Diagnostic performance of on-site, CT-derived fractional flow reserve in predicting invasive fractional flow reserve and absolute myocardial blood flow

Abstract Background Computed tomography (CT)-derived fractional flow reserve (FFR) improves the diagnostic accuracy of coronary CT angiography (CCTA) but requires off-site transfer of images and long processing times. A novel deep learning-based, on-site FFR algorithm offers rapid (8-minute) assessment with promising diagnostic performance but requires external validation before clinical implementation. Purpose To examine the agreement and diagnostic performance of CT-derived FFR with invasive FFR (iFFR) as the reference for functionally significant coronary artery disease and to assess their correlation and diagnostic performance against myocardial perfusion with [15O]H2O positron emission tomography (PET) Methods Post-hoc analysis of the Danish Study of Non-Invasive Testing in Coronary Artery Disease 3 (Dan-NICAD 3) study. Patients with suspected chronic coronary syndrome and ³50% diameter stenosis on CCTA underwent [15O]H2O PET and invasive coronary angiography with three-vessel iFFR. The primary outcome was the agreement between CT-derived FFR and iFFR , and secondary outcomes included the diagnostic accuracy of CT-derived FFR in predicting abnormal iFFR and the diagnostic accuracy of CT-derived FFR and iFFR in predicting core lab-adjudicated ischemia. Results In total, 38 patients (mean age 64 ± 8.0 years, 76% male) with 87 vessels were included for analysis. iFFR ≤0.8 was present in 21/38 (55%) patients and 23/87 (26%) vessels. Median invasive and CT-derived FFR were 0.92 (IQR 0.80–0.97) and 0.84 (IQR 0.69–0-91) (median paired difference 0.083, 95% CI 0.063–0.10), respectively, with a rho of 0.74. Against invasive FFR ≤0.80 on a per-patient level, CT-derived FFR ≤0.8 had an accuracy of 84% (95% confidence interval (CI) 70–93%), sensitivity of 100% (95% CI 86–100%), specificity of 65% (95% CI 41–83%), positive predictive value (PPV) of 78% (95% CI 59–89%), negative predictive value (NPV) of 100% (95% CI 74–100%), negative likelihood ratio (LR) (LR-) of 0.0 (95% CI 0.0–0.21), and positive LR (LR+) of 2.8 (95% CI 1.5–5.4). In comparison with ≥50% diameter stenosis on CCTA, CT-derived FFR ≤0.8 against the reference of iFFR ≤0.8 had similar sensitivity, NPV, and LR- but higher diagnostic accuracy, specificity, PPV, and LR+ (Table 1 + Figure 1). Both CT-derived FFR and iFFR correlated weakly with stress myocardial blood flow (rho 0.30 and 0.36, respectively) and myocardial flow reserve (rho 0.20 and 0.38, respectively). The diagnostic accuracy of CT-derived and iFFR in predicting ischemia was 46% (95% CI 30–64%) and 61% (95% CI 42–76%), respectively, on a per-patient level, and 69% (95% CI 57–79%) and 75% (95% CI 63–84%), respectively, on a per-vessel level. Conclusions A novel on-site CT-derived FFR algorithm demonstrates fast and very high rule-out capability and incremental rule-in potential beyond diameter stenosis for functionally significant disease in patients with suspected stenosis at CCTA.Figure 1