Development of a Highlighting System for Surgical Instruments in Total Knee Arthroplasty

Abstract

Currently, there is a shortage of nurses worldwide. In total knee arthroplasty, many surgical instruments of similar shapes and sizes are available. As a result, medical accidents have occurred due to errors in the selection of these instruments. To solve these problems, several surgical instrument detection systems using object detection methods have been developed. However, these studies do not mention how to display the detection results. It is inefficient for scrub nurses to look at the display to check the detection results, and mistakes can occur. Therefore, in this study, we have developed a system that detects surgical instruments and uses the detection results to highlight the instruments using a projector like projection mapping. This study also describes a method for creating training data and an inference method to reduce the additional training time required when the number of detection targets is increased. In the experiments, we verify the effectiveness of the calibration system for the camera and projector. In addition, object detection is performed on a surgical instrument to verify the accuracy of the system. The results showed that the error after calibration is 0.15 cm. The IoU and F-measure of object detection are 0.911 and 0.958, respectively, indicating the effectiveness of the system.

Keywords

• Machine learning
• medical robotics
• object detection