MIT has teamed up with Massachusetts General Hospital (MGH) to create a new deep learning model that can predict from a patients mammogram if they are likely to develop breast cancer in the future. The model was trained using mammograms with known outcomes from over 60,000 patients that were treated at MGH. Using that dataset, the model learned subtle patterns in breast tissue that are precursors to malignancy.
The goal of the MIT system is to allow doctors to customize screening and prevention programs at the individual level and make late diagnosis a relic of the past. Currently, the recommendation from the American Cancer Society is annual screenings starting at 45. The US Preventative Task Force recommends bi-annual screening starting at 50.
The team at MIT wants to eliminate the one-size fits all approach and personalize screenings around the risk of developing cancer. The team found that its model was “significantly” better at predicting cancer risk than the existing approaches. It was able to accurately place 31% of all cancer patients in its highest-risk category compared to only 18% for traditional models.
The researchers trained their deep learning model to induce the patterns directly from the data on over 90,000 mammograms. The model is able to pick up on patterns that are too subtle and complex for the human eye to detect. The new model is also more accurate for minorities as most of the early detection models were developed on populations of white women.
Researchers say that is particularly important for African American women who are 43% more likely to die than white women. The new system could one day enable doctors to use mammograms to see if patients are at a greater risk for other health problems like cardiovascular disease or other cancers.