Utilizing Deep Learning-based Computed Tomography Fractional Flow Reserve on Coronary Artery Disease Diagnosis and Treatment

PURPOSE: Machine learning-based coronary computed tomography fractional flow reserve (CT-FFR) holds great potential for assessing coronary ischemic status. The current literature lacks a comprehensive description of the routine implementation of CT-FFR in real world. To investigate the clinical characteristics and acceptance of CT-FFR in clinical decision-making among Chinese patients and subsequently assess the diagnostic accuracy of invasive coronary angiography as the reference. MATERIALS AND METHODS: In this retrospective single-center study, 4564 patients were included. In the first part, we conducted a baseline analysis of patients and their epicardial coronary arteries. Then, we analyzed hospitalization and revascularization in the context of application of CT-FFR, using logistic regression and Sankey diagrams. Finally, we performed a diagnostic analysis of 2718 vessels in 906 patients. RESULTS: The baseline analysis included a total of 4564 patients. A statistically significant distinction was observed in the traditional risk factors for coronary heart disease between 2 groups with CT-FFR 0.8 cutoff values. Logistic regression analysis and Sankey plots revealed a association between CT-FFR ≤0.8 and subsequent hospitalization. Finally, a diagnostic analysis was performed on 2718 vessels, and the optimal diagnostic model efficacy was achieved by using a CT-FFR cutoff value of 0.8 in conjunction with stenosis ≥70% for CCTA. CONCLUSIONS: Our study provides evidence that machine learning-based CT-FFR values exhibit a probably positive correlation with individuals presenting high-risk factors for coronary artery disease. Furthermore, we observed a influence of CT-FFR on the clinical decisions made by physicians. The integration of CT-FFR and CCTA has the potential to enhance diagnostic efficacy.