Enhancing Intraoperative Safety Through an Intelligent Surgical Instrument Verification System Based on Computer Vision Technologies

BACKGROUND: In operating room management, surgical instrument counting is a critical process for ensuring patient safety and preventing intraoperative risks. Traditional manual counting takes 15-30 minutes and is prone to errors under high-pressure environments. With the increasing variety and quantity of instruments, identification difficulty and the risk of omissions during surgery are further heightened. METHODS: The proposed system is structured around four integrated modules: image acquisition and annotation, YOLOv11-based surgical instrument detection, PaddleOCR-based digital extraction of electronic scale readings, and an intelligent comparison and alert system. This study focuses on the systematic design, integration, and technical validation of a multimodal verification framework tailored to a high-risk orthopedic instrument counting scenario. Designed with patient safety as its core objective, the system operates in near real time to support instrument tracking and discrepancy checking under the experimental setting of this study. RESULTS: The results demonstrate that the system achieved high technical accuracy and processing efficiency under the study setting. Specifically, the YOLOv11 model achieved a precision of 0.91, recall of 0.88, F1-score of 0.89, and a mAP50 of 0.93 in surgical instrument identification. These metrics indicate that the proposed framework has strong technical feasibility for instrument counting verification in the target orthopedic scenario. Furthermore, the PaddleOCR component attained 99.7% accuracy in digit recognition with an average response time of under 0.5 seconds, demonstrating its capability for near-real-time digital extraction. Taken together, these findings suggest that the system has potential clinical relevance for perioperative instrument management and may support safer and more efficient verification processes. Future testing in real or simulated clinical settings will further evaluate its usability, user acceptance, and compatibility with real clinical workflow. CONCLUSIONS: This study established an intelligent surgical instrument verification prototype and demonstrated the feasibility of a multimodal safety-checking framework in a specific orthopedic instrument counting scenario. The main contributions of this study are threefold: (1) the development of a workflow prototype that integrates instrument detection, electronic scale reading extraction, and discrepancy alerting; (2) the use of weight-based cross-verification to complement visual recognition and improve checking reliability; and (3) the deployment of a modular system architecture with potential for future extension to perioperative safety management. Future studies incorporating user testing in real or simulated clinical settings will further evaluate system usability and workflow compatibility.