YuLin Zhen, Photography Editor

Doctors at the School of Medicine are debating whether artificial intelligence mammography could be used to diagnose and predict breast cancer more accurately.

Detecting breast cancer is particularly challenging. For decades, doctors have tried to improve breast cancer diagnosis using computer-aided detection, but their efforts were generally fruitless. Recent advances in AI and machine learning, however, may enhance breast cancer detection rates. AI technologies can be employed to analyze mammographic images and identify subtle patterns that might indicate early signs of breast cancer that could be missed by the human eye. 

These AI models assist radiologists by acting as a second reader, flagging potential areas of concern that warrant closer inspection. This dual-review process, combining AI with human expertise, aims to refine the screening process, making it not only faster but also more accurate.  However, AI mammography is hamstrung by several limitations.

“I think when used as a second reader to a human, the combination of human and AI has the potential and will almost certainly be slightly better than a human alone or AI alone,” John Lewin, an associate professor and chief of radiology and biomedical imaging at the School of Medicine, told the News.

Liane Philpotts, professor and chief of Radiology and Biomedical Imaging, on the other hand, offers a grounded perspective. She noted that while mammograms often correctly identify patients with breast cancer, they are less successful when encountering dense breast tissue. 

“Where mammography suffers the most is in women with dense tissue because that’s where most of the cancers are going to get hidden,” Philpotts said. 

AI mammography may help overcome these limitations. According to Lewin, when two physicians read a mammogram, it produces more accurate diagnoses than just one physician, and the combination of a physician and AI is better than one physician and about the same as two physicians.

“Detecting breast cancer on mammography is incredibly challenging at times,” Philpotts told the News. “It tends to help some of the more junior people, the less experienced readers can benefit perhaps from the AI, but the AI doesn’t do better than the experienced readers.”

Both Lewin and Philpotts are cautious about over-reliance on AI. According to Lewin, it is important to integrate AI judiciously, ensuring it complements rather than supplants human expertise. 

“AI is not magic. It can get as good as a human, but it’s not going to drastically increase things,” Lewin told the News. “Based on studies of double reading, you might find maybe 10 percent, maybe 15 percent more cancers, but when you’re doing screen mammography, there [are] four, five, cancers per thousand screening mammograms. It takes thousands of women for that to make a significant difference.” 

Additionally, Andrejeva Liva-Wright, an associate professor of Radiology and Biomedical Imaging, raised concerns about the potential risks associated with AI, such as inducing biases in clinical judgment. 

“Studies that show that it does affect your judgment in a way. Regardless of how expert you may be, it can give you biases,” Liva-Wright said. 

Despite these risks, the doctors agree that using AI mammography could facilitate personalized medicine. According to Liva-Wright, AI mammography could help radiologists issue more personalized and accurate screening recommendations.

In 2019, 69.1 percent of women aged 40 and older had a mammogram within the past 2 years.