Deep Learning Predicts Mutations and Outcomes in Gastrointestinal Stromal Tumors.

Deep learning accurately predicts mutations, treatment response, and recurrence risk in 8,398 gastrointestinal stromal tumor cases, with AUCs up to 0.96 for key mutations.

• Why it matters: GIST prognosis and therapy are limited by inconsistent risk models and the challenge of molecular classification, especially for wild-type variants that respond poorly to existing treatments.
• What they did: A deep learning approach was trained on whole-slide images from multiple international centers to classify mutations, predict treatment sensitivity, and estimate recurrence-free survival, using a large dataset of over 7,000 cases with molecular and clinical data.
• The result: The models achieved high accuracy, comparable to traditional pathology scores, enabling improved molecular and prognostic predictions that could guide personalized treatment strategies in GIST.