Artificial intelligence–enhanced microsurgical training: a systematic review

Abstract

Artificial intelligence (AI) offers objective, adaptive tools for skill enhancement in microsurgical training, but evidence is fragmented. This systematic review evaluates AI-enhanced training efficacy compared to traditional methods, focusing on technical performance, learning efficiency, and skill retention. Following PRISMA guidelines, databases (MEDLINE, Embase, Cochrane, IEEE Xplore, Web of Science) were searched from January 2010. Data on study characteristics, AI models, outcomes (time, errors, skill metrics), risk of bias, evidence certainty (GRADE), methodological quality, and reporting quality were extracted and synthesized narratively. From 2,056 records, 13 studies were included, involving 3-50 participants, mostly single-centre with varied designs. AI/ML models, such as Mask R-CNN, YOLOv2, ResNet-50, and other convolutional neural networks, were primarily used for assessment or guidance/coaching, focusing on instrument tracking (30.8%), motion analysis (23.1%), tutoring/guidance (15.4% each). Median accuracy 83.8% (IQR 78.4-88.2%). AI improved technical skills (reduced errors) and learning curves via real-time feedback, with promising retention outcomes. RoB high; evidence certainty very low. Reporting quality high/moderate, external validation poor. AI enhances microsurgical training with objective metrics and personalised feedback, showing promising technical advantages in simulations. However, heterogeneous, low-quality evidence limits generalisability. Future research needs multi-centre RCTs, standardised outcomes, external validation, and ethical considerations for clinical translation.

Ähnliche Arbeiten