Imagine enduring a painful, expensive and scar-inducing surgery—only to find out afterwards that it wasn’t necessary.
This is the situation for many women with high-risk breast lesions—areas of tissue that appear suspicious on a mammogram and have abnormal but not cancerous cells when tested by needle biopsy. Following surgical removal, 90% of these lesions end up being benign.
A change in the standard of care could be on the horizon thanks to researchers at Massachusetts General Hospital and MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) who have found a more precise and less invasive way to separate harmful lesions from benign ones.
“The decision about whether or not to proceed to surgery is challenging, and the tendency is to aggressively treat these lesions [and remove them],” said Manisha Bahl, MD, Director of the Breast Imaging Fellowship Program at Mass General, in a recent interview.
Bahl, along with a team of researchers, have harnessed the power of artificial intelligence (AI) to develop a more accurate and less invasive screening method for high-risk lesions. When tested, the machine correctly diagnosed 97 percent of 335 high-risk breast lesions as malignant and reduced the number of benign surgeries by more than 30 percent compared to existing approaches. These results were recently published in Radiology.
The team developed an AI system that uses machine learning to distinguish between high-risk lesions that need to be surgically removed from those that should just be watched over time. They created this model by feeding it data on over 600 high-risk lesions, including information on the patient’s demographics and pathology reports, and then tasked it to identify patterns among the different data elements.
Through a process called deep learning, the machine uses the data to create an algorithm that can be used to predict which high-risk lesions should be surgically removed. This process differs from traditional software programming in that the researchers did not give the machine the formula for diagnosis, but rather let it analyze the data and identify patterns on its own.
“To our knowledge, this is the first study to apply machine learning to the task of distinguishing high-risk lesions that need surgery from those that don’t,” said collaborator Constance Lehman, MD, PhD, chief of the Breast Imaging Division at Mass General’s Department of Radiology, in a recent interview. “We believe this could support women to make more informed decisions about their treatment and that we could provide more targeted approaches to health care in general.”
Lehman says Mass General radiologists will begin incorporating the model into their clinical practice over the next year.