The influence of image quality on diagnostic performance of a machine learning-based fractional flow reserve derived from coronary CT angiography.

Abstract

To investigate the effect of image quality of coronary CT angiography (CCTA) on the diagnostic performance of a machine learning-based CT-derived fractional flow reserve (FFR_CT). This nationwide retrospective study enrolled participants from 10 individual centers across China. FFR_CT analysis was performed in 570 vessels in 437 patients. Invasive FFR and FFR_CT values ≤ 0.80 were considered ischemia-specific. Four-score subjective assessment based on image quality and objective measurement of vessel enhancement was performed on a per-vessel basis. The effects of body mass index (BMI), sex, heart rate, and coronary calcium score on the diagnostic performance of FFR_CT were studied. Among 570 vessels, 216 were considered ischemia-specific by invasive FFR and 198 by FFR_CT. Sensitivity and specificity of FFR_CT for detecting lesion-specific ischemia were 0.82 and 0.93, respectively. Area under the curve (AUC) of high-quality images (0.93, n = 159) was found to be superior to low-quality images (0.80, n = 92, p = 0.02). Objective image quality and heart rate were also associated with diagnostic performance of FFR_CT, whereas there was no statistical difference in diagnostic performance among different BMI, sex, and calcium score groups (all p > 0.05, Bonferroni correction). This retrospective multicenter study supported the FFR_CT as a noninvasive test in evaluating lesion-specific ischemia. Subjective image quality, vessel enhancement, and heart rate affect the diagnostic performance of FFR_CT. • FFR_CTcan be used to evaluate lesion-specific ischemia. • Poor image quality negatively affects the diagnostic performance of FFR_CT. • CCTA with ≥ score 3, intracoronary enhancement degree of 300-400 HU, and heart rate below 70 bpm at scanning could be of great benefit to more accurate FFR_CTanalysis.