Automated detection of large vessel occlusion using deep learning: a pivotal multicenter study and reader performance...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Large vessel occlusion (LVO) is a critical factor in ischemic stroke.
• Early and accurate detection of LVO in CT angiography (CTA) is vital for timely treatment.
• Evaluating AI software for LVO detection and its impact on physician performance is essential.

Purpose of the Study:

• To assess the standalone efficacy of AI software in detecting LVO on CTA.
• To determine if AI software improves diagnostic accuracy for early-career physicians.
• To evaluate the impact of AI on stroke workflow efficiency.

Main Methods:

• A multicenter study involving 595 ischemic stroke patients.
• AI software performance benchmarked against expert consensus for LVO detection.
• Diagnostic accuracy of physicians with and without AI assistance was compared using AUROC, sensitivity, and specificity.

Main Results:

• The AI software demonstrated high sensitivity (0.858) and specificity (0.969) for LVO detection.
• Within 24 hours of symptom onset, AI achieved an AUROC of 0.973.
• AI assistance significantly improved physician sensitivity by 4.0% and AUROC by 0.024 (P<0.001).

Conclusions:

• AI software exhibits high efficacy in detecting LVO on CTA.
• AI significantly enhances the diagnostic accuracy of early-career physicians in identifying LVO.
• The integration of AI streamlines stroke workflow in emergency settings.