionmethodrdquo Physics Procedia vol 24 pp 2075ndash2082 2012

[17] D N Sotiropoulos and G A Tsihrintzis ldquoArtificial immunesystem-based classification in class-imbalanced image prob-lemsrdquo in Proceedings of the 8th International Conference onIntelligent InformationHiding andMultimedia Signal Processingpp 138ndash141 July 2012

[18] M Ramon FMartınez-Pastor O Garcıa-Alvarez et al ldquoTakingadvantage of the use of supervised learning methods forcharacterization of sperm population structure related withfreezability in the Iberian red deerrdquoTheriogenology vol 77 no8 pp 1661ndash1672 2012

[19] S G Goodson Z Zhang J K Tsuruta W Wang and DA OrsquoBrien ldquoClassification of mouse sperm motility patternsusing an automatedmulticlass support vectormachinesmodelrdquoBiology of Reproduction vol 84 no 6 pp 1207ndash1215 2011

[20] D T Lin and D C Pan ldquoIntegrating a mixed-feature modeland multiclass support vector machine for facial expressionrecognitionrdquo Integrated Computer-Aided Engineering vol 16no 1 pp 61ndash74 2009

[21] C G Cheng Y M Tian and W Y Jin ldquoA study on the earlydetection of colon cancer using the methods of wavelet featureextraction and SVM classifications of FTIRrdquo Spectroscopy vol22 no 5 pp 397ndash404 2008

[22] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004

[23] V R Nafisi M H Moradi and M H Nasr-Esfahani ldquoSpermidentification using elliptic model and tail detectionrdquo WorldAcademy of Science Engineering and Technology vol 6 pp 205ndash208 2005

[24] W J Yi K S Park and J S Paick ldquoParameterized charac-terization of elliptic sperm heads using Fourier representationand wavelet transformrdquo in Proceedings of the 20th AnnualInternational Conference of the IEEE Engineering in Medicineand Biology Society vol 20 pp 974ndash977 1998

[25] C C Chang andC J Lin LIBSVMmdasha library for support vectormachines Version 312 httpwwwcsientuedutwsimcjlinlibsvm

[26] CW Hsu C C Chang and C J Lin ldquoA practical guide to sup-port vector classificationrdquo 2010 httpwwwcsientuedutwsimcjlin

[27] A Vedaldi and B Fulkerson ldquoVLFeat an open and portablelibrary of computer vision algorithmsrdquo 2008 httpwwwvlfeatorg

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

BioMed Research International 13

to 160 sperm image samples and provided 76875 accuracyof judgment We then applied the fused SVM method tothe images and reached accuracy of 875 The result provesthat the proposed approach is superior to the previousapproaches

References

[1] N R Pal and S K Pal ldquoA review on image segmentationtechniquesrdquo Pattern Recognition vol 26 no 9 pp 1277ndash12941993

[2] A Yezzi L Zollei and T Kapur ldquoA variational frameworkfor integrating segmentation and registration through activecontoursrdquoMedical ImageAnalysis vol 7 no 2 pp 171ndash185 2003

[3] R Chrastek M Wolf K Donath et al ldquoAutomated segmenta-tion of the optic nerve head for diagnosis of glaucomardquoMedicalImage Analysis vol 9 no 4 pp 297ndash314 2005

[4] H W Kang and S Y Shin ldquoEnhanced lane interactive imagesegmentation by incremental path map constructionrdquo Graphi-cal Models vol 64 no 5 pp 282ndash303 2002

[5] J Peters O Ecabert C Meyer R Kneser and J Weese ldquoOpti-mizing boundary detection via Simulated Search with appli-cations to multi-modal heart segmentationrdquo Medical ImageAnalysis vol 14 no 1 pp 70ndash84 2010

[6] G Behiels F Maes D Vandermeulen and P Suetens ldquoEvalu-ation of image features and search strategies for segmentationof bone structures in radiographs using Active Shape ModelsrdquoMedical Image Analysis vol 6 no 1 pp 47ndash62 2002

[7] M Blanchard K Haguenoer A Apert et al ldquoSperm morphol-ogy assessment using Davidrsquos classiffication time to switch tostrict criteria Prospective comparative analysis in a selectedIVF populationrdquo International Journal of Andrology vol 34 no2 pp 145ndash152 2010

[8] MKass AWitkin andD Terzopoulos ldquoSnakes active contourmodelsrdquo International Journal of Computer Vision vol 1 no 4pp 321ndash331 1988

[9] TMcInerney andD Terzopoulos ldquoT-snakes topology adaptivesnakesrdquoMedical Image Analysis vol 4 no 2 pp 73ndash91 2000

[10] J Liang T McInerney and D Terzopoulos ldquoUnited SnakesrdquoMedical Image Analysis vol 10 no 2 pp 215ndash233 2006

[11] M Zhao Z Meng K K Tseng J S Pan and C Y Hsu ldquoSym-metry auto-detection based on contour and corner modelsrdquo inProceedings of the 5th International Conference on Genetic andEvolutionary Computing (ICGEC rsquo11) pp 345ndash349 September2011

[12] Y Mingqiang K Kidiyo and R Joseph ldquoA survey of shapefeature extraction techniquesrdquo Pattern Recognition vol 15 no7 pp 43ndash90 2008

[13] Y Wang and H Huang ldquoAnalysis of Human Heartbeat withEKG Signalsrdquo 2010

[14] Q Hua A Ji and Q He ldquoMultiple real-valued K nearestneighbor classifiers system by feature groupingrdquo in Proceedingsof the IEEE International Conference on Systems Man andCybernetics (SMC rsquo10) pp 3922ndash3925 October 2010

[15] P Viswanath and T Hitendra Sarma ldquoAn improvement to k-nearest neighbor classifierrdquo in Proceedings of the IEEE RecentAdvances in Intelligent Computational Systems (RAICS rsquo11) pp227ndash231 September 2011

[16] P Wu and Q Chen ldquoA novel SVM-based edge detectionmethodrdquo Physics Procedia vol 24 pp 2075ndash2082 2012

[17] D N Sotiropoulos and G A Tsihrintzis ldquoArtificial immunesystem-based classification in class-imbalanced image prob-lemsrdquo in Proceedings of the 8th International Conference onIntelligent InformationHiding andMultimedia Signal Processingpp 138ndash141 July 2012

[18] M Ramon FMartınez-Pastor O Garcıa-Alvarez et al ldquoTakingadvantage of the use of supervised learning methods forcharacterization of sperm population structure related withfreezability in the Iberian red deerrdquoTheriogenology vol 77 no8 pp 1661ndash1672 2012

[19] S G Goodson Z Zhang J K Tsuruta W Wang and DA OrsquoBrien ldquoClassification of mouse sperm motility patternsusing an automatedmulticlass support vectormachinesmodelrdquoBiology of Reproduction vol 84 no 6 pp 1207ndash1215 2011

[20] D T Lin and D C Pan ldquoIntegrating a mixed-feature modeland multiclass support vector machine for facial expressionrecognitionrdquo Integrated Computer-Aided Engineering vol 16no 1 pp 61ndash74 2009

[21] C G Cheng Y M Tian and W Y Jin ldquoA study on the earlydetection of colon cancer using the methods of wavelet featureextraction and SVM classifications of FTIRrdquo Spectroscopy vol22 no 5 pp 397ndash404 2008

[22] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004

[23] V R Nafisi M H Moradi and M H Nasr-Esfahani ldquoSpermidentification using elliptic model and tail detectionrdquo WorldAcademy of Science Engineering and Technology vol 6 pp 205ndash208 2005

[24] W J Yi K S Park and J S Paick ldquoParameterized charac-terization of elliptic sperm heads using Fourier representationand wavelet transformrdquo in Proceedings of the 20th AnnualInternational Conference of the IEEE Engineering in Medicineand Biology Society vol 20 pp 974ndash977 1998

[25] C C Chang andC J Lin LIBSVMmdasha library for support vectormachines Version 312 httpwwwcsientuedutwsimcjlinlibsvm

[26] CW Hsu C C Chang and C J Lin ldquoA practical guide to sup-port vector classificationrdquo 2010 httpwwwcsientuedutwsimcjlin

[27] A Vedaldi and B Fulkerson ldquoVLFeat an open and portablelibrary of computer vision algorithmsrdquo 2008 httpwwwvlfeatorg

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology