A Hybrid Deep Learning and Graphics method for Vessel Centerline Extraction and Vessel Segment Labeling to Assist Plaque Detection and Evaluation

Motivation: Manual centerline Extraction based on black-blood Magnetic resonance vessel wall imaging is a difficult and time-consuming but important step for further analysis of plaques. Goal(s): To propose a method for quickly, automatically and accurately extracting the centerline and label the segments of the target arteries. Approach: This study proposes a method that combines deep learning and traditional graphics method for the automatic and accurate centerline extraction and even segments labeling, which is applicable to flexible MR sequences (only black-blood images, only bright-blood images, or both). Results: Compared with the ground truth, the proposed method achieved excellent completeness and accuracy in a short time. Impact: This study introduces a flexible (MRVWI or/and TOF-MRI), swift, and precise approach for extracting and labeling the centerline of target vessel. With this method, radiologists can more conveniently and efficiently observe potential abnormal areas around vessel and diagnose vessel-related diseases.