Source: Stanford Report — 2026-07-07
Summary
Stanford Medicine researchers, with collaborators from the Broad Institute, Harvard Medical School, and MD Anderson, unveiled CANVAS, an AI platform that infers spatial multicellular "neighborhoods" from standard H&E pathology slides — information that normally requires expensive spatial proteomics — and uses it to predict immunotherapy resistance more accurately than current clinical methods.
Key Takeaways
- Trained by mapping CODEX spatial-proteomics data (18M+ cells, 457 non-small-cell lung cancer patients) onto ordinary H&E slide images, then inferring the same cellular-neighborhood information from routine slides alone.
- Identified 10 distinct cellular neighborhoods; a neutrophil-rich neighborhood correlated with worse prognosis and immunotherapy resistance.
- Designed to work across nine cancer types using pathology infrastructure hospitals already have, with no new expensive equipment required.
- The underlying spatial dataset and method were published in Cell on 2026-06-16; this write-up is the institutional dissemination piece, with clinical-trial validation planned next.