Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Massachusetts General Hospital (MGH) have created a deep learning model that can predict if a woman will get breast cancer in the future based on a mammogram she has done today.
The researchers trained the AI by having it review mammograms and the known outcomes from more than 60,000 patients treated at MGH. The AI was then able to learn to identify the subtle patterns in breast tissue that are precursors to cancer.
All in all, the AI could accurately predict about 31% of all cancer patients in a high-risk category. Thirty-one percent might not sound like that much, but the AI performed significantly better than traditional ways of predicting breast cancer risks, which only identify about 18% of all cancer patients in a high-risk category.
What’s even better is MIT’s AI modeling works just as well for black women as it does for white women. Previous techniques were better at detecting signs of breast cancer in white women than they could in African-American populations because those modeling techniques were based on populations of white women. MIT’s new AI, on the other hand, was trained on data from women of both populations.