Learning large number of local statistical models via variational Bayesian inference for brain voxel classification in magnetic resonance images

As an essential step in brain studies, measuring the distribution of major brain tissues, including gray matter, white matter and cerebrospinal fluid (CSF), using magnetic resonance imaging (MRI) has attracted extensive research efforts over the past years. Many brain tissue differentiation methods resulted from these efforts are based on the finite statistical mixture model, which however, in spite of its computational efficiency, is not strictly followed due to the intrinsically limited quality of MRI data and may lead to less accurate results. In this paper, a novel large-scale variational Bayesian inference (LS-VBI) learning algorithm is proposed for automated brain MRI voxels classification. To cope with the complexity and dynamic nature of MRI data, this algorithm uses a large number of local statistical models, in each of which all statistical parameters are assumed to be random variables sampled from conjugate prior distributions. Those models are learned using variational Bayesian inference and combined to predict the class label of each brain voxel. This algorithm has been evaluated against several state-of-the-art brain tissue segmentation methods on both synthetic and clinical brain MRI data sets. Our results show that the proposed algorithm can classify brain voxels more effectively and provide more precise distribution of major brain tissues.