clinical study

Clinical outcome of incidental pulmonary embolism detected by artificial intelligence software: a retrospective analysis

Materials & Methods

iPE algorithm was applied retrospectively to 2793 consecutive patients undergoing Chest CT between 05/2020 to 01/2021 and compared to the original radiology report. Concordant cases between the original report and AI were considered ground truth. Discordant cases deemed positive by AI and negative by the report were reassessed by the Radiologist.


iPE prevelance was 2.3% (65/2792). 45 cases were positive by both AI and radiologist report and AI detected 23 additional discordant positive cases.87% (20/23) were considered TP on secondary review. The AI enhanced detection rate was 44.4% (20/45). 10% (2/20) were chronic PE, 90% (18/20) were acute/subacute PE. 70% (14/20) were subsegmental PE. 40% (8/20) had follow-up imaging: In 50% (4/8) cases iPE was not resolved. In 38% (3/8) patients iPE was noticed on follow-up and reported. In all 3, anticoagulation therapy was initiated. The average treatment delay was 132 days. In 35% (3/20) discrepant anticoagulation was given for other reasons. In the remaining 65% (13/20) cases, iPE was unnoticed by the radiologist or clinician and the patient did not receive any treatment. 25% (5/20) patients died of unrelated causes.


The improved detection rate of iPE by AI may bring significant benefits for prompt management in selected individuals, especially the ones at risk for recurrent thromboembolic events.