by
Gus Iversen, Editor in Chief | July 09, 2025
Radiologists using AI decision support systems detected more breast cancer lesions in screening mammograms without spending additional time on interpretations, according to a study published July 8 in Radiology.
The study, led by researchers at Radboud University Medical Center in Nijmegen, Netherlands, investigated how AI affects visual search patterns and performance in breast cancer screening. Using eye tracking technology, the team compared how 12 radiologists interpreted 150 mammography cases — half with confirmed cancer — both with and without AI assistance from ScreenPoint Medical's Transpara (v2.1.0).
"By analyzing this data, we can determine which parts of the mammograms the radiologist focuses on, and for how long, providing valuable insights into their reading patterns," said joint first author Jessie J. J. Gommers, M.Sc., of the Department of Medical Imaging.
Ad Statistics
Times Displayed: 17839
Times Visited: 48 Brand-New FDA-cleared Advanced Ultrasound Medical Device available for sale or lease to Wound Care Centers or any other Medical Facilities.The Arobella 1000D is designed for non-contact or debridement ultrasound wound healing therapy, or any other wounds
The results showed that radiologists more accurately identified breast cancer when aided by AI, with no measurable change in average sensitivity, specificity or time spent per case. Eye tracking data indicated that radiologists spent more time reviewing areas that contained lesions when AI support was active.
"Radiologists seemed to adjust their reading behavior based on the AI's level of suspicion," Gommers said. Low AI scores appeared to reassure readers, allowing them to move more quickly through normal cases, while higher scores prompted additional scrutiny.
The AI tool used region-specific markings that acted as visual cues, directing radiologists' attention to areas of interest. According to Gommers, this functioned as “an additional set of eyes,” enhancing both accuracy and workflow.
However, she also noted potential risks, including overreliance on incorrect AI suggestions. “Educating radiologists on how to critically interpret the AI information is key,” Gommers said.
Ongoing research aims to refine how and when AI support should be offered, and to assess whether AI uncertainty can be predicted in advance to further guide its use.
