Purpose Angiographic Imaging is used in minimally-invasive interventions to visualize vessels in order to plan (preoperatively) the operation and navigate (intraoperatively) a catheter to the region of interest. Registering 3D preoperative angiographic data and 2D intraoperative images would enable the physician to lay a detailed roadmap over the 2D angiogram avoiding too much injection of contrast agent and radiation exposure to the patient. A Computed Tomography Angiography (including early arterial phase) and fluoroscopic or digitally subtracted angiograms are the modalities used in chemoembolizations where cytostatica are applied locally and tumor-supporting blood vessels are blocked. Method Since vessel structures are not rigid in the abdominal area mainly due to respiratory movement an estimation of rotation and translation using intensity-based methods is not sufficient to accurately overlay 3D and 2D data. Thus, all information of the data sets are discarded except the structures of interest, the vessels, which are transformed to a 3D and a 2D graph respectively and matched considering topological and geometrical information about bifurcation points (nodes), and connectivity between vessel segments (edges). The 3D graph is obtained from a CTA volume scan in the early arterial phase with an isotropic voxel size of 0.6mm. A wave propagation approach is used to extract the topology of the vessel tree starting at the arteria hepatica propria and to assess its topology. Due to the simplicity of the 2D angiogram, the 2D graph is currently extracted interactively. The algorithm projects the 3D graph with the current pose and determines putative node correspondences using the graduated assignment algorithm for graph matching. Inherently, it determines the Euclidean distance between the putative matches and uses it as a cost function that is optimized over in an iterative process. Results In order to test the quality and accuracy of the registration approach, rigid vessel structures have been used to create a ground truth registration and simulate deformability. In a first step, the algorithm has been used to rigidly register two data sets and compare the result with traditional methods. A CT volume and a 2D angiogram (acquired with a Siemens Axiom Artis System) of a head phantom with rigid vessels have been registered using intensity-based registration (Abb1 shows an overall setup, Abb2 shows a checkerboard image of the registered data sets). A hardware-accelerated volume renderer has been utilized to create Digitally Reconstructed Radiographs that are compared with the 2D angiogram using the Normalized Cross Correlation similarity measure. A Best-Neighbor optimization technique iteratively determined the optimal pose. After creation of 3D and 2D graphs the above described method has been used to align the two data sets feature-based. The resulting 3 translation and 3 rotation parameters could be recovered using this approach (Abb3 shows an overall setup, Abb4 shows and overlay of projected graph (purple) and 2D graph (turquoise)). See the images on the curac website. Conclusion We have developed a feature-based registration algorithm to align 3D to 2D angiographic data sets. It has been shown that it yields the same results as traditional registration algorithms in the rigid case. Future work will model deformations on the vessel structures to align abdominal data sets subject to respiratory movement.
«
Login
BibTeX