Abstract Number: Esoc2026a1899 Performance Limitations of Rapid AI in Posterior Circulation Stroke Detection: A Single-Center Retrospective Analysis

Abstract Background and aims Posterior circulation strokes (PCS) account for up to one-quarter of ischemic strokes and often present with nonspecific symptoms, making diagnosis challenging. RAPID AI is an automated CT perfusion (CTP) interpretation platform validated mainly for anterior circulation ischemia. However, CTP has known technical limitations in posterior fossa imaging, and RAPID AI performance in detecting PCS remains uncertain. We evaluated the diagnostic performance of RAPID AI perfusion output in suspected PCS compared with brain MRI. Methods This retrospective single-center study included consecutive acute stroke patients who underwent CTP processed by RAPID AI. Of 577 CTP scans performed, 535 generated automated output and 218 patients underwent MRI. The final cohort included 71 patients with clinical or imaging suspicion of PCS and available MRI; patients with isolated anterior circulation infarction were excluded. RAPID results were categorized as true positive, false positive, or false negative using MRI as the reference standard. Because true negatives were not captured, only sensitivity and positive predictive value were calculated. Results MRI confirmed posterior circulation infarction in 32 of 71 patients. RAPID AI detected posterior perfusion abnormalities in 52 patients, of which 13 were true positive and 39 false positive, yielding a PPV of approximately 25%. Nineteen infarcts were not detected (sensitivity ~41%). Nearly three-quarters of RAPID-positive findings were not confirmed by MRI. Conclusions RAPID AI demonstrated modest sensitivity and low PPV for PCS detection. Automated perfusion output alone should not guide diagnosis or management. Comprehensive clinical assessment and multimodal imaging remain essential, and algorithm refinement is needed for posterior circulation Stroke Conflict of interest Amro Biadsee: nothing to disclose