Robust rigid registration for non invasive Computer Assisted Orthopedic Surgery. Preliminary results

In order to automatically guide orthopedists during the interventions, computers need to localize and track bones to be operated. Localization may be done thanks to a calibrated ultrasound (US) probe for non invasive Computer Assisted Orthopedic Surgery (CAOS) instead of implanting trackers in the bone. In this approach, rigid registration is required for the transfer and update of the plan. In this paper, a new approach inspired by Robust Point Matching (RPM) [1] is proposed for rigid registration. It is able to handle outliers encountered when one dataset is a partial view of the other one or when the datasets locally overlap. The algorithm is tested on data acquired in the proximal part of femur phantoms by means of computed tomography (CT) and 3D US probe. The obtained results show that the proposed registration is robust to outliers and initial misalignment.