The world is being quietly rearranged by people who write very long documents.


The title they went with MUST: Modality-Specific Representation-Aware Transformer for Diffusion-Enhanced Survival Prediction with Missing Modality Noisy translates that to

AI learns to predict cancer survival with incomplete medical records


Researchers built a machine learning system that can predict how long cancer patients will survive even when some of their medical data is missing — a common real-world problem because collecting every scan, pathology slide, and genetic test is expensive and time-consuming. The system works by figuring out which information is truly irreplaceable and which can be inferred from other available data, then filling in the gaps using a generative model trained on what's recoverable.
Clinical AI systems today often fail when real-world data is incomplete, which makes them fragile in actual hospitals. If this approach works reliably at scale, it removes a major bottleneck: hospitals wouldn't need to delay diagnosis or skip tests just to feed complete datasets into AI systems.

If you insist
Read the original →