High-Speed Onsite Deep-Learning Based FFR-CT Algorithm: Evaluation Using Invasive Angiography as Reference Standard.

Background: Estimation of fractional flow reserve (FFR) from coronary CTA (FFR-CT) is an established method to assess coronary lesions' hemodynamic significance. However, clinical implementation has progressed slowly, partly related to offsite data transfer with long turnaround times while awaiting results. Objectives: We aimed to evaluate the diagnostic performance of FFR-CT computed onsite with a high-speed deep-learning based algorithm, using invasive hemodynamic indices as reference standard. Methods: This retrospective study included 59 patients (46 men, 13 women; mean age 66.5±10.2 years) who underwent coronary CTA (including calcium scoring) followed within 90 days by invasive angiography with invasive FFR and/or instantaneous wave-free ratio (iwFR) measurements from December 2014 to October 2021. Coronary artery lesions were considered to show hemodynamically significant stenosis in presence of invasive FFR ≤0.80 and/or iwFR ≤0.89. A single cardiologist evaluated CTA images using an onsite deep-learning based semiautomated algorithm employing a 3D computational flow dynamics model to determine FFR-CT for coronary artery lesions detected by invasive angiography. Time for FFR-CT analysis was recorded. FFR-CT analysis was repeated by the same cardiologist in 26 randomly selected examinations, and by a different cardiologist in 45 randomly selected examinations. Diagnostic performance and agreement were assessed. Results: Invasive angiography identified 74 lesions. FFR-CT and invasive FFR showed strong correlation (r=0.81), and, in Bland-Altman analysis, showed bias of 0.01 and 95% limits of agreement of -0.13 to +0.15. FFR-CT had AUC for hemodynamically significant stenosis of 0.975. At cutoff of ≤0.80, FFR-CT had accuracy of 95.9%, sensitivity of 93.5%, and specificity of 97.7%. In 39 lesions with severe calcifications (≥400 Agatston units), FFR-CT had AUC of 0.991, with cutoff of ≤0.80 yielding sensitivity of 94.7%, specificity of 95.0%, and accuracy of 94.9%. Mean analysis time per patient was 7 minutes 54 seconds. Interobserver and intraobserver agreement were good-to-excellent (intraclass correlation coefficient, 0.944 and 0.854; bias -0.01 and -0.01; 95% limits of agreement, -0.08 to +0.07, and -0.12 and +0.10, respectively). Conclusion: A high-speed onsite deep-learning based FFR-CT algorithm showed excellent diagnostic performance for hemodynamically significant stenosis, with high reproducibility. Clinical Impact: The algorithm should facilitate the FFR-CT technology's implementation into routine clinical practice.