Skip to content

Closing the Performance Gap between Generalists and Breast Imaging Specialists Using a Nationally Deployed AI Workflow for Screening Mammography

July 21, 2026
2 min

July 21, 2026

Radiology

Matthew P. McCabe, PhD; Edgar A. Wakelin, PhD; Leeann D. Louis, PhD; Annie Y. Ng, PhD; Diana S. M. Buist, PhD; Christoph I. Lee, MD; A. Gregory Sorensen, MD; Bryan Haslam, PhD

Purpose

General radiologists interpret most screening mammograms in the United States and previous research has reported  a gap in screening performance between general radiologists and fellowship-trained breast imaging specialists. This study examined clinical performance before and after implementation of a multistage artificial intelligence-supported workflow, with the goal of evaluating product impact on general radiologists compared to breast imaging specialists.

Methods and Materials

This prospective, multicenter, observational study included 577,742 screening digital breast tomosynthesis examinations interpreted by 95 radiologists at 109 imaging facilities between September 2021 and December 2022. The radiologist group included 60 general radiologists and 35 fellowship-trained breast imaging specialists.

The workflow combined an AI-driven computer-aided detection and diagnosis algorithm Mammo DX with the Safeguard Review. Investigators compared cancer detection rate, recall rate, and positive predictive value of recalls before and after implementation.

Results

Among general radiologists, the adjusted cancer detection rate increased from 3.76 to 4.99 cancers per 1,000 examinations following implementation of the AI-supported workflow (P < .001). Their adjusted positive predictive value of recalls increased from 3.38% to 3.89% (P = .02). The adjusted recall rate also increased, from 9.06% to 10.40% (P = .007).

During the AI workflow period, no statistically significant differences were observed between general radiologists and breast imaging specialists in adjusted cancer detection rate (P = .53) or positive predictive value of recalls (P = .64). Among breast imaging specialists, cancer detection rate, recall rate, and positive predictive value of recalls did not change significantly between the two study periods.

Conclusion

Implementation of the multistage AI-supported workflow was associated with improved cancer detection and positive predictive value of recalls among general radiologists in routine U.S. screening practice. During the AI workflow period, their adjusted cancer detection rate and positive predictive value of recalls were not statistically different from those of fellowship-trained breast imaging specialists.

Together, these results suggest that the multi-stage AI driven workflow could be an effective tool to close the performance gap between generalist and specialist radiologists in the real world.

Read full article

Disclaimers: DeepHealth Prostate is manufactured as Quantib Prostate by Quantib BV for DeepHealth Inc., DeepHealth Lung is manufactured as Veye Lung Nodules by Aidence BV for DeepHealth Inc., DeepHeath Lung Tracker is manufactured as Veye Clinic by Aidence BV for DeepHealth Inc. and DeepHealth Brain is manufactured as Quantib ND by Quantib BV for DeepHealth Inc. Not all products and functionalities are commercially available in all countries. For clearance and commercial availability in your geography of functionalities listed and compatibility with other systems, please contact us