AI Support Improves Accuracy in Breast Cancer Screening

Artificial intelligence (AI) can help radiologists detect breast cancer more accurately by guiding their attention toward suspicious regions in mammograms, a new study published in Radiology has shown. Researchers used eye tracking data to assess how AI alters the way radiologists view mammographic images during cancer screening.

Previous studies have found that AI decision support systems improve radiologists' sensitivity without increasing reading time. However, little has been known about how AI affects their visual search behavior.

To explore this, researchers at Radboud University Medical Center in the Netherlands analyzed eye tracking data from 12 radiologists who reviewed 150 mammography cases, half of which included breast cancer. The study compared unaided readings with those performed using an AI decision support system.

Tracking gaze to map attention

The research team used a device with two infrared lights and a camera, mounted in front of the radiologist’s monitor. By capturing infrared reflections from the eyes, the system computed the radiologist’s gaze position on the screen in real time. This allowed the team to understand how long and where radiologists focused their attention during image interpretation.

The results showed a consistent pattern: radiologists spent more time examining actual lesion regions when AI support was available. Detection accuracy improved with AI, though sensitivity, specificity, and reading time remained statistically unchanged.

Visual prompts guide diagnostic behavior

The AI system displayed suspicious regions on the images using markers. These acted as visual cues, prompting radiologists to look more closely at highlighted areas. When AI scores indicated low suspicion, radiologists tended to move through cases more quickly. When high suspicion was indicated, they revisited areas with greater scrutiny.

According to the data, this shift in visual behavior contributed to improved performance without an increase in time spent per case. The AI appeared to function as an auxiliary reader, complementing the radiologist’s own interpretation by highlighting potentially concerning regions.

Balancing AI benefits with human oversight

Despite the improved outcomes, researchers cautioned that overreliance on AI could pose risks if the system misclassifies findings. Errors might lead to missed cancers or unnecessary follow-up imaging. As a result, ensuring high AI accuracy and maintaining radiologist accountability remain essential.

To support this, ongoing research is evaluating when AI support should be presented during the diagnostic process and developing methods to identify when the AI’s confidence is low.

Reference: Gommers JJJ. Verboom SD, Duvivier KM, et al. Influence of AI decision support on radiologists’ performance and visual search in screening mammography. Radiology. 2025. doi: 10.1148/radiol.243688

This article has been republished from the following materials. Note: material may have been edited for length and content. For further information, please contact the cited source. Our press release publishing policy can be accessed here.

This content includes text that has been generated with the assistance of AI. Technology Networks' AI policy can be found here.