Mammography is among the first areas in radiology where artificial intelligence (AI) is making a tangible impact. A recent study from Radboud University Medical Center in Nijmegen, Netherlands, published in Radiology (2025; DOI: 10.1148/radiol.243688 ), investigated how AI influences both the accuracy and visual search patterns of radiologists while reading screening mammograms.
The study involved 12 radiologists who reviewed mammograms from 150 women, 75 with breast cancer and 75 without. Each radiologist examined the images twice: once without AI assistance and once with AI highlighting suspicious areas and assigning a risk score from 0 to 100.
To track visual behavior, researchers used an eye-tracking system. A small camera with infrared lights positioned in front of the radiologist’s screen captured reflections from the eyes, allowing software to determine precisely which areas of the mammogram were being examined and for how long.
The findings confirmed that AI support enhances diagnostic performance:
• Sensitivity increased from 81.7% to 87.2%
• Specificity improved from 89.0% to 91.1%
• Overall diagnostic accuracy (ROC AUC) rose from 0.93 to 0.97
Eye-tracking data provided further insights:
• Radiologists spent more time on areas marked as high-risk by AI, ensuring careful review of subtle lesions.
• Low-risk AI scores allowed radiologists to move more quickly through clearly normal images, reducing unnecessary time spent.
• Overall examination time increased slightly (from 29.4 to 30.8 seconds per image), while the portion of the image that radiologists focused on decreased (9.5% vs. 11.1%), showing more targeted attention.
Despite the benefits, caution is needed to avoid “automation bias,” where overreliance on AI could lead radiologists to overlook obvious findings or make unnecessary recalls. Jessie Gommers, MSc, the study’s joint first author, emphasized that radiologists must remain accountable for their decisions and critically interpret AI suggestions.
Future research will focus on optimizing AI integration, such as determining when AI guidance should be shown immediately or on request and developing methods to indicate when AI is uncertain about its predictions.
This study demonstrates that AI does not replace radiologists; instead, it enhances their ability to detect cancer accurately and efficiently. By combining human expertise with AI assistance, radiologists can focus on the most relevant areas, potentially improving outcomes for women undergoing breast cancer screening.
References:
1. Gommers J. J. J., et al. Influence of AI Decision Support on Radiologists’ Performance and Visual Search in Screening Mammography. Radiology. 2025; DOI: 10.1148/radiol.243688
2. 2. RSNA. Behavioral Research Will Help Us Understand When AI Will Help Radiologists. Radiology Editorial, 2025.
