Preoperative 3D models used in surgical navigation bring complementary information on structures of interest during surgery. Such models can be reconstructed pre-operatively using different imaging modalities (CT, MRI, radiographs) that differ from that available intra-operatively (radiographs or video). Registration of pre-operative 3D models with 2D intra-operative images is challenging due to the spine shape variation of different patient positioning between the image acquisitions. As a first attempt to address this technological problem, the objective is to propose a 2D/3D registration technique based on a 3D MRI model and only one per-operative calibrated radiograph.
The registration was performed using an optimization technique, which minimized the retroprojection errors of 14 landmarks on each vertebra. Additional constraints were incorporated based on rigid intervertebral transformations. The technique was tested on 3 scoliotic cases with their pre-operative 3D models from supine MRI and their pre-operative standing 2D posteroanterior radiographs. A mean 3D difference of 3.8mm was found between the positions of the landmarks for all vertebrae of the 3 cases. Part of the difference could be explained by the elapsed time between the image acquisitions up to 1 year and the manual identification of landmarks. The preliminary results show the potential of the technique to update the pre-operative MRI model using only one radiograph to compensate the difference in posture between the two acquisitions. Further work is under way to improve the precision by minimizing landmarks error identification avoiding manual identification by using semi-automatic detection of vertebral contours.