Artificial Intelligence–Driven Personalized Learning Improves Operating Room Instrument Training: A Prospective Observational Study

Summary

This prospective observational study demonstrated that an artificial intelligence-powered personalized learning system significantly improved long-term operating room instrument competency, reduced safety incidents, and enhanced training efficiency compared to traditional instruction.

Abstract

Traditional operating room instrument training often relies on one-size-fits-all teaching, produces poor long-term skill retention, and contributes to preventable surgical errors. The present study hypothesized that an artificial intelligence (AI) powered personalized learning system could improve technical competency, patient safety, and training efficiency by tailoring practice to individual learner profiles. An AI-powered personalized learning system (APLS) was developed that combines data-driven learner phenotyping, deep-learning-based instrument recognition, ensemble prediction of competency trajectories, and reinforcement learning to deliver adaptive multimodal feedback during simulation-based training. In a prospective observational study with cluster-based departmental allocation (introducing quasi-experimental elements) at a tertiary hospital, 107 multidisciplinary operating room staff completed either the AI-driven curriculum or standard instructor-led training. The primary outcome was 12-month retention of instrument-handling competency, measured by the perioperative instrument proficiency scale (PIPS); secondary outcomes included operating room safety incidents, training time, and cost per competent professional. Compared with traditional training, the personalized system was associated with substantially higher 12-month competency scores (APLS 84.1 ± 9.2 vs. Control 58.9 ± 18.7, p < 0.001), fewer safety-related events in clinical practice, and nearly half the training time while reducing overall training costs by 38.3%. The AI ensemble also outperformed its individual machine-learning components when predicting learner performance and selecting feedback strategies. These findings suggest that integrating unsupervised phenotype discovery, supervised prediction, and reinforcement learning into a unified platform can meaningfully enhance operating room instrument training and may offer a scalable framework for data-driven workforce development in perioperative care.

Introduction

The operating room environment represents one of healthcare's most complex settings, in which effective multidisciplinary collaboration directly impacts patient safety outcomes and surgical quality^1,2. Inadequate instrument competency training contributes to an estimated 15,000 preventable surgical errors annually in Chinese hospitals alone, with communication failures and technical deficiencies accounting for 72% of perioperative adverse events³. Traditional training approaches rely on standardized methodologies that fail to account for individual learning differences, cognitive pr....

Protocol

This study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Sir Run Run Shaw Hospital, Zhejiang University School of Medicine (Approval Number:0480, approved on August 10, 2025). All participants provided written informed consent before participation.

Study design and setting
A prospective observational study with cluster-based departmental allocation was conducted from January 2024 to February 2025 at Sir Run Run Shaw Hospital, Zhejiang University School of Medicine, Hangzhou, China. The department-based allocation strategy introduces quasi-experimental elements to ....

Results

Participant characteristics
The final sample included 107 participants with excellent retention (Figure 1). Attrition analysis comparing completers (n = 107) with withdrawals (n = 13) revealed no significant differences in baseline characteristics: age (t = 0.89, p = 0.38), gender (χ² = 0.21, p = 0.64), professional role (χ² = 1.45, p = 0.48), baseline PIPS scores (t = 0.62, p = 0.54), or group assignment (χ² = 0.03, p = 0.86), supporting.......

Discussion

This prospective observational study, incorporating quasi-experimental elements through cluster-based departmental allocation, provides evidence that an AI-powered personalized learning system integrating multiple machine learning algorithms is associated with substantial improvements in operating room instrument training outcomes compared to traditional pedagogical approaches. The findings have implications for both artificial intelligence methodology and surgical education practice; however, several important caveats t.......

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Haptic Feedback Surgical Simulator (Touch X)	3D Systems (formerly Sensable)	PHANToM Premium 3.0	Six-degree-of-freedom haptic device; used as primary interface for AI-driven force feedback delivery during simulated tissue manipulation and suturing tasks
Simulation Software Platform (SimSurgery™ AI Suite)	Custom / In-house developed	N/A	Proprietary AI-driven training software integrating convolutional neural network (CNN) modules for real-time performance classification and adaptive feedback generation
Force/Torque Sensor	ATI Industrial Automation	Nano17 SI-12-0.12	Measures applied instrument forces (resolution: 0.003 N) during tissue handling and suturing exercises; data fed into the AI classification algorithm
Motion Tracking System	Northern Digital Inc. (NDI)	Polaris Vega ST	Optical tracking of surgical instrument trajectories; 0.12 mm RMS volumetric accuracy; used for instrument navigation accuracy assessment
Laparoscopic Instrument Set (Training)	Karl Storz SE & Co. KG	26003 AA / 33310 DB	Standard 5 mm laparoscopic grasper and needle driver used in both AI-driven and conventional training arms
High-Fidelity Tissue Phantom (Soft Tissue Model)	SynDaver Labs	Surgical Abdominal Model (SKU: SYN-ABD)	Validated synthetic tissue surrogate for suturing, dissection, and tissue handling exercises; replaced every 50 uses per manufacturer guidance
GPU Workstation	NVIDIA / Dell	Dell Precision 7920 with NVIDIA A6000 (48 GB VRAM)	Computational hardware for running real-time AI inference (CNN classification latency < 15 ms); dedicated server connected to haptic device
Neural Network Framework	Open-source (Meta AI)	PyTorch v2.1.0	Deep learning framework used for model training, validation, and real-time inference of the adaptive feedback algorithm
Statistical Analysis Software	IBM Corp.	IBM SPSS Statistics v29.0	Used for mixed-effects modeling, independent samples t-tests, and repeated-measures ANOVA in all primary and secondary outcome analyses
Data Visualization and Graphing Software	GraphPad Software	GraphPad Prism v10.0	Used for generating all manuscript figures, including error bars (SD and 95% CI), retention curves, and grouped bar charts
Video Recording System	Stryker Corp.	1688 AIM 4K Platform	Intra-operative and simulation session video capture for blinded post-hoc expert assessment (OSATS scoring)
Structured Assessment Tool (OSATS)	N/A — Validated published instrument	Martin et al., 1997 (Br J Surg)	Objective Structured Assessment of Technical Skills; seven-item global rating scale (1–5 per item) used for blinded expert evaluation of suturing and tissue handling
Randomization and Data Management Software	Open-source	REDCap v13.7.2	Electronic data capture for participant allocation (cluster-based departmental), demographic data collection, and longitudinal follow-up score tracking
Audio Feedback Module	Custom / In-house developed	N/A	Software module delivering real-time corrective auditory cues (tone-coded alerts) integrated into the multimodal feedback system
Institutional Questionnaire (Learner Satisfaction Survey)	Custom / In-house developed	N/A	18-item Likert-scale (1–5) questionnaire assessing perceived training effectiveness, system usability, and learner confidence; administered post-training and at 24-week follow-up