To evaluate the diagnostic efficacy of commercially available artificial intelligence (AI) software to detect incidental pulmonary embolism (IPE) on chest CT and shorten the time to diagnosis using worklist prioritization.

In this prospective study, AI software was deployed in a clinical environment to analyze routinely acquired chest CT scans of adult oncology patients. Three time periods of 15 weeks each were compared: routine workflow without AI, manual triage without AI, and worklist prioritization with AI. Diagnostic accuracy of the tool was evaluated on both prospectively and retrospectively collected data. Temporal endpoints including Detection and Notification Times (DNT) were assessed.

A total of 11 736 CT scans were evaluated. Prevalence of IPE was 1.2% (n=143). The AI software detected 131 true positives, 12 false negatives, 31 false positives, and 11559 true negatives. Sensitivity was 91.6%, specificity 99.7%, NPV 99.9%, and PPV 80.9%. When applied retrospectively, the AI software found IPEs in 47 CTs (44.8%) that were missed in the radiology report. The median DNT for IPE positive examinations was 7714, 4973, 87 min for the respective time periods. The difference in DNT between positive and negative CTs was largest when using AI assistance and was significantly different from both workflows without AI. In contrast, no statistically significant difference was found between routine workflow and manual triage without AI.

A commercially available AI tool wafound to have a high diagnostic efficacy in detecting IPE on CT of oncology patients. AI assisted worklist prioritization washown to be effective in significantly reducing the time to diagnosiof IPE casecompared to the routine clinical workflow.

Incidental pulmonary embolism is a common comorbidity in oncology patients. Despite being clinically unsuspected, IPE can be an urgent and life-threatening event. AI based worklist prioritization can assist the radiologist to shorten the time to diagnosis of critical imaging findings.