Breast cancer is a condition that claims the lives of 40,000 women every year in the U.S. alone. The problem isn’t necessarily cancer itself, it’s the detecting of it at an early enough stage we struggle with. While mammograms are useful, not they’re not perfect and often give off false positives, leading to unneeded biopsies and unnecessary surgery. High-risk lesions are one common cause of false positives. When tested with a biopsy needle they show abnormal cells and they appear as suspicious on mammograms too. So then, what can we do to improve these methods and ultimately save lives?
According to researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), Harvard Medical School, and Massachusetts General Hospital, the answer lies in artificial intelligence? They collaborated in order to develop a machine learning AI system that was able to predict if high-lesions detected by a needle biopsy after a mammogram would result in cancer. Results from the study revealed that out of 335 high-risk lesions tested, 97% were correctly diagnosed as malignant by the machine learning model, and as a result reduced the number of surgeries required by more than 30% compared to traditional methods.
“Because the diagnostic tools are so inexact, there is an understandable tendency for doctors to over-screen for breast cancer,” says MIT’s Delta Electronics Professor of Electrical Engineering and Computer Science, Regina Barzilay. “When there’s so much uncertainty in data, machine learning is exactly the tool that we need to improve detection and prevent over-treatment.”
The AI works by looking for patterns among various elements of data including past biopsies, demographics, and family history. “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,” says Constance Lehman, a professor at Harvard Medical School and chief of the Breast Imaging Division at MGH’s Department of Radiology. “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.”
How it works at the moment is first a mammogram detects the lesion, then a needle biopsy is carried out on the lesion to determine whether or not it’s cancerous. Current statistics show that around 70 percent are benign while 20 percent are malignant, and 10 percent are actually high-risk lesions. Not all high-risk lesions are treated through surgery. Some doctors will only carry out surgery on lesions that have high cancer rates, such as atypical ductal hyperplasia (ADH) or a lobular carcinoma in situ (LCIS).
“The vast majority of patients with high-risk lesions do not have cancer, and we’re trying to find the few that do,” says Manisha Bahl, co-author of the study and fellow doctor at MGH’s Department of Radiology. “In a scenario like this there’s always a risk that when you try to increase the number of cancers you can identify, you’ll also increase the number of false positives you find.”
Using their method, the researchers demonstrated how they could diagnose more cancerous lesions compared to traditional methods, meaning more surgeries could be avoided. “This work highlights an example of using cutting-edge machine learning technology to avoid unnecessary surgery,” said the director of clinical informatics in the Department of Radiology and Biomedical Imaging at the University of California, Marc Kohli. “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.”
The team will continue to work on and improve the model. “In the past, we might have recommended that all high-risk lesions be surgically excised,” says Lehman. “But now, if the model determines that the lesion has a very low chance of being cancerous in a specific patient, we can have a more informed discussion with our patient about her options. It may be reasonable for some patients to have their lesions followed by imaging rather than surgically excised.” Moving forward the team are hopeful the model can be used to detect other types of cancer or disease too. “A model like this will work anytime you have lots of different factors that correlate with a specific outcome,” says Barzilay. “It hopefully will enable us to start to go beyond a one-size-fits-all approach to medical diagnosis.”