Using a Deep Learning-Based Decision Support System to Predict Emergent Large Vessel Occlusion Using Non-Contrast...

Area of Science:

• Neurology
• Radiology
• Medical Imaging

Background:

• Retrospective, multi-reader, blinded trial evaluating an AI-based clinical decision support system.
• Focus on improving clinician detection of emergent large vessel occlusion (ELVO) using brain non-contrast computed tomography (NCCT).

Purpose of the Study:

• To assess the performance of an AI system in detecting ELVO compared to unassisted clinician readings.
• To evaluate the impact of AI assistance on diagnostic accuracy metrics like sensitivity, specificity, and AUROC.

Main Methods:

• 477 patients enrolled; 112 with anterior circulation ELVO, 365 controls.
• Clinicians performed unassisted and AI-assisted readings of NCCT images after a 2-week washout period.
• Primary endpoints: sensitivity and specificity; Secondary endpoints: AUROC and individual-level accuracy.

Main Results:

• AI-assisted readings significantly improved sensitivity (92.0% vs 75.9%) and specificity (92.6% vs 83.0%) compared to unassisted readings (p < 0.01).
• AI assistance also increased accuracy (92.5% vs 81.3%) and AUROC (0.95 vs 0.87) (p < 0.01).
• The AI system demonstrated high standalone performance with 88.4% sensitivity and 91.2% specificity.

Conclusions:

• AI-based clinical decision support systems enhance the detection of ELVO on NCCT scans.
• AI can facilitate acute stroke reperfusion therapy by improving patient triage, especially in centers without thrombectomy capabilities.