1. Image processing for selected biological experiments J. Schier, B. Kovář ÚTIA AV ČR, v.v.i.

2. 2 Contents Counting of yeast colonies (Future projects): Microarray scans Images from the FISH analysis (Fluorescence In-Situ Hybridization)

3. 3 Counting of yeast colonies Where, why? Application area: microbiology Testing influence of substance in the growth medium on innoculated colonies (size, growth rate) Test setup Colonies innoculated on Petri dishes Grown in a growth box (constant temperature and humidity) Dishes sampled by digital camera

4. Yeast colonies: examples

5. 4.2.20105 Growth box and imaging workplace

6. Area and number of colonies Quantitative analysis of images Manual counting Time consuming Limited number of samples Limited precision Automated counting

7. Evaluation of the dish Manual counting: Time consuming Limited number of samples Limited precision More complicated images cannot be evaluated at all..

8. Extreme example – densely covered dish

9. Problems Darkroom – controlled environment, but… Random factors: Varying position of the dish Varying illumination, zoom setting Dispersion of colony size & morphology Colonies are often touching each other

10. Two phases of dish processing Preprocessing Image checking and thresholding Dish localization, ROI extraction Evaluation of characteristics Relative area Colony diameter estimation Segmentation – counting of colonies Output filtration

11. Preprocessing No fancy math, but necessary: Detect and reject faluty images!! Localization ROI extraction

12. Background thresholding Elimination of faulty images:

13. Localization – dish rim? First solution: correlation of mask with dish rim Not sufficiently robust – rim variations (shape, width, reflections,..)

14. Localization - projections Binary image Projections

15. Position check – dish out of image: Only rim – OK, use Least Squares to refine Inner part of dish – REJECT! Localization – cont’d

16. Preprocessing Dish localization Thresholding Binary image Localization, inner thresholding

17. Eliminate rim, find ROI

18. Counting methods Convolution method based on convolution with circular pattern Fast Radial Symmetry (Loy&Zelinsky) Orientation & magnitude image computed from gradient Both methods need estimate of colony radius Adaboost, Hough Transform etc. not used – noisy, learning,...

19. Radius estimation RoundIrregular MinRadius, MaxRadius radii=[.....] Colony counting

20. Convolution – original flow

21. Colony counting I Convolution method radii vector→circular convolution patterns Colony Centers

22. Colony counting II Fast radial transform (Loy&Zelinsky 2003) Image gradient Orientation and Magnitude Matrices Result – symmetry matrix

23. Output filtration Reject centers in the background - “out of colony” Use “non-maxima suppression” Dilate output image of counting, look for common points with original output

24. Tool

25. Results - overview 117 images evaluated, containing from 9 to 106 colonies Fast Radial Transform: 81 images – no error 105 images – all colonies detected (some detected multiple times) Convolution: 36 images – no error 45 images - all colonies detected (some detected multiple times)

26. Test data 24 images35 images 23 images

27. Test data 2 23.2.2010Prezentace ZS

28. Results 23.2.2010Prezentace ZS # ColoniesSamples Fast Radial TransformConvolution Missed [%] False [%] Missed [%] False [%] 0-20240,26046,51043,12503,9063 20-25340,26635,19313,46212,3968 25-303704,31733,41371,6064 >30221,89524,347811,48270,3344

29. Results cont’d 23.2.2010Prezentace ZS #ColoniesSamples 0-2031 20-2537 25-3030 >3019 Minimum number of colonies: 9 Maximum number of colonies: 106

30. Results – cont’d 23.2.2010Prezentace ZS Rel. coverage [%] Samples 0-224 2-325 3-538 >530 Minimum coverage: 0,44% Maximum coverage: 12,48%

31. Detection examples Convolution Fast Radial Symmetry

32. Typical detection errors 23.2.2010Prezentace ZS Fast Radial Symmetry Convolution

33. Difficult example from the beginning… Result: Coverage 42.73% Total 598 colonies Detected 462 Missed 136 (Fast Radial Symmetry)

34. Conclusions Semi-automated processing of batches of Petri dish images Two methods proposed Interactive graphical editor of the result Evaluation of efficiency Improved process over manual evaluation

35. Outcomes of the research Tool deployed and in practical use: Yeast Colony Group Department of Genetics and Microbiology Faculty of Sciences Charles University Journal paper in review process: Computer Methods and Programs in Biomedicine (Elsevier)

36. Outcomes of the research Established cooperation with two groups Yeast Colony Group (YCG) Department of Biology and Medical Genetics, Charles University in Prague - 2nd Faculty of Medicine (UBLG) 2 grant proposals: Image processing for microarrays (GAČR, UTIA+YCG) System for FISH analysis evaluation (TAČR, UTIA+FIT+UBLG+CAMEA s.r.o.)

37. DNA microarray processing 14000 spots with red and green fluorescence(ratio of mRNA content of a given gene for two samples) Image processing: determination of exact location and size of the spot elimination of the spots with strong background Currently: high ratio of manual processing

38. FISH analysis FISH = Fluorescence In-Situ Hybridization sample, containing DNA to be examined, is hybridized with a probe probe: DNA fragment marked with a fluorescence dye if the DNA sample contains complementary fragments, the probe will match this can be observed in a fluorescence microscope the presence of signal indicates presence of a chromosome containing the DNA sequence

39. FISH analysis Application Detection of Turner syndrom (1 out of 2500 newborn girls) Growth distortions, infertility,.. X monosomy in mosaic form with frequency <1% ! Detection of Klienefelter syndrome, etc. etc.

40. FISH analysis