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

This retrospective, multi-reader, blinded, pivotal trial assessed the performance of artificial intelligence (AI)-based clinical decision support system used to improve the clinician detection of emergent large vessel occlusion (ELVO) using brain non-contrast computed tomography (NCCT) images. We enrolled 477 patients, of which 112 had anterior circulation ELVO, and 365 served as controls. First, patients were evaluated by the consensus of four clinicians without AI assistance through the identification of ELVO using NCCT images. After a 2-week washout period, the same investigators performed an AI-assisted evaluation. The primary and secondary endpoints in ELVO prediction between unassisted and assisted readings were sensitivity and specificity and AUROC and individual-level sensitivity and specificity, respectively. The standalone predictive ability of the AI system was also analyzed. The assisted evaluations resulted in higher sensitivity and specificity than the unassisted evaluations at 75.9% vs. 92.0% ( < 0.01) and 83.0% vs. 92.6% ( < 0.01) while also resulting in higher accuracy and AUROC at 81.3% vs. 92.5%, ( < 0.01) and 0.87 [95% CI: 0.84-0.90] vs. 0.95 [95% CI: 0.93-0.97] ( < 0.01). Furthermore, the AI system improved sensitivity and specificity for three and four readers, respectively, and had a standalone sensitivity of 88.4% (95% CI: 81.0-93.7) and a specificity of 91.2% (95% CI: 87.9-93.9). This study shows that an AI-based clinical decision support system can improve the clinical detection of ELVO using NCCT. Moreover, the AI system may facilitate acute stroke reperfusion therapy by assisting physicians in the initial triaging of patients, particularly in thrombectomy-incapable centers.