Department of Medical Imaging, Jinling Hospital, Medical School of Nanjing University, Nanjing, Jiangsu 210002, China.
Department of Radiology, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, 100730, China.
Eur J Radiol. 2019 Jul;116:90-97. doi: 10.1016/j.ejrad.2019.04.011. Epub 2019 Apr 23.
To evaluate the diagnostic performance of coronary computed tomography angiography derived fractional flow reserve (CT-FFR) with invasive fractional flow reserve (FFR) in patients with coronary artery disease" before "with invasive fractional flow reserve serving as the reference standard.
CT-FFR values based on a machine learning algorithm (cFFR) in 183 vessels of 136 patients from four centers were measured with invasive FFR as reference standard. The diagnostic performance from our multicenter study was combined into a meta-analysis following a literature search in Web of Science, PubMed, Cochrane library to identify studies comparing diagnostic performance of coronary computed tomography angiography (CCTA) and CT-FFR. Sensitivity, specificity, accuracy were analyzed on both per-vessel and per-patient basis for intermediate lesions and by algorithm.
Our multicenter study demonstrated sensitivities, specificities, and accuracies of cFFR and CCTA of 0.85, 0.94, 0.90, and 0.95, 0.28, 0.55 on a per-vessel basis, respectively. For our meta-analysis, pooled sensitivities, specificities, and accuracies of CT-FFR and CCTA were 0.85, 0.82, 0.82, and 0.85, 0.57, 0.65 with AUC of 0.86 (95%CI: 0.83˜0.89) and 0.83 (95%CI: 0.79˜0.86) on a per-vessel basis, respectively. The sensitivity, specificity and accuracy for intermediate lesions using cFFR were 0.84, 0.92, and 0.89. No significant difference was found among different algorithms of CT-FFR (P < 0.001).
This multicenter study with meta-analysis showed that CT-FFR had a high diagnostic accuracy in determining ischemia-specific lesions and intermediate lesions. There was no significant difference when comparing the combined diagnostic performance of different algorithms of CT-FFR with invasive FFR as the reference standard.
以有创血流储备分数（FFR）为参考标准，评估基于机器学习算法的冠状动脉计算机断层血管造影术衍生的血流储备分数（CT-FFR）在冠状动脉疾病患者中的诊断性能。
对来自四个中心的 136 名患者的 183 个血管的 CT-FFR 值，使用有创 FFR 作为参考标准进行测量。根据文献检索 Web of Science、PubMed、Cochrane library 中比较冠状动脉计算机断层成像（CCTA）和 CT-FFR 诊断性能的研究，将多中心研究的诊断性能进行合并分析。基于血管和患者的个体，对中间病变和算法分别进行敏感性、特异性和准确性分析。
多中心研究结果显示，cFFR 和 CCTA 的敏感性、特异性和准确性分别为 0.85、0.94、0.90 和 0.95、0.28、0.55，基于血管的个体。对于我们的荟萃分析，CT-FFR 和 CCTA 的汇总敏感性、特异性和准确性分别为 0.85、0.82、0.82 和 0.85、0.57、0.65，AUC 为 0.86（95%CI：0.83˜0.89）和 0.83（95%CI：0.79˜0.86），基于血管的个体。使用 cFFR 时，中间病变的敏感性、特异性和准确性分别为 0.84、0.92 和 0.89。不同 CT-FFR 算法之间的敏感性、特异性和准确性无显著差异（P＜0.001）。
这项多中心研究结合荟萃分析表明，CT-FFR 对确定缺血性特异性病变和中间病变具有较高的诊断准确性。当以有创 FFR 为参考标准比较不同 CT-FFR 算法的联合诊断性能时，差异无统计学意义。
