Deep learning, digital pathology, challenges - February-March, 2020

To start off this newsletter, I'm going to make a few posts with links to my earlier blog posts from my PhD research.

I started the research blog in February, 2020. Great timing. For some reason, I didn’t really keep up for long with my initial plans of one post per week…

I started things off with introductions of my thesis topic, starting with digital pathology, then deep learning.

Then, I looked at how the two have been combined with a short history of deep learning in digital pathology, from Cireşan et al.’s mitosis detection and Cruz-Roa et al’s skin cancer detection, to the rise of U-Net and its successors, and some thoughts about where all of that was going.

Finally, on March 9th, 2020, I took a more in-depth look at the MITOS12 challenge, which had a big influence in the direction that my thesis took. The MITOS12 challenge was one of the first “digital pathology competitions”, and it was where deep learning methods were first shown to be one step above the rest. It also had some serious issues in its design. Problems with digital pathology challenges, their evaluations, and how much we can trust their results ended up making a fairly large portion of my thesis.