How did you acquire patch-level labels when calculating patch loss?

Question

How did you acquire patch-level labels when calculating patch loss?

Closed this issue 3 years ago · 1 comments

Since only WSI-level labels were available.

Answer 1 · 2021-08-20T03:57:28.000Z

Hey @yuanpinz, we assume patch-level labels to be the same as WSI-level labels (weak assumption). Mainly, we make this assumption to stabilize small data size training. You can train without patch-level labels also and evaluate the performance.