LongMIL

Version 1. WSI analysis with Alibi-2d positional embedding, supported by xformer.

step1: generate predefined 2-d alibi attention bias matrix by

python alibi.py

step2: use CLAM or other packages to preprocess your WSIs dataset.

step3: replace their WSI model during training by

LongMIL.py

Version 2. WSI analysis with our 2d local-global operations, implmented by our chucked attention. (NeurIPS-2024)

codes coming soon.