New AI Model Predicts Cancer Survival Rates from Histology Slides

Researchers developed a Vision Transformer-based AI model capable of predicting cancer subtypes, TP53 mutation status, and survival outcomes from routine H&E-stained whole-slide images across 32 solid tumor types.

The model achieved an area under the receiver operating characteristic curve of 0.766 for TP53 mutation detection and an overall classification accuracy of 0.659 during independent validation on 1,729 slides.

Co-lead investigator Alex W. Hewitt noted the tool is designed to complement molecular testing by providing decision support in remote or underprivileged clinical settings where genomic testing is inaccessible.