World's first real-time artificial intelligence-assisted mechanical thrombectomy for acute ischemic stroke.

Abstract

Several recent reports have discussed real-time artificial intelligence (AI) assistance in elective neuroendovascular procedures such as carotid artery stenting, coil embolization for cerebral aneurysms, and liquid embolization. We present our preliminary experience with real-time AI-assisted mechanical thrombectomy (MT) for acute ischemic stroke (AIS) under local anesthesia. A real-time deep learning-based AI software (Neuro-Vascular Assist, iMed technologies, Tokyo, Japan) was used during MT for 16 consecutive patients with AIS. The AI system provided real-time notifications when the guiding catheter (GC) moved out of the fluoroscopic image. The efficacy, accuracy, and safety of the notifications were evaluated using video recordings. The AI system functioned properly in all cases. The mean number of notifications per case for the GC moving out of view was 8.1. The overall precision and recall of AI notifications were 97% and 99%, respectively. Of the 126 true-positive AI notifications, 25 (20%) prompted operators to reposition the GC within 10 s when it became unobservable in both frontal and lateral imaging planes. This response indicates the potential effectiveness of the AI system. No adverse events or delays in procedures due to the AI system occurred. In this preliminary study, the software was sufficiently accurate and safe for MT procedures, suggesting its usefulness. To the best of our knowledge, this is one of the first studies to report using a real-time AI system for MT in patients with AIS as an emergent procedure. Large-scale studies are warranted to validate its impact on procedural workflow and clinical outcomes.ABBREVIATIONS: AI = artificial Intelligence; MT = mechanical thrombectomy; AIS = acute ischemic stroke; GC = guiding catheter; TP = true positive; FP = false positive; FN = false negative; CAS = carotid artery stenting.