When Covid Came in Massachusetts, he forced Constance Lehman change the way Massachusetts General Hospital screens women for breast cancer. Many people were skipping regular checks and scans due to concerns about the virus. Thus, the Lehman codirects center began to use a artificial intelligence algorithm to predict who is most at risk of developing cancer.
Since the start of the epidemic, Lehman says, about 20,000 women have skipped routine screening. Normally, five out of every 1000 women screened show signs of cancer. “These are 100 cancers that we haven’t diagnosed,” she says.
Lehman says the AI approach has helped identify a number of women who, when persuaded to come for routine screening, turn out to have early signs of cancer. Women reported by the algorithm were three times more likely to develop cancer; previous statistical techniques were no better than chance.
The algorithm scans past mammograms and seems to work even when doctors haven’t seen any warning signs in those past scans. “What AI tools do is extract information that my eye and brain cannot,” she says.
Researchers have long touted the potential of AI analysis in medical imaging, and some tools have found their way into medical care. Lehman has worked with researchers at MIT for several years on ways to apply AI to cancer screening.
But AI is potentially even more useful in predicting risk more accurately. Screening for breast cancer sometimes involves not only looking at a mammogram for cancer precursors, but also collecting information about the patients and feeding them both into a statistical model to determine the need for screening. feedback.
Adam yala, a doctoral student at MIT, began developing the algorithm Lehman uses, called Mirai, before Covid. According to him, the goal of using AI is to improve early detection and reduce the stress and cost of false positives.
To create Mirai, Yala had to overcome issues that hampered other efforts to use AI in radiology. He used a contradictory machine learning approach, where one algorithm tries to fool another, to account for differences between x-ray machines, which could mean that patients who face the same risk of breast cancer get different scores. The model was also designed to aggregate data from multiple years, making it more accurate than previous efforts that include less data.
The MIT algorithm analyzes the four standard views of a mammogram, from which it then infers information about a patient that is often not collected, such as a history of surgery or hormonal factors such as menopause. It can help if this data has not already been collected by a doctor. Details of the work are described in an article published today in the journal Scientific translational medicine.