42
1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘ The ‘ microtubules microtubules project project

1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

Embed Size (px)

Citation preview

Page 1: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

1

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Cédric Dufour

( LTS-IBCM Collaboration )

The ‘The ‘microtubulesmicrotubules’ project’ project

Page 2: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

2

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

The ‘The ‘microtubulesmicrotubules’ project’ project Morphological filteringMorphological filtering Markers extractionMarkers extraction Microtubules extractionMicrotubules extraction ResultsResults Algorithm testingAlgorithm testing Final assessmentFinal assessment

Content

Page 3: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

3

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Obtain Obtain specific proteins’ density statisticsspecific proteins’ density statistics, related to , related to the neural cell microtubules structures.the neural cell microtubules structures.

Markers

The ‘microtubules’ project The goal

Microtubules

Page 4: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

4

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

The ‘microtubules’ project The various steps

Isolate the markers Isolate the markers maskmask..

Isolate the microtubules Isolate the microtubules mask mask andand skeleton skeleton..

Compute the microtubules’ surface and length.Compute the microtubules’ surface and length.

Compute the markers’ quantity, overall and near the Compute the markers’ quantity, overall and near the

microtubules.microtubules.

Page 5: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

5

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

The ‘microtubules’ project The way it is done

Markers extraction :Markers extraction :

Morphological filtering and local maximum Morphological filtering and local maximum

detection.detection.

Microtubules extraction :Microtubules extraction :

Selective filtering using linear oriented structuring Selective filtering using linear oriented structuring

element’s correlation with thresholded image.element’s correlation with thresholded image.

Page 6: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

6

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Morphological filtering What is it ?

Morphological filteringMorphological filtering is a filtering method is a filtering method

originated from the theory of originated from the theory of mathematical mathematical

morphologymorphology..

The base of all morphological processing are the The base of all morphological processing are the

‘‘erosionerosion’ and ‘’ and ‘dilationdilation’ morphological functions.’ morphological functions.

Page 7: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

7

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Morphological filtering Binary ‘erosion’

Bb

B bXXbyByyX

,,

Original set Eroded set

Structuring elmt.

Page 8: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

8

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Morphological filtering Binary ‘dilation’

Bb

B bXBbXxbxX

,,

Original set Dilated set

Structuring elmt.

Page 9: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

9

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Morphological filtering Binary ‘opening’

XX BBB

Original set Opened set

Structuring elmt.

Page 10: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

10

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Morphological filtering Binary ‘closing’

XX BBB

Original set Closed set

Structuring elmt.

Page 11: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

11

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Replace the Replace the intersectionintersection//unionunion operators with operators with infimuminfimum//supremumsupremum operators operators

yxfxf

yxfxf

ByB

ByB

sup

inf

Morphological filtering Function (or ‘gray scale’) morphology

Original fct.

Dilated fct.

Eroded fct.

Original fct.

Closed fct.

Opened fct.

Page 12: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

12

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Base imageBase image

Markers extraction Step by step : 1

Page 13: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

13

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Opening with a 3x3 squareOpening with a 3x3 square

Markers extraction Step by step : 2

Page 14: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

14

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Subtraction with original image (Tophat)Subtraction with original image (Tophat)

Markers extraction Step by step : 3

Page 15: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

15

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

NormalizationNormalization

Markers extraction Step by step : 4

Page 16: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

16

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

‘‘Loosy’ thresholdLoosy’ threshold

Markers extraction Step by step : 5

Page 17: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

17

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Remove small elements (artefacts)Remove small elements (artefacts)

Markers extraction Step by step : 6

Page 18: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

18

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Mask base image ( and normalize )Mask base image ( and normalize )

Markers extraction Step by step : 7

Page 19: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

19

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Local maximum detection ( in 5x5 disc neighborhood )Local maximum detection ( in 5x5 disc neighborhood )

Markers extraction Step by step : 8

Page 20: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

20

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Resulting Resulting markers maskmarkers mask

Markers extraction In the end

Page 21: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

21

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Base imageBase image

Microtubules extraction Step by step : 1

Page 22: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

22

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Opening with 21x21 square (Opening with 21x21 square ( background) background)

Microtubules extraction Step by step : 2

Page 23: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

23

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Subtraction with original image (Tophat)Subtraction with original image (Tophat)

Microtubules extraction Step by step : 3

Page 24: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

24

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Markers removalMarkers removal

Microtubules extraction Step by step : 4

Page 25: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

25

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

ThresholdThreshold

Microtubules extraction Step by step : 5

Page 26: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

26

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Resulting binary imageResulting binary image

Microtubules extraction Step by step : 6

Page 27: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

27

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Oriented linear element correlationOriented linear element correlation

Microtubules extraction Step by step : 7

Page 28: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

28

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Small elements (artefacts) removalSmall elements (artefacts) removal

Microtubules extraction Step by step : 8

Page 29: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

29

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Repeat last 2 steps for different orientationsRepeat last 2 steps for different orientations

Microtubules extraction Step by step : 9

Page 30: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

30

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Thresholding of filter accumulatorThresholding of filter accumulator

Microtubules extraction Step by step : 10

Page 31: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

31

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Closing with a 3x3 cross to remove irregularitiesClosing with a 3x3 cross to remove irregularities

Microtubules extraction Step by step : 11

Page 32: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

32

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Skeleton by thinning and cntd. points suppressingSkeleton by thinning and cntd. points suppressing

Microtubules extraction Step by step : 12

Page 33: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

33

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Resulting Resulting microtubules mask microtubules mask andand skeleton skeleton

Microtubules extraction In the end

Page 34: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

34

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Results Image 1 (demo image)

Page 35: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

35

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Results Image 2 (low density)

Page 36: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

36

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Results Image 3 (high density)

Page 37: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

37

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Errors Errors markers: 0% / 4.8% markers: 0% / 4.8%

Algorithm testing Real image (PSNR = 35dB)

Page 38: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

38

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Errors Errors microtubules: 0.7% / 0.5%; markers: 0% / 0% microtubules: 0.7% / 0.5%; markers: 0% / 0%

Algorithm testing Synthetic image (PSNR = 35dB)

Page 39: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

39

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

We’ve been able to offer the biologists…We’ve been able to offer the biologists…

– a fully automatic analysis program,

– running in a powerful and wide-spread environment

(MatLab),

– giving good results, according to the biologists’ needs.

Final assessment In general

Page 40: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

40

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Microtubules mask extraction :Microtubules mask extraction :

The poor quality of the input images (very low The poor quality of the input images (very low

contrast) leads to a low-efficiency microtubules contrast) leads to a low-efficiency microtubules

mask extraction procedure (mask extraction procedure (the algorithm misses the algorithm misses

the most evanescent structuresthe most evanescent structures).).

Final assessment Problems

Page 41: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

41

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Microtubules mask extraction :Microtubules mask extraction :

Use tracking algorithm to follow the full microtubule Use tracking algorithm to follow the full microtubule

(pseudo-linear) structure.(pseudo-linear) structure.

N.B.N.B. This is not easy because of the variable This is not easy because of the variable

number of microtubules that may cross in one point number of microtubules that may cross in one point

(resulting in tracking uncertainties)(resulting in tracking uncertainties)

Final assessment Improvements

Page 42: 1 Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne Cédric Dufour ( LTS-IBCM Collaboration ) The ‘microtubules’ project

42

Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne

Thank you for your attentionThank you for your attention