Automated Basilar Artery Lumen Segmentation for High Resolution in Black Blood MRI

The basilar artery (BA) arises from the convergence of intracranial vertebral arteries in close proximity to the medulla oblongata and pons. As the largest artery in the posterior circulation, BA is susceptible to atherosclerosis and inflammatory conditions, necessitating the assessment of lumen shape and wall thickness in basilar artery diseases. Our objective is to develop an automated image segmentation technique for detecting the lumen and wall boundaries in black blood MR vessel wall images of the BA. The method consists of two components: an initial centerline manual tracking for creating 2D cross-sectional images of BA, and a fully automated lumen segmentation deep learning model, utilizing Detectron2/Mask RCNN. The study dataset consisted of 26 MRI scans, 20 for training and 6 for test. Preliminary results demonstrate that thin vascular structures such as BA lumen and wall can be labeled effectively and accurately labeling structures through transfer learning. In this study we trained two models, arterial wall detector (AWD) and lumen detector (LD). Both models achieved high true positive detection rates for the arterial wall (95.7%), lumen (91.4%), and the combined AWD+LD (97.6%) on the 169 cross-sectional BA images of the 6 test datasets. The mean and standard of Hausdorff distances to the ground truth inner lumen wall were 1.19 +/- 1.55 and 0.55 +/- 0.78 mm by AWD and LD respectively.Clinical Relevance- The BA is the main artery provides the posterior blood supply for the brain, the proposed method can help physicians to evaluate the BA wall and lumen more accurately in the treatment of cerebrovascular diseases.