Segmentation with varying contrasts of pediatric MRI

Abstract To better understand early brain development, accurate segmentation of infant brain MR images into regions of interest, such as white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF), is one of the most important steps. In the first postnatal year, brain development can be divided into three distinct phases due to myelination and maturation process, i.e., infantile phase (≤ 5 months), isointense phase (6–8 months), and early adult-like phase (≥ 9 months). Among these three phases, the isointense phase exhibits extremely low tissue contrast in both T1-weighted and T2-weighted images, making tissue segmentation extremely challenging. To address this issue, inspired by prior domain knowledge, we developed an anatomy-guided joint tissue segmentation and topological correction framework for 6-month infant brain segmentation, which was implemented with random forest and deep neural network technologies. Experiments showed the advantages of the proposed framework and also demonstrated that anatomical guidance is a key direction to explore for 6-month infant brain segmentation. Our method is available in iBEAT V2.0 Cloud ( http://www.ibeat.cloud ).