by Heidi Mitchell
One in eight women is diagnosed with breast cancer. New AI-powered technology is poised to drastically decrease those odds.
For most of medical history, a woman’s breast cancer risk was calculated the same crude way: Did her mother have it? Her sister? Did she carry a known genetic mutation? If the answers were no, she was told, essentially, don’t worry. Come back at 40, then see you every year or two.
The trouble is that those family-history and genetic models only ever captured about 15 percent of the women who would go on to be diagnosed. “We didn’t have a way to find those 85 percent in advance,” says Dr. Constance Lehman, a professor of radiology at Harvard Medical School and a breast imaging specialist who has spent her career at the bleeding-edge of radiology research. “Until now.”
Dr. Lehman’s aha moment occurred in 2016, when she began collaborating with a colleague at Massachusetts Institute of Technology on imaging and machine learning. Radiologists had spent decades training computers to spot cancer indicators humans had long been trained to see. Dr. Lehman wondered if, with the rapid advancements in artificial intelligence, a machine could look at a woman’s mammogram today and predict whether or not she was likely to develop cancer in five years.
Turns out, it could. By feeding the model thousands of mammograms from women who later developed cancer alongside thousands from women who never did, Dr. Lehman built an algorithm that could detect faint biological “signals”— subtle patterns in breast tissue shaped by body mass index, hormonal history, pregnancy, breastfeeding, menopause, and other factors—so faint that no radiologist could reliably read them, but that were, in effect, written onto the tissue itself. “All of those weak signals that people had studied… those biological events are laid down in the breast tissue,” Dr. Lehman explains. “The AI can extract those predictive signals.”
“None of this would have happened without BCRF,” she says, of the support the organization gave to develop her research. When Dr. Lehman decided to leave the lab and build a company, BCRF went further, with the organization’s board and CEO, Donna McKay, pledging its first venture philanthropy investment in this early, game-changing product. The result was Clairity, a for-profit company, built on a genuinely global dataset spanning the US, South America, and Europe. This corrected what had become an established blind spot in AI medicine: Roughly 95 percent of health care AI models had historically been trained on patients from just three states.
Today, Clairity has been folded into national screening guidelines, which also changed the recommended baseline breast cancer risk assessment from age 40 to age 35, also a major win, since breast cancer patients have been trending younger. Every woman screened with Clairity receives a score that represents her individualized five-year breast cancer risk. Any score above 1.7 percent qualifies a patient for enhanced screening; above 3 percent, and she’s classified as high risk, which triggers supplemental imaging and risk-reduction conversations around anything from diet, exercise, and alcohol consumption to taking preventative medications to pre-emptive mastectomies, that, until now, were reserved almost exclusively for women with a strong family history or known genetic mutation.
What excites Dr. Lehman most isn’t just earlier detection—which is a big deal—it’s what becomes possible once breast cancer risk itself is something doctors can measure and track. “The whole field of cancer prevention has just opened up,” says Dr. Lehman.
