AI enhances radiologists’ ability to detect breast cancer in real-world exams

Artificial intelligence models can actually help detect cancer and reduce the burden on doctors, according to the largest study of its kind. Radiologists who chose to use AI were able to identify an additional 1 in 1,000 breast cancers.

Alexander Katalinic and his colleagues at the University of Lübeck in Germany worked with about 200 board-certified radiologists to test an AI trained to identify signs of breast cancer from mammograms. Radiologists examined 461,818 women at 12 breast cancer screening centers in Germany between July 2021 and February 2023, allowing each woman to choose whether or not to use AI. As a result, 260,739 patients were examined by AI and a radiologist, and the remaining 201,079 patients were examined by a radiologist only.

Those who chose to use AI were able to detect breast cancer at a rate of 6.7 per 1000 scans. This is 17.6% higher than the 5.7 cases per 1000 scans for people who chose not to use AI. Similarly, when women diagnosed with suspected cancer underwent a biopsy, women diagnosed with AI were 64.5% more likely to undergo a biopsy in which cancer cells were found. Among women for whom AI was not used, the rate was 59.2%.

The scale of improvement in breast cancer detection with AI is “very positive and exceeded our expectations,” Katalinic said in a statement. “We were able to demonstrate that AI significantly improves cancer detection rates in breast cancer screening.”

“The goal was to show noninferiority,” says Stefan Bank of Vara, an AI company also participating in the study. “If we can prove that AI is as good as radiologists, it becomes an interesting scenario where we can reduce the workload. We were surprised that we were able to show an advantage.”

Over-reliance on AI in healthcare is a concern for some, as it risks missing signs of symptoms and could lead to a two-tiered treatment system where those who can pay are afforded the luxury of human touch. are. Radiologists spent less time examining scans that the AI ​​had already suggested were “normal,” meaning cancer was unlikely to be present, and scans that the AI ​​could not examine took an average of 16 seconds to examine. In contrast, there is some evidence that radiologists spend less time performing exams. Not classified. But these latest discoveries have been welcomed by those who specialize in the safe implementation of AI in healthcare.

“This study provides further evidence of the benefits of AI in breast cancer screening and should be a further wake-up call for policymakers to accelerate the adoption of AI,” she said. Ben Glocker At Imperial College London. “The results confirm what we have seen time and time again: With the right integration strategy, the use of AI is safe and effective.”

He welcomes the study's ability to empower radiologists to make their own decisions about when to use AI, and hopes to see more testing of AI in a similar way. . “This cannot be easily evaluated in the lab or in simulations, and instead we need to learn from real-world experience,” Glocker says. “The technology is ready. We need policies to follow now.”