Pictured, left to right, are: Manisha Bahl, director of the Massachusetts General Hospital Breast Imaging Fellowship Program; MIT Professor Regina Barzilay (center); and Constance Lehman, professor at Harvard Medical School and chief of the Breast Imaging Division at MGH’s Department of Radiology.
Image: Jason Dorfman/CSAIL
There are over 200,000 cases of breast cancer every year in the United States and 40,000 women die every year due to this. One of the best and most common ways to diagnose breast cancer is through mammograms. However, a drawback of using a mammogram is that they are still imperfect and result in a great many false positives which lead to unnecessary surgeries and biopsies. A cause of these false positives are high risk lesions that appear suspicious on mammograms and are often removed through surgeries. Nonetheless, 90% of these lesions are benign, meaning that thousands of women must go through painful, scarring and unnecessary surgeries.
Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory, Massachusetts General Hospital and Harvard Medical School turned to artificial intelligence for the answer. The model is trained on more than 600 existing high risk lesions and looks for patterns within family history, demographics, genetics, etc. By using the “random-forest classifier” the model diagnosed 97% of cancers. As the name suggests, a random-forest is made up of multiple decision trees. A decision tree is a predictive model which goes from observations about an item, or the branches, to conclusions about this item, represented by the leaves. Each of the decision trees come to a conclusion and vote on what the data set could be. Majority rules.
Researchers hope that this model can be incorporated into clinical practice in the next year. The team, including Regina Barzilay (MIT’s Delta Electronics Professor of Electrical Engineering and Computer Science), Constance Lehman (professor at Harvard Medical School and chief of the Breast Imaging Division at MGH’s Department of Radiology) and Manisha Bahl of MGH. Along with CSAIL graduate students Nicholas Locascio, Adam Yedidia, and Lili Yu, they published an article in the medical journal Radiology.