Using AI to Improve Early Breast Cancer Detection

A common cause of false positives is “high-risk” lesions, which appear suspicious on mammograms and have abnormal cells when tested by needle biopsy. Generally, the patient undergoes surgery to have the lesion removed in such cases, although the lesions turn out to be benign at surgery 90 percent of the time. As a result, thousands of women have to unnecessarily undergo painful, expensive, scar-inducing surgeries.

As a first project to apply AI for improving detection and diagnosis, the teams have collaborated to develop an AI system, which is trained on information about more than 600 existing high-risk lesions and looks for patterns among many different data elements, including demographics, family history, past biopsies, and pathology reports. Using a method known as a “random-forest classifier,” the model when tested on 335 high-risk lesions correctly diagnosed 97 percent of the breast cancers as malignant and reduced the number of benign surgeries by more than 30 percent compared to existing approaches.

“This work highlights an example of using cutting-edge machine learning technology to avoid unnecessary surgery,” said Marc Kohli, director of clinical informatics in the Department of Radiology and Biomedical Imaging at the University of California at San Francisco. “This is the first step toward the medical community embracing machine learning as a way to identify patterns and trends that are otherwise invisible to humans.”

“Because diagnostic tools are so inexact, there is an understandable tendency for doctors to over-screen for breast cancer,” said Regina Barzilay, MIT’s Delta Electronics Professor of Electrical Engineering and Computer Science and a breast cancer survivor herself. “When there’s this much uncertainty in data, machine learning is exactly the tool that we need to improve detection and prevent over-treatment.”

In the near future, the model could also be easily tweaked for application in other types of cancer as well as for other completely different diseases. “A model like this will work anytime you have lots of different factors that correlate with a specific outcome,” said Barzilay. “It hopefully will enable us to start to go beyond a one-size-fits-all approach to medical diagnosis.”

Related Links:
Massachusetts Institute of Technology