Background: To evaluate the feasibility of non-invasive fractional flow reserve (FFR) estimation using histologically-validated assessment of plaque morphology on coronary CTA (CCTA) as inputs to a predictive model further validated against invasive FFR.
Methods: Patients (n = 113, 59 ± 8.9 years, 77% male) with suspected coronary artery disease (CAD) who had undergone CCTA and invasive FFR between August 2013 and May 2018 were included. Commercially available software was used to extract quantitative plaque morphology inclusive of both vessel structure and composition. The extracted plaque morphology was then fed as inputs to an optimized artificial neural network to predict lesion-specific ischemia/hemodynamically significant CAD with performance validated by invasive FFR.
Results: A total of 122 lesions were considered, 59 (48%) had low FFR values. Plaque morphology-based FFR assessment achieved an area under the curve, sensitivity and specificity of 0.94, 0.90 and 0.81, respectively, versus 0.71, 0.71, and 0.50, respectively, for an optimized threshold applied to degree of stenosis. The optimized ridge regression model for continuous value estimation of FFR achieved a cross-correlation coefficient of 0.56 and regression slope of 0.59 using cross validation, versus 0.18 and 0.10 for an optimized threshold applied to degree of stenosis.
Conclusions: Our results show that non-invasive plaque morphology-based FFR assessment may be used to predict lesion-specific ischemia resulting in hemodynamically significant CAD. This substantially outperforms degree of stenosis interpretation and has a comparable level of sensitivity and specificity relative to publicly reported results from computational fluid dynamics-based approaches.
Keywords: Computational fractional flow reserve; Coronary artery disease; Coronary computed tomography angiography; Invasive coronary angiography; Plaque morphology.
Copyright © 2021. Published by Elsevier B.V.