Highly automated segmentation of arterial and venous trees from three-dimensional magnetic resonance angiography (MRA)

Abstract

With the development of new magnetic resonance (MR) contrast agents that have longer persistence in the blood, contrast-enhanced magnetic resonance angiography (MRA) facilitates non-invasive imaging of the cardiovascular system at high resolution in large anatomic volumes. These high resolution 'steady state' images have simultaneous enhancement of both the artery and vein blood pools. Consequently, separation of arteries and veins is an emerging challenge in MRA analysis. Because of the complexity of the vascular structure, manual approaches to cardiovascular tree analysis are impractical. A novel, highly-automated low extremity vessel segmentation and display methodology is reported that consists of five main steps: (1) Binary mask generation, (2) tree-structure generation, (3) optimal vessel path calculation, (4) vessel segment labeling, and (5) conflict resolution. The method's performance was tested in computer phantoms and the in vivo data sets. In the computer-generated phantoms, vessel volume errors ranged from 1.0 to 8.8%. In the in vivo data, the labeling errors ranged from 0.1 to 15.5%. The method provided high quality results in individual data sets and demonstrated segmentation robustness.