Artificial intelligence-based computer-assisted detection/diagnosis (AI-CAD) for screening mammography: Outcomes of AI-CAD in the mammographic interpretation workflow

PurposeTo evaluate the stand-alone diagnostic performances of AI-CAD and outcomes detected abnormalities when applied to mammographic interpretation workflow.MethodsFrom January 2016 December 2017, 6499 screening mammograms 5228 women were collected from a single facility. Historic reads three radiologists used as radiologist interpretation. A commercially-available was for analysis. One not involved in had retrospectively reviewed abnormality features assessed significance (negligible vs. need recall) marks. Ground truth terms cancer, benign or absence confirmed according histopathologic diagnosis negative results on next-round screen.ResultsOf mammograms, 6282 (96.7%) negative, 189 (2.9%) benign, 28 (0.4%) cancer group. 5 (17.9%, 28) 9 cancers that intially interpreted negative. Of 648 recalls, 89.0% (577 648) marks seen examinations group, 267 (41.2%) considered be negligible. Stand-alone has significantly higher recall rates (10.0% 3.4%, P < 0.001) with comparable sensitivity detection (P = 0.086 0.102, respectively) compared radiologists’ interpretation.ConclusionAI-CAD 17.9% additional mammography initially overlooked by radiologists. In spite detection, clinical workflow, which are mammograms.