Related Experiment Video
Updated: Sep 9, 2026
Simulator Training for Endovascular Neurosurgery
Published on: May 6, 2020
Real-Time AI-Generated Guidewire-Out-of-View Notifications in Diagnostic Cerebral Angiography: Higher Observed Rates
Syun Takano1, Yoshikazu Matsuda2, Kenichi Kono3,4
1Department of Neurosurgery, Showa Medical University Koto Toyosu Hospital, Tokyo, Japan.
Purpose:
Evaluation of technical proficiency plays a critical role in ensuring the safety and efficacy of neuroendovascular procedures. However, objective and quantitative assessment methods that are suitable for clinical practice remain scarce or unavailable. To address this limitation, this study examined whether real-time notifications from intraoperative artificial intelligence (AI)-assisted systems can serve as objective indicators of procedural skills.
Methods:
We retrospectively analyzed 200 consecutive digital subtraction angiography (DSA) cases performed at our institution using a real-time AI-assisted system called neuro-vascular assist (iMed Technologies, Tokyo, Japan). The procedures were performed by either board-certified neuroendovascular specialists with extensive experience or noncertified trainees. We compared AI notifications and procedural parameters between the groups and evaluated the association between certification status and AI notifications using mixed-effects logistic regression with a random intercept for operator and adjustment for case-mix.
Results:
Univariate analysis revealed that trainees exhibited significantly higher frequency of AI notifications (62.5% vs 41.9%; P = 0.01) and longer fluoroscopy times (P = 0.02) compared to specialists. Furthermore, multivariable mixed-effects logistic regression accounting for case-mix and within-operator clustering showed twofold higher odds of an AI notification in trainees, but the association was not statistically significant (adjusted OR 2.10, 95% CI 0.68-6.49; P = 0.20).
Conclusion:
This study provides preliminary evidence that intraoperative AI notifications may reflect differences in technical behavior according to operator certification status. Although the adjusted association was not statistically significant, these findings suggest the potential of AI-assisted systems as proof-of-concept tools for evaluating technical proficiency. Further validation involving a larger number of operators is required.
