3D Reconstruction of Patient Specific Bone Models for Image Guided Orthopaedic Surgery

Abstract

Visualisation of patient-specific fractured bone in 3D plays an important role in image guided orthopaedic surgery. Existing research often focuses on intra-operative registration of the patients' anatomy with pre-operatively obtained 3D volumetric data (e.g. CT scans) utilising fiduciary markers. This expensive and invasive approach is not routinely available for diagnostics, and a majority of fracture reduction procedures currently solely relies on 2D x-ray/fluoroscopic images. This research aims to assist orthopaedic surgeons in all steps of the femur fracture reduction procedure by introducing 3D anatomical visualisation. Studies conducted on femur fracture reduction has confirmed that computer-aided systems can significantly improve the accuracy of orthopaedic procedures by augmenting the current 2D image guidance with an interactive display of 3D bone models. This research indicates that the positioning errors, which generate bone misalignments and complications, will be reduced through the introduction of 3D bone fragment visualisation during surgical procedures. Consequently 3D visualisation of anatomy plays an important role in image guidedd orthopaedic surgery and most importantly contributes to minimally invasive procedures. The research goals of this thesis are achieved through the construction of a 3D model of a fractured bone, and the real-time tracking (pose estimation) of the bone segmentsintra operativelyly. The first component of the research is the innovative 3D reconstruction technique proposed for preoperative planning on procedures involving the femur, tibia and iliac. Pre-operative planning plays an essential role in the management of orthopaedic injuries because many of the technical problems that may arise during surgery can be anticipated during this preparatory phase. The novel reconstruction algorithm is based on two conventional orthogonal (in anterior and lateral views) 2D radiographic images and a 3D model of an intact (healthy) bone. This intact model is customised through a non-rigid registration process to the shape of the patient's bone. The customisation involves a fracture incorporation process that separates the bone into the proximal and distal segments and identifies the pose (position and orientation) of each fragment. This generally applicable framework is a significant contribution over current literature which is hindered by proprietary models that limit usage, are only available for small regions of the bone and have time consuming feature matching requirements. Furthermore, tests conducted involving cadaveric bone models conveyed a millimetre level accuracy in reconstruction, which is superior to comparable literature. The second component of the research is the intra-operative pose estimation conducted for cases involving bone segment motion tracking (e.g. femur shaft fracture reduction procedure). Here the pre-operatively reconstructed 3D model will be utilised intra-operatively in the 2D-3D registration process for real-time pose estimation. This novel intra-operative registration is performed solely utilising bony anatomical features extracted from fluoroscopic images. This contribution is an enhancement over the current literature which is a proponent of utilising invasive external fiduciary markers. Experiments conducted through phantom studies and cadaveric fractured bones identified a millimetre level accuracy in translation and less than a two degree level accuracy in rotation.