Reply to: Clarifying multimodal inputs and attribution in colorectal cancer survival prediction - Takeaways - MDSpire

Reply to: Clarifying multimodal inputs and attribution in colorectal cancer survival prediction

Share

1

The survival prediction results were obtained using CT-only inference in the HydraMamba framework.
2

Cross-Modal Representation Learning trains a shared backbone using CT and endoscopic images for improved anatomical representation.
3

The ViT pooling (CT+Endo) baseline refers to a training dataset composition, not additional test-time inputs.
4

The survival head in HydraMamba receives a CT-derived representation from late fusion of anatomy tokens and style vectors.
5

Endoscopic data aids in training by providing textural priors, enhancing the CT features used for survival prediction.