MRI is conventionally employed in neonatal brain diagnosis and research studies. However, the traditional segmentation protocols omit differentiation between heterogeneous white matter (WM) tissue zones that rapidly evolve and change during the early brain development. There is a reported correlations of characteristics of the transient WM compartments (including periventricular regions, subplate, etc.) with brain maturation [?,?] and neurodevelopment scores [?]. However, there are no currently available standards for parcellation of these regions in MRI scans. Therefore, in this work, we propose the first deep learning solution for automated 3D segmentation of periventricular WM (PWM) regions that would be the first step towards tissue-specific WM analysis. The implemented segmentation method based on UNETR [?] was then used for assessment of the differences between term and preterm cohorts (200 subjects) from the developing Human Connectome Project (dHCP) (dHCP) project [?] in terms of the ROI-specific volumetry and microstructural diffusion MRI indices.