Whether you’re building AI models or looking to apply those models in your work, you’ll want to see this new study examining AI’s performance in detecting hard-to-find breast cancers.

In this study, partially funded by NCI, researchers applied a deep-learning tool to search for interval breast cancers (IBCs)—those cancers that arise between routine screening mammograms and which tend to be particularly aggressive and hard to treat.

By examining images from the same patient at different times, the researchers were hoping to answer the question, “Was this cancer missed by the radiologist in the earlier screening, or was the cancer simply not visible at that time?”

The researchers found that, of the 131 mammograms they scored, AI flagged 76% for possible IBCs. These cancers were mostly missed-reading errors or were subtle changes not found on the earlier scan.

These findings reinforce a role for AI, potentially serving as an adjunct to humans by flagging images that need more attention.

Corresponding author on the study, Dr. Tiffany Yu, of the University of California-Los Angeles’ Department of Radiological Sciences, noted, “Our study shows that AI has the potential to increase screening sensitivity and could help support the radiologist’s decision making. However, it will be up to the radiologist to accurately assess those flagged AI exams to confirm cancer.”

She added, “In general, our study reaffirms the value in radiologists reviewing and classifying IBCs and shows how tools like AI could be used to improve screening sensitivity. These are exciting initial results, but we still have a lot to explore to understand AI’s limitations and how it can be used in a real-world setting.”