Template-based 3D-2D rigid registration of vascular structures in frequency domain from a single view

Official URL: http://risc01.sabanciuniv.edu/record=b1606643 (Table of Contents)

Image guided interventions in angiography are performed with a real-time X-ray sequences acquired by a C-arm device which provides the surgeon two dimensional visualization needed to guide the surgical instruments. This visualization may be augmented by registering a three dimensional preoperative volume with the interventional images to provide additional information such as depth, removal of occlusions and alternative views of vessel paths. This thesis presents two novel methods for rigid registration of vascular structures in the preoperative volume to the interventional X-ray image for enhancing visualization in Image Guided Interventions. In the first part of this thesis, estimation of rotation and translation are decoupled. Rotation is estimated by comparing rotated projections of the segmented vessels of the volume with segmented X-ray vessels in frequency domain. Translation is then estimated by minimizing the distances and maximizing the overlap ratio between segmented vessels. The registration results are reported in mean Projection Distances. The second part of the thesis adds separation of out-of-plane translation estimation to the first part and replaces segmentation by gradients. Rotation and out-of-plane translation are estimated by comparing rotational projected templates of volume with depth templates formed by scaling the X-ray image in the Fourier Magnitude Domain. The in-plane translation is then estimated by a Fourier Phase correlation. The registration results are evaluated by a Gold Standard dataset on cerebral arteries. This method is robust against occlusions and noises due to its usage of gradients and frequency domain similarity, has high capture range and fast, fixed computation times for every step due to template based framework.