A multimodal deep learning model for CRC prognosis prediction based on histopathological images and clinical variables.

1. UNI2 is used for feature extracting from patches.
2. ABMIL is used for feature aggregation to generate WSI-level embedding.
3. Cross-attention is used for multimodal fusion.
4. NLL is used for survival prediction.

WSI preprocessing is based on TRIDENT (https://github.com/mahmoodlab/TRIDENT). Thanks the authors for making it open source!