Efficient Hybrid Registration Method Using a Shell Volume for PET and High Resolution MR Brain Image Fusion

Abstract

This paper presents an efficient hybrid registration method using a shell volume that consists of high contrast voxels for combining PET and high resolution MR (HR-MR) brain images. This approach automatically selects a brain shell volume from the PET image, and then transforms only the voxels in the brain shell volume into the coordinate space of HR-MR images. Based on the corresponding voxels in HR-MR images, it finally calculates the best-matching voxel positions using normalized mutual information (NMI). The shell volume reduces the computation time by using smaller number of corresponding voxels to be matched, and it even enables a more robust registration. Experimental results on clinical data sets showed that our method successfully aligned all PET and HR-MR image pairs without losing any diagnostic information, while the conventional NMI method failed to align some cases.