Computer vision in neuropharmacology
•Download as PPTX, PDF•
0 likes•90 views
ISS Art shares experience of using Computer Vision in neuropharmacology on the example of the portfolio project. ISS Art CTO describes the technical part of developing a Computer Vision script for neuropharmacology at HappyDevLite conference http://happydevlite.1der.link/site. You will know about the script we wrote for the company that creates drugs for diseases related to the death of nerve cells. Well-known examples of these diseases are Parkinson and Alzheimer. What details are disclosed: -how to identify an object that move a bit -how to identify the core of the object (in our case it was the core of a cell) -how highlight an object and to identify the affiliation etc.
Report
Share
Computer vision in neuropharmacology
- 1. The use of methods of computer vision in neuropharmacology Sergey Maslov, CTO, ISS Art
- 2. Student Student+programmer Teacher Tester(didn’t like it) Programmer Head of development team Architect CTO 2 About me Omsk State Polytechnical University
- 3. Customer A company from the USA which researches and creates drugs for illnesses connected with nerve cell death (e.g. Alzheimer, Parkinson diseases). 3 Neuropharmacology - the field of development of drugs for diseases of the nervous system.
- 4. Process 4 Grow cells in Petri dishes Drip drugs Watch what happens with cells
- 5. Watching cells 5 Digital microscope Immunofluorescence Automatic monitoring of cell conditions (alive-dead) process of immunofluorescence
- 6. Example of photo Original photo from digital microscope 6 Photo after pre- processing Shown for visibility
- 7. Nerve cell structure of nerve cell (from school book of biology) 7 Photo after pre- processing Main criteria of condition: Brightness, presence of neurites
- 8. Issues 8 Customer’s tools: Slow (2-3 hours for processing 1 photo) Inaccurate
- 10. Pre-processing 10 Photo after pre- processing Make it black-and-white Brightness/ Contrast
- 11. 11 Finding cell centers Found cells Classical computer vision technique: Blob detection For geeks: bright spot - is local maximum of Laplace transform from Gauss blurring. Must know: divergence of vector field
- 12. 12 Finding neutrites Ridge line, actually Color - the height Natural analogue: ridge line Photo after pre- processing Must know: classification of quadric surfaces, eigenvalues of matrix of second derivatives
- 13. Watershed algorithm 13 Neurites affiliation to cells algorithm working scheme
- 14. 14 Neurites affiliation to cells kind of
- 15. 15 Input - output Collection of daily photos Daily cell conditions Quantity and length of neurites Chart
- 16. 16 Results Faster (2-3 minutes instead of 2-3 hours per photo) Much more accurate (hard to say in numbers) Use of indicators that were not used before (quantity and length of neurites) Tools for analyzing cell photos: further cooperation with the customer
- 17. What to read 17 Nerve cells Immunofluorescence Opencv tutorial Mathematics, programming and English - study on your own :) For those who is too lazy to google
- 18. Thank you! Sergey Maslov, CTO, ISS Art LLC Email: crm@issart.com