AI model predicts which cancer patients will respond to immunotherapy by reading tissue slides

Researchers trained a single AI model on 1.2 million tissue samples across 18 organs to predict gene expression from standard histology slides. This means hospitals can now forecast immunotherapy response from cheap, routine pathology slides instead of running expensive molecular tests, potentially speeding treatment decisions for thousands of patients.

Until now, predicting whether a cancer patient will respond to immunotherapy required expensive molecular testing (spatial transcriptomics) that most hospitals don't do routinely. This model cuts that bottleneck by reading the same H&E slides every pathologist already looks at, then outputting the molecular prediction that matters clinically. The practical effect: hospitals can start using immunotherapy response prediction as a standard part of cancer diagnosis without building new infrastructure or waiting weeks for molecular labs. The catch is this hasn't been tested in real clinical workflows yet, only on retrospective data.