Researchers at Massachusetts General Hospital, Harvard Medical School and MIT’s Computer Science and Artificial Intelligence Laboratory have developed a new system that utilizes artificial intelligence to distinguish between high-risk lesions requiring surgical removal and those that simply need to be watched over time.

If effective, the system could be used to reduce over-screening of breast cancer, as well as false positives and the number of benign surgeries that take place, thereby sparing many patients painful and scar-inducing procedures and the expenses that come with them.

“Currently, there are no features – imaging or otherwise – that reliably allow us to distinguish between high-risk lesions that warrant surgery from those that can be safely followed over time,” Manisha Bahl, director of the breast imaging fellowship program at Massachusetts General Hospital, told HCB News. “Machine learning allows us to incorporate the full spectrum of diverse and complex data that we have available, such as patient risk factors and imaging features, in order to predict which high-risk lesions are likely to be upgraded to cancer and, ultimately, to help our patients make more informed decisions about surgery versus surveillance.”

High-risk lesions are areas of tissue that appear suspicious on mammograms and are found to have abnormal cells by needle biopsy tests. Surgical procedures are typically recommended, as some can be upgraded to cancer at surgery. However, following such procedures, 90 percent of these lesions are found to be benign, making surgery unnecessary.

The system is trained with information on nearly 700 existing high-risk lesions and analyzes a range of data elements, such as demographics, family history, past biopsies and pathology reports, to identify patterns. The information is then used to create an algorithm to predict which high-risk lesions require surgery and which should be monitored.

In a test consisting of 335 high-risk lesions, the system, using a method known as random-forest classifier, showed a 97 percent accuracy rate for identifying malignancies, and reduced the number of benign surgeries by more than 30 percent, compared to current available approaches.

Bahl says there is expressed interest in possibly using the system for risk-stratification of patients with ductal carcinoma in situ (DCIS), a noninvasive breast cancer, and that in regard to high-risk lesions, shows much potential in making sure patients receive the best course of action for treatment.

“Our study provides proof of concept that a machine learning model can be applied to predict the risk of upgrade of high-risk breast lesions to cancer,” Bahl said. “Use of our model has the potential to decrease unnecessary surgery by nearly one-third in women with high-risk breast lesions and could support shared decision-making with regard to surgery versus surveillance of high-risk lesions. We are actively working to incorporate this risk prediction tool into our daily clinical practice and hope to use it to guide clinical decision making very soon.”

The next step will focus on incorporating actual mammographic images and histopathology slides. Mass General radiologists are expected to begin incorporating the system into their clinical practice over the next year.

Cardiology Homepage