Document is current

Article History

Received: 4 June 2025

Revised: 18 September 2025

Accepted: 1 December 2025

First Online: 13 January 2026

Compliance with ethical standards

:

: The scientific guarantor of this publication is Solveig Hofvind.

: Dr. C. Lee reports textbook royalties from McGraw-Hill, Inc., Oxford University Press, and UpToDate, Inc., research consulting fees from DeepHealth/RadNet, and personal fees for journal editorial board work from the American College of Radiology, all outside the submitted work. K.L. is affiliated with Unilabs. The other authors declare no conflicts of interest.

: No complex statistical methods were necessary for this paper, but Marthe Larsen is a statistician.

: Written informed consent was waived by the Institutional Review Board. The data was disclosed with a legal basis in the Cancer Registry Regulations section 3–1 and the Personal Health Filing System Act section 19a to 19h.

: Institutional Review Board approval was obtained, i.e., the study was approved by the Regional Committee for Medical and Health Research Ethics (2018/13294).

: Overall performance and possible strategies for combining AI and radiologists for almost an identical sample of the screening examinations (122,969 examinations) have previously been reported for version 1.7 ( ). The sample with results from 1.7 has also been included in an analysis of AI score on mammograms preceding a cancer diagnosis ( ). Since we could merge results from version 1.7 and 2.1 at an individual examination level, we reported the overall performance for 1.7 in this paper as well, since the sample included about 5000 fewer examinations.

: