Investigations on 3D reconstruction of bones in surgery using a handheld trinocular camera system

Abstract

Knee arthroplasty benefits significantly from computer-assisted navigation, which improves the accuracy of prosthesis placement. However, current methods require invasive optical locators to track the position of the knee, which carries risks such as infection and prolonged healing times. To address these limitations, this work uses markerless trinocular SLAM to achieve accurate 3D reconstruction of the knee during surgery. The approach integrates SuperGlue for robust feature matching and incorporates segmentation to mask the knee, improving reconstruction accuracy despite challenges such as low-texture surfaces, reflections and spotlight illumination. The accuracy of the handheld trinocular camera system is evaluated under dynamic conditions, simulating camera movement during surgery to ensure accurate reconstruction during real-time surgery. In addition, a robot-guided dataset will be used to assess the repeatability and robustness of the SLAM approach. This research focuses on positional accuracy in motion and aims to advance real-time, non-invasive navigation solutions for knee arthroplasty, contributing to safer and more efficient surgical outcomes.