“What are the odds that you will get breast cancer in the next few years?
Soon, AI might tell you the answer.
Researchers and companies are designing artificial-intelligence models to predict a woman's near-future breast-cancer risk, as well as risks of lung cancer and other diseases. The AI algorithms are trained on mammograms from past patients, some of whom would go on to develop breast cancer, to pinpoint who is most at risk before the disease is visible to the human eye.
Clairity and DeepHealth are among the companies looking to soon bring their own predictive algorithms into the U.S. market, pushing AI beyond its current medical use as a diagnostic helper.
The new tools could help doctors decide if a patient needs more screening or even preventive drugs.
"I cannot look at a mammogram and predict a woman's five-year risk of breast cancer with any accuracy, nor can any of my colleagues," says Dr. Connie Lehman, Clairity's founder and professor of radiology at Harvard Medical School. "We're in a domain where the human isn't able to do this."
Clairity's AI model is designed to predict a woman's five-year breast-cancer risk from a routine mammogram. The model was trained on more than 400,000 routine mammograms from existing records, Lehman says, matched with the women's breast-cancer status five years later. That enabled the AI to identify patterns in breast tissue that would predict future cancer, researchers suspect. The signs are so subtle and detailed that humans have yet to differentiate them on their own.
Experimenting with the tool, the team saw that the AI could also predict a woman's age, menopausal status and whether she had birthed children, after analyzing just one of her mammograms.
"A woman's life experience is laid down in the image of her breast tissue," Lehman says.
The Clairity AI model received Food and Drug Administration authorization in May, the first of its kind to do so. The company is planning to introduce it in clinics later this year and is speaking with insurers about potential coverage.
Tools currently available to determine breast-cancer risk are lower tech. Today, women are quizzed about their age, race, family history, breast density and other factors that influence their odds of breast cancer. From there, calculators estimate a woman's near-term or lifetime risk; a five-year risk of around 3.0% is considered high, as is a lifetime risk of 20% or more.
The new AI models that analyze a mammogram often outperform the older risk-score calculators, studies show. Around 60% of the time, the calculators correctly give a higher risk score to a woman who will develop cancer over a woman who won't. For the AI models, that increases to around 70% of the time or more, data show, depending on the group of patients and the specific model.
"Although these sound like small numbers, it's a really big, stepwise change," says Dr. Vignesh Arasu, a radiologist and AI researcher at Kaiser Permanente Northern California Division of Research, who has published research comparing the traditional risk calculators with the newer AI tools.
Using the AI without the calculators could save time in the clinic, says Dr. Graham Colditz, a cancer-prevention researcher at Washington University School of Medicine in St. Louis.
Colditz and his team at WashU are planning a clinical trial to test their own model in development. The AI algorithm will predict the participants' five-year breast-cancer risk during their regular screening and then sort them into risk groups. Then, the trial would triage women to get either more or less screening, based on their level of risk, and track them. Earlier this year, WashU's AI risk-prediction tool was acquired by Lunit, a Seoul-based medical AI company.
Women flagged as higher-risk could get MRIs or other tests or preventive medications such as tamoxifen, which blocks estrogen and reduces breast-cancer risk. That would enable them to find the disease in its infancy or avoid it altogether. Low-risk women, on the other hand, might need less screening.
"We can make this process really much more tailored to a woman's risk," says Colditz. "If high-risk women were appropriately identified, we'd see a substantial reduction in breast cancer being diagnosed."
Still, some doctors hesitate to embrace the technology, wanting more evidence that it reduces advanced disease and death. Studies so far haven't looked at long-term patient outcomes. Some of the tools might flag too many women who could develop low-risk breast lesions that wouldn't have caused them harm.
"I think this is wonderful, but I do think this may be early still for us to know how effective it is," says Dr. Stamatia Destounis, chair of the American College of Radiology Breast Imaging Commission. "Do they have any impact on mortality outcomes? We don't have that information."
The companies and researchers say that the quiz-based calculators that are already in use don't have data on mortality outcomes. And clinical trials, some of which have started, take years to complete; the AI models could be outdated by the time the trial is done.
Some 30% of radiology practices in the U.S. already use some form of AI in breast-imaging care, according to survey data from the American College of Radiology. But most aren't using it to predict a person's risk. Right now, some patients can pay to add on an additional AI reading of their mammogram, on top of the radiologist, and some clinics use AI to triage suspicious-looking scans to the top of the to-do list.
Doctors also worry that Al models might not work as well in diverse patient populations, but data so far suggest that some are up to the task. An AI model called Sybil, designed to predict lung-cancer risk from a CT scan, was highly accurate among a predominantly Black population in Illinois, recent data show. Previous U.S. studies had tested the algorithm on mostly white patients, the researchers say. The model has also been successfully tested on patient data from South Korea and Taiwan.
The algorithm from researchers at the Massachusetts Institute of Technology and Mass General Brigham could help doctors move beyond smoking history when thinking about who might be at risk for lung cancer; right now, only those with a significant cigarette history are eligible for screening.
Researchers are also designing algorithms to identify which chronic kidney-disease patients will likely progress to end-stage renal disease.
Another team used more than one million electronic health records to develop AI models to predict whether a person's genetic mutation will actually lead to a corresponding disease, including for some cancers and a hereditary heart disease.
At DeepHealth, which has its own AI risk predictor for breast cancer already available in Europe, researchers are looking at how a single mammogram could predict health risk even beyond cancer.
"I'm screening for cancer, but can I look at the calcification in the same exam and have a risk assessment of cardiovascular disease?" says Dr. Niccolo Stefani, business and product leader for population health and clinical AI at DeepHealth. "You deliver double value."” [1]
1. AI to Help Predict Breast-Cancer Risk --- New models could identify patients at risk of developing the cancer and who might need more screening. Abbott, Brianna. Wall Street Journal, Eastern edition; New York, N.Y.. 30 Sep 2025: A10.
Komentarų nėra:
Rašyti komentarą