Automatic Segmentation of Coronary Artery Lumen via Anisotropic Graph-cuts.

Abstract

Coronary fractional flow reserve (FFR) is the ratio of distal to proximal pressure in a stenotic lesion and defines its presence of ischemia (i.e. FFR≤0.8). However, FFR requires pressure to be measured invasively across the lesion, which is barrier to wider use of the methods. Computational fluid dynamics (CFD) applied to computed tomography coronary angiography (CTCA) enable the computation of FFR (i.e. CT-FFR) noninvasively. The accurate coronary artery geometrical model is the prerequisite for CT-FFR. We aimed to develop a method to automatically segment the coronary artery lumen based on anisotrophic graph-cuts. 5 patients with coronary artery disease underwent CTCA. CTCA images were enhanced by Hessian Matrix filter and then to derive vesselness images and vessel directions. Centrelines were extracted using the shortest path algorithm. We modified the edge energy term based on the dissimilarity of neighboring voxels intensities. The inner product of vessel directions of two neighboring voxels was multiplied as well. All cases were automatically segmented successfully with average computational time of 79.5 seconds. Our anisotropic graph-cuts method reduced unnatural bumps on the lumen than classic graph-cuts methods. Our anisotrophic graph-cuts based method improved coronary lumen segmentation and downstream coronary geometry modelling, will play an important role for real-time CT-FFR computation in clinical practice.