Personalized coronary blood flow model based on CT perfusion to non-invasively calculate fractional flow reserve

Abstract

Coronary outlet resistance, influenced by quantification and distribution of total hyperemic coronary blood flow, is critical for accurately estimating the computation of fractional flow reserve from computed tomography angiography (FFRCT). However, accurate estimation of FFRCT is still a challenging task due to the impractical of non-invasively measuring coronary outlet resistance in the clinic. While traditional methods have been used to estimate outlet resistance through population-averaged physiological assumption models, individual differences between patients limit the reliability of the models. In this work, we propose a personalized coronary flow model based on dynamic stress computed tomographic perfusion (CTP), called CTPV, to achieve a noninvasive estimation of the outlet resistance and FFRCT. The CTPV model employs myocardial blood flow (MBF) obtained from CTP to quantify the total hyperemic coronary blood flow directly. Then, myocardial perfusion territories corresponding to coronary artery branches are proposed to distribute the outlet coronary blood flow. For this purpose, we use the Voronoi algorithm to segment the myocardial perfusion territory. Finally, the outlet resistance and FFRCT are estimated based on the outlet coronary blood flow. Experiment studies on a clinical dataset consisting of 138 vessels from 86 subjects show the superiority of our CTPV model against the CTPD and LVMD models, including the method of quantifying total hyperemic coronary blood flow based on left ventricular myocardial mass (LVM) and distributing the outlet coronary blood flow based on outlet diameter. The CTPV model estimates a higher AUC for FFRCT (AUC = 0.96) than the CTPD model (AUC = 0.93) and the LVMD model (AUC = 0.92). The accuracies of CTPV, CTPD, and LVMD models are 90.6%, 87.7%, and 84.1%, respectively. Additionally, we use different outlet truncation locations to evaluate the repeatability of the CTPV and CTPD models for estimating FFRCT on a random dataset of 10 vessels. The result shows that the repeatability of the CTPV model is better than the CTPD model. These compelling results evince that the proposed CTPV model can accurately estimate the coronary outlet resistance and improve the diagnostic performance and repeatability of FFRCT.