Automatic tissue and structural segmentation of neonatal brain MRI using Expectation-Maximization

Accurate automated image segmentation in neonates is challenging due to the lower contrast-to-noise ratio compared to adult scans, the partial volume effect and large anatomical variation. In this paper, we present a technique for brain segmentation into different tissues and structures of interest. Atlas priors and subject-specific tissue priors are used to initialize an ExpectationMaximization (EM) scheme. The proposed implementation incorporates a Markov Random Field (MRF) regularization to account for the spatial dependency of labels, prior relaxation to adapt the priors according to the individual brain appearance and partial volume (PV) correction. The algorithm is evaluated against manually segmented data from the NeoBrainS12 MICCAI challenge.