How AI Is Making Mammograms More Accurate

The AI revolution is changing medicine, especially in the realm of diagnostics. Here’s how it’s helping doctors detect breast cancer sooner

Sari Harrar,

AARP

Comments

Published September 30, 2025

Glenn Harvey

When Teresa McKeon arrived for her annual mammogram in August 2024, she said yes to an extra, AI-assisted review of the images. It cost $40.

“My sister was diagnosed with breast cancer at age 47,” says McKeon, 57, of Sherman Oaks, California. “I have dense breasts, which raises risk. Breast cancer is something you have to stay on top of with every tool in the toolbox.”

Her mammogram found tiny white calcifications in her left breast. A biopsy the next day confirmed that McKeon had ductal carcinoma in situ — cancerous cells in milk ducts.

“It was very small, very early, very treatable,” says McKeon, vice president of production at an entertainment marketing company.

A lumpectomy last October removed the carcinoma before it could turn into an invasive cancer. After 25 radiation sessions, she now takes the estrogen-lowering drug anastrozole to reduce the risk of recurrence.

McKeon is one of more than 1.5 million American women who opt for an AI-assisted review of their screening test for breast cancer annually.

AI doesn’t create the mammogram images or replace the doctors who read them, says radiologist Jason McKellop, M.D., medical director of breast imaging in Southern California for RadNet and Breastlink. It simply provides a second set of eyes.

“The ultimate burden falls on the human being,” McKellop says.

Conventional screening mammograms miss 20 percent of existing breast cancers, according to the National Cancer Institute. “AI can provide meaningful benefit, especially for cancers that could be missed,” says radiologist Manisha Bahl, M.D., associate professor of radiology at Harvard Medical School and a specialist in breast imaging at Massachusetts General Hospital in Boston.

For a recent study, Bahl and her colleagues evaluated the mammograms of 764 women obtained between 2016 and 2018. They found that while the traditional, non-AI readings identified 73 percent of cancers, the AI-assisted system detected 94 percent, and did so with a significantly lower false-positive rate.

Most Popular

“Will AI for mammograms save lives?” Bahl asks. “We assume that AI will lead to improved cancer detection rates, which will in turn lead to better long-term outcomes for patients. But we currently lack the real-world data to support this.”

In the future, AI reviews of mammograms may be able to determine which suspicious findings are likely to be cancer, cutting the need for biopsies for low- to moderate-risk patients in half, according to a 2023 study from Houston Methodist Neal Cancer Center.

AI may also help track the success of chemotherapy before breast cancer surgery and even predict future cancer risk by examining normal breast tissue on cancer-free mammograms. A system that does this, called Clairity Breast, received FDA approval in June 2025. For now, women who want an AI mammogram reading will likely have to pay an extra $40 to $100.

“The closest thing we have to curing breast cancer at this point is early detection,” McKeon says. “I’m all for doing whatever we can.”

Other AI advances in cancer diagnosis

Identify skin cancers. A handheld, AI-enabled device the size of a cellphone helped primary care doctors identify potential skin cancers in the office more accurately than a visual inspection, per a 2025 University of North Carolina at Chapel Hill study. The FDA-cleared device, developed by DermaSensor of Miami, emits bursts of light over suspicious skin areas and analyzes the reflections.

Find prostate cancers. An AI-assisted review added to a doctor’s reading of magnetic resonance imaging improved the detection of clinically significant prostate cancer in a recent study of 360 men (median age: 65) from the international Prostate Imaging-Cancer AI Consortium. The extra reading increased detection by 3.3 percent. The algorithm is still in research.

Detect lung cancers. Using clues from primary care doctors’ notes, an experimental AI algorithm found potential lung cancers an average of four months sooner than doctors might have on their own. The study from the Netherlands, published in March 2025, was based on data from 525,526 patients, including 2,386 with lung cancer.

%{postComment}%