PathPrism, an interpretable AI framework, was evaluated for spatial biomarker discovery in histopathology whole-slide images across 7,000 colorectal cancer patients in 11 cohorts, targeting prognosis, molecular alterations (MSI, BRAF, TP53), and chemotherapy benefit prediction in stage II/III disease.
PathPrism identified hundreds of spatial biomarkers predictive of survival, MSI status, BRAF and TP53 mutations, and stratified chemotherapy benefit in stage II/III colorectal cancer; it also introduced VirtualWSI for semantic in-silico tissue perturbation experiments.
Study is limited to colorectal cancer, so generalizability to other tumor types is unproven. The framework relies on LLMs as auxiliary hypothesis generators, which may introduce model-specific biases. Validation is retrospective across cohorts without a prospective clinical trial.
PathPrism offers a transparent, scalable approach to extract prognostic and predictive spatial biomarkers directly from routine pathology slides — no additional molecular assays needed. Clinicians may eventually use such tools to guide chemotherapy decisions in stage II/III colorectal cancer, pending prospective validation.
Explore related topics
