Multi-class brain segmentation using atlas propagation and EM-based refinement

In recent years, multi-atlas segmentation has emerged as one of the most accurate techniques for the segmentation of brain magnetic resonance (MR) images, especially when combined with intensity-based refinement techniques such as graph-cut or expectation-maximization (EM) optimization. However, most of the work so far has focused on intensity-based refinement strategies for individual anatomical structures such as the hippocampus. In this work we extend a previously proposed framework for labeling whole brain scans by incorporating a global and stationary Markov random field that ensures the consistency of the neighbourhood relations between structures with an a-priori defined model. In particular we improve the segmentation result of a locally weighted multi-atlas fusion method for 41 different structures simultaneously by applying a subsequent EM optimization step. We evaluate the proposed approach on 30 manually annotated brain MR images and observe an improvement of label overlaps to a manual reference by up to 6%. We also achieved a considerably improved group separation when the proposed segmentation framework is applied to a volumetric analysis of 404 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort.