Artificial Intelligence Based Fully-Automated One-Click On-Site CT-Derived Fractional Flow Reserve: A Stepwise Development and Validation from Multicohort China CT-FFR Studies
Bang Jun Guo,Meng Chun Jiang,Xiang Guo,Chun Xiang Tang,Jian Zhong,Meng Jie Lu,Chun Yu Liu,Xiao Lei Zhang,Hong Yan Qiao,Fan Zhou,Peng,Yi Xue,Minwen Zheng,Yang Hou,Yi Ning Wang,Jiayin Zhang,Bo Zhang,Dai Min Zhang,Lei Xu,Xiu Hua Hu,Chang Sheng Zhou,Jian Hua Li,Xin Sheng Mao,Zhi Wen Yang,Guangming Lu,Longjiang Zhang
DOI: https://doi.org/10.2139/ssrn.4455336
2023-01-01
Abstract:Background: CT-derived fractional flow reserve (CT-FFR) has been developed and used with a promising potential in routine clinical practice. However, currently available CT-FFR is time consuming and complex due to needs of many manual editing. Thus, a fully-automated on-site CT-FFR technology has an urgent clinical appeal. Its feasibility, diagnostic and prognostic value in patients with chronic coronary syndromes or acute chest pain have not yet been evaluated. Methods: The fully-automated, one-click, and on-site CT-FFR software was developed based on two major steps: including automated coronary plaque segmentation and luminal extraction to generate the accurate coronary tree, and combined with the reduced order 3D computational fluid dynamics (CFD) model to accelerate the CT-FFR calculation process. Three cohorts were retrospectively collected to validate its diagnostic performance and prognostic value. A total of 463 consecutive patients with 600 vessels from updated China CT-FFR study in cohort 1 underwent CCTA and invasive FFR within 90 days were collected for diagnostic performance evaluation. For cohort 2, a total of 901 chronic coronary syndromes patients with index CT-FFR and clinical outcomes at a median 35 months follow-up were retrospectively analyzed. In cohort 3, the association between index CT-FFR from triple-rule-out CTA and major adverse cardiac events (MACE, including all cause of death, nonfatal myocardial infarction, and unexpected revascularization after the index visit) in patients with acute chest pain in the emergency department were further evaluated. Diagnostic performance of CT-FFR, CCTA and invasive coronary angiography (ICA) in identifying functional myocardial ischemia on a per-vessel and per-patient basis were evaluated, moreover, the clinical friendliness between our proposed CT-FFR and cFFR (Siemens) and uCT-FFR (United-Imaging) were compared. Finally, the associations of index CT-FFR and MACE in patients with chronic coronary syndromes or acute chest pain were evaluated. Findings: On a per-vessel basis, sensitivities, specificities, and accuracies were 0·80, 0·83, and 0·82 for CT-FFR, 0·90, 0·22, and 0·48 for CCTA, and 0·98, 0·19, and 0·50 for ICA, respectively, using invasive FFR as reference standard. Per-patient sensitivities, specificities, and accuracies were 0·84, 0·81 and 0·82 for CT-FFR, 0·92, 0·18, and 0·50 for CCTA, and 0·98, 0·17, and 0·52 for ICA, respectively. There was higher AUC for CT-FFR than CCTA and ICA on a per-patient (0·82 vs 0·55 vs 0·57) and per-vessel (0·81 vs 0·56 vs 0·59) basis (p < 0·001 for all). Compared to the manual CT-FFR techniques, the average consumed time and number of clicks were significantly lower for our proposed CT-FFR technique than cFFR and uCT-FFR (average time: 3·96 ± 0·04 minutes vs 13·06 ± 0·28 minutes vs 12·27 ± 0·28 minutes and number of clicks: one-click vs 59·20 ± 12·13 vs 62·41 ± 8·82, p < 0·001 for all). Technical successful rates of our proposed CT-FFR were 99.8% and 99.4% for Cohort 2 and Cohort 3, respectively. After adjustment for positive CCTA results, a positive CT-FFR result was associated with an increased risk of MACE in patients with chronic coronary syndromes (hazard ratio [HR] 18·37, 95% CI: 9·76-34·60; p< 0·001) and acute chest pain (HR 7·24, 95% CI: 1·65-31·69; p<0·001). Compared to CCTA alone, CT-FFR was a superior predictor for MACE both in patients with chronic coronary syndromes (C-index: 0·82 vs 0·65, p< 0·001) and acute chest pain (C-index: 0·81 vs 0·72, p<0·001). Interpretation: A novel artificial intelligence based fully-automated, one-click, on-site CT-FFR is feasible, user-friendly and accurate in detecting functionally significant stenosis in chronic and acute chest pain patients. Positive CT-FFR results were associated with higher incidence of MACE, and CT- FFR was superior to CCTA or triple-rule-out CTA alone for MACE prediction both in patients with chronic coronary syndromes or acute chest pain. Based on these findings, the fully-automated, one-click, on-site CT-FFR can be integrated into routine clinical practice.Funding: Jiangsu Province Key Project of Comprehensive Prevention and Control of Chronic Diseases (grant BE2020699).Declaration of Interest: All authors declare no competing interests.Ethical Approval: The collection of patient data was approved by the local ethics board at each institution. Written informed consent was waived in Cohort 1 and Cohort 3, while written informed consent was obtained for all participants in Cohort 2.