The abstract is read at the publisher; the summary is JClub's.
PRISM, an interpretable multiscale graph-learning model using pretreatment H&E whole-slide images, predicts early treatment-specific objective responses in NSCLC across 7,082 patients.
- Why it matters: Treatment response prediction in NSCLC is difficult, and clinicians need models that can estimate outcomes for specific systemic regimens from routine pathology.
- What they did: The framework combines multiresolution tissue features with a cell-level TME hypergraph, then is tested in targeted therapy (T), EGFR-targeted therapy (T_EGFR), immune checkpoint inhibitor therapy (ICI), chemoimmunotherapy (CI), and chemotherapy (C) using internal cross-validation, independent external multicenter validation, and prospective silent validation.
- The result: PRISM stays competitive versus 18 multiple-instance learning and computational pathology baselines, and its multiscale attribution and H&E-derived TME analyses provide spatial context for predicted early responses, supporting further prospective multicenter evaluation.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.