A technique for improving 2D X-ray to 3D MRI registration in fluoroscopy-guided orthopedic interventions

Introduction Orthopedic interventions are often performed using x-ray fluoroscopic guidance. Although boney structures can be seen on fluoroscopic images, soft tissue contrast in such images is poor. In addition, fluoroscopy can only provide 2-dimensional projection images, which has no spatial resolution along the projection axis. For these reasons it would be beneficial to use preoperative 3D data with adequate soft tissue contrast for real-time intra-procedural navigation. Three-dimensional or multi-slice MRI can provide such data. However, using 3D MRI data during the intervention requires registration of the data to intra-operative fluoroscopic images. Registering MRI to x-ray data is technically challenging because fundamental differences in physical imaging principles lead to very different tissue contrasts. For vascular interventions, the enhanced vasculature itself can be used for x-ray angiogram to MRA registration, but in orthopedic interventions there are no obvious landmarks or structures similar in appearance in both techniques that can be used. To circumvent this problem Van de Kraats et. al. [1] have investigated the possibility to define a mapping function from multispectral MRI signal intensity values to computed tomography (CT) Hounsfield units in order to create CT-like data from multiple MRI datasets. Such image data can be used to digitally reconstruct radiographs that would have a similar appearance as fluoroscopic X-ray images of the same object and this method would therefore improve registration. We have elaborated on this idea in order to make it more robust and applicable to scan sequences commonly used in clinical practice. The method we propose now is based upon a voxel classification strategy. Its feasibility is demonstrated in ex vivo animal models.