Synthesizing contrast-enhancement map from non-contrasted black-blood images for brain lymphatic imaging using a deep neural network

Abstract

In this study, we proposed non-invasive brain lymphatic region mapping by synthesizing contrast-enhancement maps (CEM) from non-contrast enhanced black blood imaging (non-CEBB). T1 images were used as secondary input along with non-CEBB, which helped the network to better distinguish lymphatic regions from blood vessels. From the reconstructed 3D CEM segmentation, enhancement was mainly distributed in dorsal parasagittal dura, parasagittal regions, brain basal region and around choroid plexus, consistent with previous studies. This study could be applied to the segmentation of the brain lymphatic region with less ambiguity and may help automatic segmentation rather than intensity-based segmentation by adapting self-supervised learning.