Markerless Navigation in Computer Assisted Orthopaedic Surgery Using RGB-D Data: a Phantom-Based Comparative Study of...

Computer Assisted Orthopedic Surgery (CAOS) plays a major role to improve surgical outcomes. Existing navigation rely on optical markers that are fixed to the bones and surgical instruments and tracked with an optical camera in order to provide, during the intervention, an optimal and real-time assistance to the surgeon. However, these solutions have, still today, several drawbacks that limit their wide adoption, in particular, the need to use markers. Based on RGB-D data, we propose to evaluate two algorithms allowing the 3D localization without markers of surgical objects of interest for orthopedics such as bony structures or instruments. The performance of a recent deep-learning (DL)-based algorithm is therefore compared to the Point-Pair Features (PPF) algorithm using a surgical-based database that has been created for this purpose. The global mean error for the DL-based algorithm was 1.28 mm in translation and 1.54° in rotation, surpassing the PPF algorithm. Despite errors need to be reduced, DL-based algorithms have shown promising results.Clinical relevance- Reducing the number of markers for surgical navigation by exploiting RGB-D data would be beneficial as it could reduce the duration of the operation and could make it less invasive. Latest DL-based models to estimate the pose of bony structures and surgical instruments have yielded promising results paving the way toward markerless navigation approaches.