Brain tissue segmentation in PET-CT images using probabilistic atlas and variational Bayes inference

PET-CT provides aligned anatomical (CT) and functional (PET) images in a single scan, and has the potential to improve brain PET image segmentation, which can in turn improve quantitative clinical analyses. We propose a statistical segmentation algorithm that incorporates the prior anatomical knowledge represented by probabilistic brain atlas into the variational Bayes inference to delineate gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF) in brain PET-CT images. Our approach adds an additional novel aspect by allowing voxels to have variable and adaptive prior probabilities of belonging to each class. We compared our algorithm to the segmentation approaches implemented in the expectation maximization segmentation (EMS) and statistical parametric mapping (SPM8) packages in 26 clinical cases. The results show that our algorithm improves the accuracy of brain PET-CT image segmentation.