Efficacy of an Automated Pulmonary Embolism (PE) Detection Algorithm on Routine Contrast-Enhanced Chest CT Imaging...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Cardiovascular Imaging

Background:

• Pulmonary embolism (PE) management requires rapid diagnosis, driving AI development for dedicated chest CT scans (CTPA).
• The diagnostic accuracy of AI for incidental PE in non-dedicated CT imaging remains largely uninvestigated.

Purpose of the Study:

• To evaluate the performance of a commercial AI algorithm (Viz PE) in detecting incidental PE within non-dedicated contrast-enhanced CT chest imaging.
• To assess the accuracy, sensitivity, and specificity of the AI algorithm on both dedicated and non-dedicated CT datasets.

Main Methods:

• The Viz PE AI algorithm was applied to 130 dedicated and 63 non-dedicated contrast-enhanced CT chest exams.
• Performance metrics including accuracy, sensitivity, and specificity were calculated for PE detection in both imaging types.

Main Results:

• The AI algorithm achieved 90.48% accuracy for incidental PE in non-dedicated CT scans (sensitivity 0.14, specificity 1.00).
• Overall performance across both datasets showed 90.16% accuracy, with 96% specificity and 41% sensitivity.
• High specificity suggests AI can effectively rule in PE, but low sensitivity indicates a potential to miss incidental cases.

Conclusions:

• Commercial AI tools show potential as supportive systems for PE detection, especially in clinically indicated scans.
• Current AI limitations, particularly low sensitivity in identifying incidental PEs, necessitate ongoing radiologist oversight.
• Further development is needed to improve AI sensitivity for incidental findings in non-dedicated CT imaging.