Segmentation of Myelin-like Signals on Clinical MR Images for Age Estimation in Preterm Infants

Abstract

Myelination is considered to be an important developmental process during human brain maturation and closely correlated with gestational age. Assessment of the myelination status requires dedicated imaging, but the conventional T2-weighted scans routinely acquired during clinical imaging of neonates carry signatures that are thought to be associated with myelination. In this work, we propose a new segmentation method for myelin-like signals on T2-weighted magnetic resonance images that could be used to assess neonatal brain maturation in clinical practice. Firstly we define a segmentation protocol for myelin-like signals, and delineate manual annotations according to this protocol. We then develop an expectation-maximization framework through which we obtain the automatic segmentations of myelin-like signals. We incorporate an explicit class for partial volume voxels whose locations are configured in relation to the composing pure tissues via second-order Markov random fields. We conduct experiments in the thalami and brainstem where the majority of myelination occurs during the perinatal period for 16 test subjects aged between 29 and 44 gestational weeks. The proposed method performs accurately and robustly in both regions with respect to the manual annotations over a range of intensity percentile thresholds that are used to generate the initial segmentation estimates. Finally, we construct spatio-temporal growth models for myelin-like signals in the thalami and brainstem to demonstrate the applicability of the proposed method for age estimation in preterm infants.