Rigid registration of MR and biplanar fluoroscopy

Abstract

This thesis discusses the the rigid registration of MR data and biplanar, fluoroscopic radiographs of the head, by means of an intensity-based, measure of similarity. We investigate a number of similarity measures which have been applied in other multimodal registration contexts, including that of CT-fluoroscopy registration. The similarity measures are evaluated by probing the space of transformations between the coordinate frames of an MR volume and biplanar, digitally-reconstructed radiographs (DRRs) produced from a corresponding CT volume. This method allowed us to know the "ground truth" of the registration, which could be established by proven methods for the 3D-3D CT-MR alignment. Furthermore, we propose a method of DRR production called voxel-projection which drastically reduces processing time relative even to optimized ray-casting methods. The computational efficiency of voxel-projection makes it a useful tool for the investigation of similarity measures in this 3D-2D context. Its speed may also enable 3D-2D registration methods which rely on evaluation of a similarity measure over the entirety of a DRR and model radiograph. Based on similarity measure characeristics observed in our probing experiments, and using the voxel-projection method, we adapted the uphill-simplex optimization algorithm to implement an intensity-based MR-fluoroscopy rigid registatration engine.