Auto-Registration of Cardiac PET/CT Images with a 3D Weighted Gradient Correlation Algorithm
H Ai1,2*, T Pan1, (1) The University of Texas M.D.Anderson Cancer Center, Houston, TX, (2) The University of Texas Graduate School of Biomedical Science at Houston, TXSU-D-217A-5 Sunday 2:15:00 PM - 3:00:00 PM Room: 217A
Purpose: To design a novel 3D automatic registration algorithm for cardiac PET/CT registration using gradient information and to evaluate the performance of the algorithm with clinical PET/CT datasets.
Methods: The 512x512x47 CT images are at first resized to 128x128x47 to match the matrix size of PET. In order to maximize the gradient information at boundaries in CT, a conventional fuzzy c-means clustering algorithm (number of cluster = 7) is implemented to suppress signals from tissues that do not contribute much useful information for the registration purpose (fat, lung, and bones). The 3D Image gradient map, consisting of all three orthogonal components, is derived from the PET images and the post-clustering CT images. The mis-registration is modeled as 3D rigid body translation in this study, though it can be extended to include rotations as well. The details of the gradient-based objective function are described in the support document. Optimal registration is determined by searching for the maximum of the objective function over a range of potential translation positions (e.g. 8.2x8.2x2.8cm). This process is repeated at a higher matrix size (512x512x47) to refine the result of registration.
Results: We applied this auto registration technique on 55 patient data sets of cardiac PET/CT images. The CT images were average-CT images, and the PET images were without attenuation correction. 54 out of the 55 cases produced satisfactory registration.
Conclusions: The proposed weighted gradient correlation algorithm is a viable solution for auto registration of cardiac PET/CT images. More work is needed to further improve the robustness of the algorithm.