Artificial Intelligence Enhances Detection of Breast Lesions in Mammograms

Artificial intelligence (AI) is revolutionizing breast cancer screening by helping radiologists identify more potential lesions in mammograms, thereby increasing diagnostic accuracy. A recent study published in Radiology demonstrates that AI support enables radiologists to focus more efficiently on suspicious areas, improving their performance without extending reading times.

Previous research has established that decision-support AI can boost radiologists’ sensitivity for cancer detection. However, its impact on how radiologists visually search and interpret mammograms was less understood. To explore this, researchers employed an eye-tracking system—using infrared lights and a camera—to monitor how radiologists examine mammograms with and without AI assistance.

The study involved 12 radiologists reading 150 mammograms from women, including 75 with confirmed breast cancer and 75 without. The AI system provided visual cues by marking regions of concern, guiding radiologists’ attention to areas of potential abnormalities.

Findings showed that, with AI support, radiologists achieved higher diagnostic sensitivity. Eye-tracking data revealed they spent more time examining regions with actual lesions when aided by AI, indicating more targeted and thorough analysis. Notably, radiologists tended to adjust their reading behavior based on AI suspicion levels: low AI scores reassured them to proceed faster, while high scores prompted more careful inspection.

The AI's markings acted as visual cues, highlighting key areas and enhancing both the accuracy and efficiency of interpretation. Gommers emphasized that AI helps radiologists prioritize suspicious regions, acting as an additional set of eyes and reducing the chances of missing subtle signs of cancer.

While overreliance on AI could pose risks, such as missed diagnoses if the AI makes errors, current evidence suggests that AI performance is comparable to human expertise in mammography. Ensuring AI accuracy and training radiologists on critical interpretation are vital to mitigate potential issues.

The researchers are now investigating optimal timing for AI support—whether it should be available immediately or upon request—and methods for AI to indicate its certainty level, aiming to make AI support more precise and judicious in clinical settings.

Overall, AI shows significant promise in improving breast cancer screening by enhancing radiologists’ focus, reducing missed lesions, and streamlining workflow, supporting earlier and more accurate diagnosis.