Machine Learning and Medical Engineering for Cardiovascular Health and Intravascular Imaging and Computer Assisted Stenting

Abstract

Time-of-flight (TOF) Magnetic Resonance Angiography (MRA) is a useful imaging technique which reflects blood flow and vasculature information. However, due to the low signal and contrast of arteries in TOF MRA, it is challenging to extract vascular features such as length, volume and tortuosity, through segmentation and tracing. Hence, in this paper, a simultaneous artery tracing and segmentation method is proposed to a generate quantitative intracranial vasculature map from TOF MRA. Instead of using original images, segmentation from a neural network model is used to initiate tracing, avoiding the low signal or contrast for small arteries. A tracing method is proposed based on cross-sectional best matching, followed by an optimization scheme from the multiplanar reformatted view. Centerline positions, lumen radii and centerline deviations are jointly optimized for robust tracing within artery regions. Finally, the refined artery traces are used for better artery segmentation. The method is validated on eight TOF MRAs of both healthy subjects and patients with cerebrovascular disease, showing good agreements with human supervised tracing and segmentation results for representative features such as artery length (<4% mean absolute difference), volume (>0.80 Dice), and tortuosity (<3% mean absolute difference). Our method out-performs three other popular tracing and segmentation methods by a large margin.