1. Image analysis with EBImage

2. Introduction EBImage Set of quantitative image processing tools Goal
Image reading/writing/rendering
Image processing (linear operators, matrix algebra, morphology)
Object detection
Features extraction
Goal
Extract cellular descriptors
Classify cells
Automatic phenotyping
Process/transform images

3. Image representation Image object Array of intensity values
Lena: 512x512 matrix 21 20 28 43 53 67 54 12 31 30 41 52 71 98 78 11 14 33 49 72
110
133
144 19 29 39 57 74
121
100 16 59 18 23 27 38 50 61 62 17 24 42 48 47 15 37 36

4. Image representation Multi-dimensionnal array 2D: spatial dimensions
Other dimensions: replicate, time point, color, condition, z-slice
Nuclei
4 replicates r0 r1 r2 r3
Lena
3 color
channels R G B

5. Image rendering Rendering dissociated from representation Sequence
of images
Color
images
Nuclei
4 replicates r0 r1 r2
Lena
3 color
channels R G B

6. IO Functions readImage(), writeImage()
Reads an image, returns an array
Supports 80 formats (JPEG, TIFF, PNG…)
Supports HTTP, sequences of images
Example: image format conversion
x = readImage('sample a.tif')
writeImage(x, 'sample a.jpeg', quality=95)

7. Display Function display() GTK+ interactive: zoom, scroll, animate
Supports RGB color channels and sequence of images

8. Adjust brightness, contrast and g-factor
Matrix algebra
Adjust brightness, contrast and g-factor x
x+0.5
3*x
(x+0.2)3

9. Spatial transformation
Cropping, thresholding, rotation, resize
x[45:56, 98:112]
x > 0.5
resize(x, w=128)
rotate(x, 30)

10. Thresholding x x > 0.3 Global thresholding
Building block tool to segment cells x
x > 0.3

11. Linear filter x x  2D convolution with filter2() Low-pass filter
Smooth images, remove artefacts x
1 1 1
x 

12. Linear filter x x  High-pass filter Detect cell edges -1 -1 -1
x 

13. Nuclei segmentation Global thesholding + labelling Function bwlabel()
Labels connected sets (objects) from a binary image
Every pixels of an object is set to an unique integer value
max(bwlabel(x)) gives the number of detected objects x
x>0.2
bwlabel(x>0.2)

14. Better segmentation
Adaptive thresholding + mathematical opening + holes filling
xb = thresh(x[,,1], 10, 10, 0.05)
kern = makeBrush(5, shape='disc')
xb = dilate(erode(xb, kern), kern)
xl = bwlabel(fillHull(nuct2)) x xb xl

15. Cell segmentation Using Voronoi segmentation with propagate()
Image gradient-based metric

16. Cell segmentation
Combining channels and segmentation masks

17. Feature extraction Function getFeatures()
Extracts features from image objects
Geometric, image moment based features
Texture based features (Zernike moments, Haralick features)
100 features
g.x g.y g.s g.p g.pdm g.pdsd g.effr g.acirc
[1,]
[2,]
[3,]
[4,]
[5,]
[6,]
[7,]
[8,]
[9,]
[10,]
[11,]
[12,]
[13,]
[14,]
[15,]
[16,]
[17,]
[18,]
[19,]
[20,]
...
76 cells

18. siRNA screen
3 immunostains: DNA, Actin, Tubulin
siRluc
siCLSPN

19. Segmentation

20. Features
Cell size
Median cell size of 1024 um2 (siRluc) and 1517 um2 (siCLSPN)
Wilcoxon rank sum test, P-value < 10-12