Home - Diseases - Using Computer Models to Predict Which Breast Lesions Require Removal
Research News

Using Computer Models to Predict Which Breast Lesions Require Removal

Machine learning model identifies the high-risk breast lesions that are most likely to turn into cancer.

Mary Bates, Contributor
Fri, 11/03/2017


A new study demonstrates that a machine learning model can distinguish between high-risk breast lesions that require surgical removal versus those that are not cancerous nor likely to turn into cancer. Use of this model could decrease unnecessary surgery by nearly one-third and could help guide clinical decision making with regard to surgery versus surveillance.

Suspicious findings from mammography often lead to image-guided biopsies, which can yield one of three results: cancer, high-risk or benign. Most high-risk breast lesions are not cancerous, but surgical excision is typically recommended because some high-risk lesions can be reassessed as cancer at surgery.

Currently, there are no features that reliably allow doctors to distinguish between high-risk lesions that warrant surgery from those that can be safely monitored. The result is unnecessary surgeries of high-risk lesions that are not associated with cancer.

Researchers from the Massachusetts General Hospital/Harvard Medical School and the Massachusetts Institute of Technology applied machine learning algorithms to this challenging clinical scenario.

“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,” said Manisha Bahl, the lead author of a study published in the journal Radiology.

Some of the features input into the model were patient age, pathology result at needle biopsy and risk factors like family history, breast density and findings on mammography. Additionally, the researchers input the full text of the core biopsy pathologic report. The model incorporated all of this data and then determined if a high-risk breast lesion diagnosed with biopsy actually required surgical removal, or if it could have been safely monitored.

Bahl and her colleagues found that compared to the traditional strategy of surgically removing all high-risk lesions, use of their model would increase the number of cancers detected and decrease the number of unnecessary surgeries performed.

“We found that if those high-risk lesions categorized with the model to be at low risk for upgrade were surveilled and the remainder were excised, then 97.4 percent of malignancies would have been diagnosed at surgery, and 30.6 percent of surgeries of benign lesions could have been avoided,” said Bahl.

“This is a well-executed study that uses machine learning to identify patients who could avoid unnecessary surgery for a benign breast lesion,” said Marc Kohli, director of clinical informatics and associate professor of clinical radiology at the University of California, San Francisco, who was not involved in this study.

Bahl said that the team is actively working to incorporate this risk prediction tool into their daily clinical practice and hope that it can be used to guide clinical decision-making soon. Their next steps include adding the actual mammographic images and histopathology slides into the machine learning model.

“Medicine strives to provide the right amount of care for each and every individual, relying on study data from large populations. This leads to some patients getting over-treated, and others under-treated,” said Kohli. “Sophisticated models, like the one described in this paper, will allow us to more closely match individual patients to the right treatment by analyzing large amounts of data that are difficult for humans to process.”