Better Breast Cancer Risk Prediction

A deep-learning model developed by KI member and Delta Electronics Professor Regina Barzilay can predict from a mammogram if a patient is likely to develop breast cancer within five years. Trained on mammograms and outcomes from more than 60,000 patients at Massachusetts General Hospital, the model learned to spot patterns in mammograms that are precursors to malignant tumors. Published in Radiology, the model performed significantly better than existing approaches, and could be used in the future to build personalized breast cancer screening plans. Read more.

At last month's SOLUTIONS with/in/sight, Barzilay was joined by her co-author, Harvard Medical School Professor and Director of Breast Imaging at Massachussetts General Hospital Constance Lehman, to talk about the new model and earlier work using deep-learning models to screen for dense breast tissue. Managing Director of The Boston Globe and STAT Linda Pizzuti Henry moderated the discussion, with an introduction from MIT president emerita and KI faculty member Susan Hockfield. Watch video.