MRI-based elastic-mapping method for inter-subject comparison of brain FDG-PET images

Inter-subject anatomic differences prohibits direct image-wise statistical evaluation of brain FDG-PET images of Alzheimer's disease (AD) patients. In this study, the authors propose a MRI-based elastic-mapping method which enables image-wise evaluation. The method involves intra-subject MR-PET registration, 3-D elastic mapping of two set of MR images, and elastically transforming the co-registered PET images. The MR-PET registration used simulated PET images, which were based on segmentation of MR images. In the 3-D elastic mapping stage, first a global linear scaling was applied to compensate for brain size difference, then a deformation field was obtained by minimizing the regional sum of squared difference between the two sets of MR images. Two groups (AD patient and normal control), each with three subjects, were included in the current study. After processing, images from all subjects have similar shapes. Averaging the images across all subjects (either within the individual group or for both groups) give images indistinguishable from original single subject FDG images (i.e. without much spatial resolution loss), except with lower image noise level. The method is expected to allow statistical image-wise analysis to be performed across different subjects.