Abstract 940: AI-Assisted Error Recognition During Neurointerventional Training: A Systematic Review And Meta-Analysis
Stroke: Vascular and Interventional Neurology · 18.10.2026 · 6 (Suppl_1)
DOI: 10.1161/svi270000_940 ↗Kurz gefasst
Systematische Übersicht in der Zeitschrift Stroke: Vascular and Interventional Neurology, erschienen am 18.10.2026. Verfasst von J. Kahen.
Abstract (Originalsprache)
Introduction/Purpose: Neurointerventional procedures demand high technical precision, yet trainee error rates remain elevated during the early learning curve. Artificial intelligence (AI) systems capable of real-time error recognition, anomaly detection, and objective performance feedback have emerged as potential training adjuncts. We conducted a systematic review and meta-analysis to evaluate the accuracy, impact on skill acquisition, and educational utility of AI-assisted error recognition tools in neurointerventional simulation and training. Materials and Methods: We searched PubMed, Embase, IEEE Xplore, Scopus, and Web of Science from inception through July 2026 using terms related to artificial intelligence, machine learning, deep learning, computer vision, error detection/recognition, anomaly detection, neurointervention, endovascular, catheter navigation, simulation, and training. Eligible studies assessed AI systems designed to identify technical errors, complications (e.g., vessel perforation, device malposition), or performance deviations during neurointerventional simulation or procedural training. Two independent reviewers performed screening, data extraction, and risk-of-bias assessment. Primary outcomes included detection sensitivity/specificity or AUC, reduction in procedural errors, improvement in technical metrics (path length, fluoroscopy time, procedure time), and trainee performance versus traditional feedback. Random-effects meta-analysis was performed when comparable continuous or diagnostic accuracy data were available; otherwise results were synthesized narratively.
Results
Twenty-four studies met inclusion criteria. Most employed convolutional neural networks, geometric deep learning, or reinforcement-learning models for real-time catheter/guidewire tracking, perforation detection, endoleak identification, or motion-based performance scoring on fluoroscopy or simulator data. Pooled diagnostic performance for error/complication detection showed high sensitivity (range 73-93%) and AUC values typically exceeding 0.85. AI feedback consistently reduced technical errors and improved objective metrics (shorter path lengths, lower fluoroscopy and procedure times) compared with traditional instructor-only or no-feedback conditions. Several studies demonstrated accelerated learning curves and superior skill retention with AI-augmented training. Real-time systems outperformed post-hoc analysis for immediate error correction. Heterogeneity in simulator platforms, error definitions, and outcome measures limited full quantitative synthesis for some endpoints. Risk of bias was moderate in the majority of studies, largely due to small sample sizes and limited external validation.
Conclusion
AI-assisted error recognition systems demonstrate strong diagnostic accuracy and measurable educational benefit in neurointerventional training environments. These tools provide objective, scalable, and real-time feedback that complements traditional apprenticeship models and may accelerate competency acquisition while reducing trainee error rates. Future priorities include multicenter external validation, standardized error taxonomies, integration with high-fidelity simulators, and prospective studies linking AI-enhanced training to clinical procedural safety outcomes.
Zitieren
APA
Kahen, J. (2026). Abstract 940: AI-Assisted Error Recognition During Neurointerventional Training: A Systematic Review And Meta-Analysis. Stroke: Vascular and Interventional Neurology, 6(Suppl_1). https://doi.org/10.1161/svi270000_940
Vancouver
Kahen J. Abstract 940: AI-Assisted Error Recognition During Neurointerventional Training: A Systematic Review And Meta-Analysis. Stroke: Vascular and Interventional Neurology. 2026;6(Suppl_1). doi:10.1161/svi270000_940
BibTeX
@article{Kahen2026,
author = {Kahen, J.},
title = {Abstract 940: AI-Assisted Error Recognition During Neurointerventional Training: A Systematic Review And Meta-Analysis},
journal = {Stroke: Vascular and Interventional Neurology},
year = {2026},
volume = {6},
number = {Suppl_1},
doi = {10.1161/svi270000_940}
}
EUVIMED