“We shall not cease from exploration and the end of all our exploring will be to arrive where we started and know the place for the first time”
- TS Eliot.
Is it possible to predict molecular biomarkers from H&E sections through deep learning and computer vision? After all the Hematoxylin and Eosin section was one of our first tools in our arsenal. Molecular diagnostics came center stage in the 1990s and 2000s. Can we know can we now go back to the H&E and predict the molecular features?
Our guest is Heather Couture of Pixel Scientia, where she accelerates machine learning projects for pathology startups. She has a PhD in Computer Science and has published in top-tier computer vision and medical imaging venues. She is the host of the Impact AI Podcast and writes regularly on LinkedIn, for her newsletter Pathology ML Insights, and for a variety of other publications.
We're going to be talking about artificial intelligence and computer vision in digital pathology. First of all, what do we even mean by artificial intelligence? What is the difference between AI, a computer program, or simply just crunching large amounts of data? How do we develop tools for digital pathology; what are some of the challenges and pitfalls? And yes the “V” word - what do we mean by validation? How do we go about validation in the digital era?
And finally, what about those pesky preanalytical factors? Will it even be possible to develop new tools without standardizing or accounting for variability in tissue processing, staining, and scanning?
