Upload
kashif-aziz-awan
View
216
Download
0
Embed Size (px)
Citation preview
8/3/2019 Presentation @ FIT
1/20
Texture analysis for liver segmentation
and classification: a survey
Saima Rathore, Muhammad Aksam Iftikhar,
Mutawarra Hussain, Abdul JalilPakistan Institute of Engineering and Applied Sciences (PIEAS), Islamabad
8/3/2019 Presentation @ FIT
2/20
Abstract
Texture is a combination of repeated patterns with regular/irregular
frequency. It can only be visualized but hard to describe in words. Liver
structure exhibit similar behavior; it has maximum disparity in intensity
texture inside and along boundary which serves as a major problem in its
segmentation and classification. The problem of representing liver textureis solved by encoding it in terms of certain parameters (called features) for
texture analysis. Numerous texture analysis techniques have been devised
for liver classification over the years some of which work equally work
well for most of the imaging modalities. In this paper, we attempt to
summarize the efficacy of textural analysis techniques devised for CT,
Ultrasound and some other imaging modalities like MRI, in terms of well-known performance metrics.
8/3/2019 Presentation @ FIT
3/20
Introduction
Liver is the largest organ of body and performsvarious critical bodily functions
Normal liver usually differs from the diseasedone in terms of intensity texture. Thisvariation helps in determining thecorresponding disease.
A Computer-Aided-Diagnosis (CAD) system is amerger of medical imaging and tissuecharacterization techniques
8/3/2019 Presentation @ FIT
4/20
Liver CAD system
Figure 1 shows top level layout of a Computer
Aided Diagnosis system employing liver
texture analysis for disease diagnosis
Figure 1
8/3/2019 Presentation @ FIT
5/20
Textural analysis techniques
Various textural analysis techniques have beenused in literature which give rise to different setof features for liver classification. A few populartechniques are named as follows. Gray Level Difference Statistics (GLDS)
Spatial Gray level Dependence Matrices (SGLDM)
Gray level Run length Statistics (RUNL)
Laws Texture Energy Measure (TEM)
Wavelet Features
Fourier Power Spectrum (FPS)
First-Order Parameters (FOP)
8/3/2019 Presentation @ FIT
6/20
Texture Features
Aforementioned techniques give rise to various
features for texture classification. E.g.
Entropy (ENT)
Run Length Distribution (RLD)
Contrast (CO)
Variance (VAR)
Energy (E)
Uniformity (U)
Short Run Emphasis (SRE)
Gray Level Distribution (GLD)
Angular Second Moment (ASM)
Correlation (CORR)
Standard Deviation (SD)
Homogeneity (H)
Mean (M)
LAWs textural energy features
8/3/2019 Presentation @ FIT
7/20
US Texture Analysis Techniques
Ultrasound is the least expansive and most available
medical imaging modality
It has been used most frequently by researchers for liver
texture analysis . Table 1 (on next slide) is a summary of the work with US
modality for classification of different liver diseases
8/3/2019 Presentation @ FIT
8/20
US Texture Analysis Techniques
Paper Features / Technique Liver Classes
[2] GLDS, FDTA Normal, Cirrhosis, Heptoma
[5] GLDS, FDTA, RUNL, SGLDM Normal, Fatty, Cirrhosis, Heptoma
[6] GLDS, SGLDM, FDTA, RUNL, FOP Normal, Fatty, Cirrhosis
[7] FDTA, SGLDM Normal, Fatty, Cirrhosis
[24] GLDS, FDTA Normal, Cirrhosis, Heptoma
[29] Wavelet Normal, Cirrhosis, Steatosis
[30] SGLDM, FDTA, RUNL, GLDS Normal, Fatty, Cirrhosis, Heptoma
[31] GLDS, SGLDM, RUNL, FOP Normal, Fatty, Cirrhosis
[34] FDTA, SGLDM Normal, Fatty, Cirrhosis
[37] Gabor Wavelet Normal, Diseased
[42] Wavelet Normal, Diseased
[43] GLDS, SGLDM, Histogram Normal, Fatty
Table 1
8/3/2019 Presentation @ FIT
9/20
CT Texture Analysis Techniques
Computed Tomography (CT) is reliable enough but expansivemedical imaging modality
Table 2 shows different research works employing different textureanalysis techniques for classification of different liver disease
Paper Features / Technique Liver Classes
[6] FOP,SGLDM, GLDM,TEM, and FDTA Normal, Fatty, Cirrhosis
[9] Wavelets Normal, Diseased
[10] Wavelets Normal, Diseased
[13] Zernike moments, Legendremoments
Normal, HepatocellularCarcinoma
[14] SGLDM Normal, Cirrhosis, Heptoma,
Hemangioma
[19] FOP Normal, Heptoma,
Hepatocellular Carcinoma
Table 2
8/3/2019 Presentation @ FIT
10/20
Other Modalities
Several other modalities have been tried byresearchers, but with less frequency, to classify liverinto different liver classes.
Table 3 presents a summary of such effortsPaper Modality Features / Technique Liver Classes
[9] MRI GLDS, Shape features Noraml, Cirrhosis
[47] Biopsy FDTA Normal, Hepatocellular
Carcinoma[49] Mammography SGLDM Normal, Diseased
Table 3
8/3/2019 Presentation @ FIT
11/20
Discussion
Authors have employed a variety of textureanalysis techniques for liver classification
Different authors have used differentperformance metrics to check classification
accuracy. Most have used classification accuracyas a measure of performance while a few haveexploited area under ROC curve.
A consistent measure of accuracy would have
been more helpful for comparative analysis A summary of classification results using different
textural measures can be of substantial value forsetting future directions.
8/3/2019 Presentation @ FIT
12/20
A very few authors have used
sensitivity/specificity as performance measure,
which is summarized in table 4 below
Results Summary / Comparative Analysis
Paper Modality Features / Technique Sensitivity Specificity
[9] CT Wavelets 96 94
[10] CT Wavelets 98 85
[36] US SGLDM, FDTA 94.9 81.3
[37] US Gabor Wavelet 85.5 78
Table 4
8/3/2019 Presentation @ FIT
13/20
Results Summary / Comparative Analysis
Following graphical summary (Figure 2) of results provides greatcomfort to gain a deeper insight into the performance (accuracy)
of different research efforts and compare them critically.
Highlighted results show better performance of two techniques(One Ultrasound and one Computed Tomography)
Legend: Normal, Fatty, Cirrhosis, Heptoma, TotalFigure 2
8/3/2019 Presentation @ FIT
14/20
Best Performance Technique for CT
Bharathi et al. [13] utilized the better featurerepresentation capability and least informationredundancy of Zernike moments and Legendremoments for classification of normal and HCC liver
using CT images. Total 200 ROIs (140 Normal, 60 HCC) were
experimented out of which 75 were used for trainingand remaining for testing
The classification result with Zernike and Legendrefeature vector for normal liver was 98.60% and 97.57%respectively
Classification accuracy for HCC was 90.00% and 81.5%.
8/3/2019 Presentation @ FIT
15/20
Best Performance Technique for US
Mojsilovik et al. [29] used 6-level quincunx
wavelet decomposition for identifying diffused
liver diseases in their work.
They estimated channel variances using wavelets
at the output of each filter of the filter bank
which were then used for liver classification.
This scheme was effective as well as simple as itclassified normal and cirrhosis liver images with
an accuracy of 94% and 90% respectively.
8/3/2019 Presentation @ FIT
16/20
Conclusion
It has been observed that techniques based on CT textureanalysis, though evaluated for a few liver diseases, havemuch better discriminating power than others.
Contrary, techniques based on ultrasound images have
been used for diagnosing a large number of diseases (thismight pertain to low cost of ultrasound) but are lessaccurate.
Moreover, texture measure methods perform better whenused in combination as compared to their standalone
application. As already indicated that statistical moments based
technique gives better performance in case of CT[13] whilein case of ultrasound wavelet feature extraction [29]outclasses others
8/3/2019 Presentation @ FIT
17/20
Future Work
Current survey of liver textural analysis can be
extended in two directions in future
testing all texture measure methods using same
data set and similar performance measures may
provide a more accurate analysis
Adding more texture measure methods, even
trying a new one, can potentially provide bettercomparative study and a useful addition to current
research database
8/3/2019 Presentation @ FIT
18/20
References[2]. M.M.Galloway, Texture classification using gray level run lengths, Computer Graphics and
Image Processing, vol. 4, pp. 172- 179, 1975.
[5]. A.H. Mir, M. Hanmandlu, and S.N. Tandon, Texture analysis of CT images, IEEE Engineering
in Medicine and Biology, 1995.
[6]. S.Gr. Mougiakakou, I. Valavanis, K.S. Nikita, A. Nikita, and D. Kelekis, Characterization of CT
liver lesions based on texture features and a multiple Neural Network classification scheme,
Proceedings of the 25 Annual lnternational Conference of the IEEE EMBS, 2003.
[7]. Goldberg, Generic algorithm in search optimization and machine learning, Addison-Wesley,
1989.
[9]. K. Mala, and V. Sadasivam, Automatic segmentation and classification of diffused liverdiseases using wavelet based texture analysis and Neural Network, IEEE Indicon Conference,
pp. 216-219, 2005.
[10]. K.Mala, and Dr.V.Sadasivam, Wavelet based texture analysis of Liver tumor from Computed
Tomography images for characterization using Linear Vector Quantization Neural Network,
IEEE, 2006.
[13]. V.S. Bharathi, M.A.L. Vijilious, and L.Ganesan, Orthogonal Moments based texture analysisof CT liver images, International Conference on Computational Intelligence and Multimedia
Applications, 2007.
[14]. S. Nawaz, and A.H. Dar, Hepatic lesions classification by ensemble of SVMs using statistical
features based on co-occurrence matrix, International Conference on Emerging
Technologies, IEEE-ICET, 2008.
[19]. K. Wu, C. Garnier, J. Louis, Coatrieux, and H. Shu, A preliminary study of moment-based
texture analysis for medical images 32nd Annual International Conference of the IEEE EMBS,2010
8/3/2019 Presentation @ FIT
19/20
References (Contd.)[24]. Pavlopoulos et al., Evaluation of texture analysis techniques for quantitative characterization of
ultrasonic liver images, 18th Annual International Conference of the IEEE Eng in Med and Bio
Society, I996.
[29]. A.Mojsilovik, S. Markovic, and M. Popovic, Characterization of visually similar diffuse diseases
from B-scan liver images with the nonseparable wavelet transform, IEEE, 1997.
[30] E. Kyriacou, S. Pavlopoulos, D. Koutsouris, P. Zoumpouli, and I. Theotoka, Computer assisted
characterization of liver tissue using image texture analysis techniques on B-scan images,
Proceedings - 19th International Conference -IEEE/EMBS, 1997.
[31] Kyriacou et al., Computer assisted characterization of diffused liver images, IEEE, 1998.
[34]. Pavlopoulos et al., Fuzzy Neural Network-based texture analysis of ultrasonic images IEEEEngineering in Medicine and Biology, 2000.
[37]. A. Ahmadian, A. Mostafa, M.D. Abolhassani, and Y. Salimpour, A texture classification method
for diffused liver diseases using Gabor wavelets, Proceedings of the IEEE Engineering in
Medicine and Biology, 2005.
[42]. Y. Huang, L. Wang, and C. Li, Texture analysis of ultrasonic liver image based on wavelet
transform and PNN, International Conference On Biomedical Engineering & Informatics, 2008.
[43]. Huang et al., Texture analysis of ultrasonic liver images base5d on spatial domain methods,
3rd International Congress on Image and Signal Processing, 2010.
[47]. S.M. Pan, and C.H. Lin, Fractal features classification for liver biopsy images using NN based
classifier, International Symposium on Computer, Communication, Control and Automation,
2010.
[49]. M.H.Mohamed, and M.M.AbdeISamea, An efficient clustering based texture feature extraction
for medical image, Proceedings of International Workshop on Data Mining and AI, 2008.
8/3/2019 Presentation @ FIT
20/20
Questions
if (questions == yes)
DoAnswer()
else
Print Thank You!