Back to the list
Congress: ECR25
Poster Number: C-18295
Type: Poster: EPOS Radiologist (scientific)
Authorblock: M. B. Bergan, M. Larsen, J. Gjesvik, Å. S. Holen, N. Moshina, T. Hovda, H. W. Koch, M. A. Martiniussen, S. Hofvind; Oslo/NO
Disclosures:
Marie Burns Bergan: Nothing to disclose
Marthe Larsen: Nothing to disclose
Jonas Gjesvik: Nothing to disclose
Åsne Sørlien Holen: Nothing to disclose
Nataliia Moshina: Nothing to disclose
Tone Hovda: Nothing to disclose
Henrik Wethe Koch: Nothing to disclose
Marit Almenning Martiniussen: Nothing to disclose
Solveig Hofvind: Nothing to disclose
Keywords: Breast, Mammography, Computer Applications-Detection, diagnosis, Cancer
Purpose To explore different combinations of radiologists’ initial interpretation scores and artificial intelligence (AI) malignancy scores, assigned to screening mammograms of women who participated in BreastScreen Norway and diagnosed with cancer.
Read more Methods and materials This retrospective cohort study included 1,027,430 screening examinations, including 5785 screen-detected cancers, performed in BreastScreen Norway, 2004-2021. All examinations were independently interpreted by two breast radiologists. The radiologists scored each breast from 1 to 5, where 1 indicated negative for abnormality; 2, probably benign; 3, intermediate suspicion; 4, probably malignant; 5, high suspicion of malignancy (1). All examinations were processed by the AI system Lunit INSIGHT MMG version 1.1.7.2, which provided a continuous malignancy score from 0 to 100, with...
Read more Results Of the screen-detected cancers, 12.9% (745/5786) had an interpretation score of 1 by one of the radiologists and 2 by the other (1+2), 12.1% (701/5786) had a score of 1+3/4/5, 12.4% (716/5786) 2+2, 17.3% (1001/5786) 2+3/4/5, and 45.3% (2623/5786) had a score of 3/4/5+3/4/5 (Table 1). For interval cancers, the percentages were 79.7 (1421/1783) for interpretation score 1+1, 13.8% (246/1783) for score 1+2, 1.5% (27/1783) for 1+3/4/5, 3.7% (66/1783) for 2+2, and 0.8% (14/1783) for 2+3/4/5.When defining 10% of the...
Read more Conclusion Almost 98% of the screen-detected cancers with an interpretation score of ≥3 by both radiologists were identified by AI at a 10% threshold, while 91% of screen-detected cancers with a score of ≥3 by one radiologist were identified. At a 10% threshold AI identified 35% of the interval cancers interpreted negative by both radiologists.
Read more References 1. Bjørnson EW, Holen ÅS, Sagstad S et al (2022) BreastScreen Norway: 25 years of organized screening. Oslo: Cancer Registry of Norway. Cited February, 2025. Avaliable from: https://www.kreftregisteret.no/Generelt/ Rapporter/Mammografprogrammet/25-arsrapport-mammografprogrammet/
Read more
GALLERY