Artificial Intelligence-Based Computer Vision Feedback to Improve Clinical Performance in Surgical Technology Education

Advancements in artificial intelligence (AI) and computer vision offer new opportunities to improve clinical skills training. This study developed a real-time, AI-based computer vision system for detecting surgical instruments and evaluated its association with clinical performance among surgical technology students.

A quasi-experimental pre-test/post-test study with a control group was conducted among 48 eligible surgical technology students at Iran University of Medical Sciences during the 2025-2026 academic year. Participants who met the predefined inclusion criteria were enrolled using census sampling. Following ethics approval (IR.IUMS.REC.1403.928), the participants were allocated to an intervention group (n=31), which received immediate AI-generated visual feedback after each instrument arrangement task, and a control group (n=17), which received routine instructor-led training without AI feedback. A YOLOv8-based deep learning model was trained using transfer learning on annotated images of surgical table setups. Clinical performance was assessed using a validated observational checklist (CVI=0.90; Cronbach's α=0.87). Data were analyzed using analysis of covariance (ANCOVA), the Wilcoxon signed-rank test, Spearman's correlation, and the intraclass correlation coefficient (ICC) in SPSS version 27. All 48 participants completed the study (attrition rate=0%).

The AI model achieved excellent object detection performance (precision=0.974, recall=0.979, mAP@0.5=0.981, mAP@0.5:0.95=0.808), with an average inference time of 1.6 seconds per image. Clinical performance was significantly higher following AI-assisted feedback (Wilcoxon signed-rank test, Z=-4.879, p<0.001, r=0.88). After adjustment for baseline performance using ANCOVA, AI-assisted feedback was associated with higher post-intervention performance scores than conventional training (F (1,45)=19.51, p<0.001, partial η²=0.302). Excellent agreement was observed between AI-generated assessments and expert human ratings (ICC=0.984).

Real-time, AI-based computer vision feedback was associated with improved surgical instrument arrangement performance among surgical technology students while providing an objective and reliable assessment of practical skills. The proposed system represents a promising proof-of-concept educational tool for simulation-based training and competency assessment and may complement conventional instructional approaches in surgical technology education.