OmniT2I: a unified table-to-image framework for survival prediction and cross-modal inference with incomplete multimodal cancer data

OmniT2I improves survival prediction from incomplete multimodal cancer data, reaching a pooled C-index of 0.783 across 10,373 patients and 32 cancer types.

• Why it matters: Prognostic modeling is limited when molecular profiles and histopathology are missing for individual patients, and cross-modal inference is needed to use whatever data are available.
• What they did: The method converts each molecular profile into a three-channel image using correlation-based feature placement, fuses available embeddings via availability-aware fusion, and predicts missing modalities; it is evaluated across 10,373 patients and 32 cancer types.
• The result: A single model outperformed nine separately trained survival methods in 240/279 pooled and 226/279 within-cancer comparisons, and adding withheld-modality predictions increased C-indices in all 180 pooled pairings and 164 within-cancer pairings.

AI summary of the abstract · may contain errors · not medical advice · read the paper.