If you count the red dots,
what can you say about a person?
A slide of cancer tissue is cut into 50 Β΅m tiles, and for each tile a CNN answers βis there a lymphocyte here?β. The 5,202 TIL maps made that way are publicly available.
Over these 20 lessons you count the pixels of those maps yourself to build an infiltration rate, then pass through KaplanβMeier, the log-rank test, multiple testing and Simpson's paradox to settle what that number entitles you to say. Python runs right inside this page. There is nothing to install.
How you'll learn
An infiltration rate does not fall out of the sky. What the denominator should be, where to draw the threshold, how large a tile to look at β you make each of those decisions yourself.
Not a stand-in dataset. The original maps from TCIA's TIL-WSI-TCGA (Saltz 2018) and the survival data from UCSC Xena, exactly as published.
Two slides from the same patient come out differently, and mixing cancer types flips the sign. Before you trust a result you feel how you get fooled with your own hands.
Every lesson has a laboratory widget β the threshold slider, the tile size, where you cut, the permutation shuffle, all turned by you.