Clinical Implementation and Performance of Real-Time AI System for Device Tracking in Neurointervention: First U.S. Evaluation of Neuro-Vascular Assist.

Abstract

Purpose: To evaluate the first U.S.-based implementation, notification accuracy, and temporal operator response of Neuro-Vascular Assist.

Methods: In a single-center study, 21 consecutive neurointerventional cases were performed using the AI system integrated with a biplane angiographic system. Procedures were classified as "Standard", "Coil", "Filter", or "Venous". Neuro-Vascular Assist tracked devices including guidewires, stents, coils, and filters in real-time. Visual and audio notifications were made based on device entry and exit in a determined region of interest. We classified these notifications as true positives (TP), false positives (FP), and false negatives (FN), and further assessed for temporal operator response (TOR).

Results: The AI system was successfully integrated into the workflow without procedural delays or system failures affecting notification delivery. A total of 232 TPs, 13 FPs, and 17 FNs were recorded, resulting in an overall precision of 94.7% and recall of 93.2%. Among the 59 TP notifications evaluated for TOR, 25 (42.4%) were followed by an observable operator response within 10 s. These 25 positive TOR events represented 10.2% of all 245 notifications.

Conclusion: Neuro-Vascular Assist was successfully implemented and demonstrated preliminary accuracy and feasibility in a U.S.-based neurointerventional setting. Larger prospective studies are warranted to determine whether real-time AI-assisted device tracking influences procedural decision-making, workflow, safety, or clinical outcomes.