Automated PET Quantification of [18F]FDG PET Images for Neurodegenerative Disorders Research

Positron Emission Tomography (PET) and Magnetic Resonance Imaging (MRI) are important tools for the study of neurodegenerative diseases. Existing tools often perform preprocessing, registration and segmentation steps separately. In this work, we propose an integrated approach that combines these steps into a unified framework for the quantification of brain [F-18]FDG PET images. The proposed pipeline receives dynamic or static PET images and a high-resolution anatomical MRI image. The PET image is registered to the MRI and both are normalized into MNI152 space. Then, the uptake values of the PET image are normalized to the cerebellum uptake. Regional FDG uptake values are computed for each region of the Hammers' atlas. Finally, a quantitative synthetic image showing a hypometabolism/hypermetabolism map is generated using a FDG atlas of healthy individuals. The pipeline was evaluated with data from our imaging facilities and with a set of images from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. We successfully processed the local dataset and 29 of the 35 ADNI images. The 6 images that failed had problems in the spatial normalization to the MNI152 space. We found agreement in the metabolic patterns between the two AD dementia patients and, on the other hand, between the CN subject and the atlas. These results validated the pipeline and showed its consistency and robustness across different datasets. This automated pipeline provides quantitative metrics of regional [F-18]FDG uptake and hypometabolism maps, which can have multiple clinical and scientific applications.