AI Now Reads Mammograms Alongside Radiologists at Yale

Artificial intelligence is moving from research labs into routine breast cancer screening, where it now reads mammograms alongside human radiologists and catches cancers that might otherwise be missed.

John Lewin, associate professor of radiology and biomedical imaging at Yale School of Medicine and division chief of breast imaging, has watched medical imaging evolve from film to digital and now to AI-assisted interpretation. In a recent interview, he outlined how the technology is reshaping mammography—and where it still falls short.

From single readers to human-plus-AI

The integration of AI into mammography follows different paths in the United States and Europe. European countries have traditionally used two radiologists to read each mammogram, with a third breaking any tie. Now AI is replacing one of those human readers, creating a human-plus-computer pairing.

In the U.S., where single-reader interpretation has been standard, AI is being added as a second set of eyes. "It's been shown that a single reader plus AI will find more cancers," Lewin said. The approach essentially brings double-reading to American screening without requiring twice the radiologist time.

The question of whether AI could eventually read mammograms alone remains open. While the technology continues improving, Lewin noted it will be difficult to surpass the performance of human and AI working together.

Predicting risk, not just detecting cancer

Beyond identifying existing tumors, AI systems are being developed to predict which patients face higher future cancer risk based on breast tissue patterns. But translating those predictions into clinical action presents challenges.

AI might classify one group of women as having 10% lifetime breast cancer risk and another at 25%. The higher-risk group could theoretically benefit from annual MRI screening, which detects cancers earlier than mammography. But most women in that group will never develop breast cancer, leading to decades of expensive scans, unnecessary biopsies, and false positives.

Connecticut and the nation lack sufficient MRI capacity and funding to screen all women at elevated risk. "What we don't know is would that translate into saving lives, how many lives would we save, and would it be worth the extra cost," Lewin said.

Why it matters

Mammography reduces breast cancer deaths by roughly one-third, but the technology has changed little since the 1960s beyond the shift to digital imaging and 3D tomosynthesis. AI represents the first fundamental change in how mammograms are interpreted rather than captured. For healthcare systems struggling with radiologist shortages and rising screening volumes, AI-assisted reading could maintain or improve cancer detection rates without proportionally increasing physician workload. The technology also opens questions about risk-stratified screening that could reshape how healthcare allocates expensive imaging resources.

A long view on medical imaging

Lewin's perspective on imaging technology spans an unusual career arc. Before medical school, he worked at Kodak helping launch the first digital spy satellite in 1983—technology so classified that the U.S. government denied spy satellites existed until 1995. That same digital scanning technology later appeared in mammography, where Lewin helped pioneer the first clinical trial of digital mammography around 2000.

Digital mammography wasn't initially better than film, but it offered the same convenience that drove consumers from film cameras to digital: immediate image review, easy sharing, compact storage, and image manipulation. The transition took radiology from developing sheets of film prone to dust and scratches to systems that could integrate with AI.

The details were first reported by Yale News in an interview with Lewin published June 11, 2026.