Design and Development of a Surgical Instrument Detection System using Faster R-CNN

Abstract

This study aims to design and develop an AI-based tool for detecting and counting surgical instruments to reduce human counting errors and improve accuracy in medical workflows. The target instruments include surgical scissors, scalpel handles, and forceps, and the proposed system uses a Faster R-CNN object detection model implemented in MATLAB as the core method. The research process includes preparing an image dataset, labeling the instruments, splitting the data into training and testing sets, and training the Faster R-CNN model so that the system can automatically localize, classify, and count instruments. The system is then evaluated under different conditions, including varying the number and positions of instruments in each image and changing background characteristics, to assess performance as the environment becomes more complex. Experimental results show that the system can effectively detect and classify surgical instruments, particularly with a white background, which yields a high accuracy rate of 95%. However, performance tends to decrease when more instruments are present in the image or when the background is changed to green, especially in terms of detection completeness and counting accuracy, reflecting the model’s limitations under more challenging visual conditions. Overall, the developed system demonstrates potential for practical use in controlled environments and provides a foundation for the future development of intelligent medical systems.