Diagnostic performance of machine learning based FFRCTin coronary artery stenosis

YU Mengmeng; LI Yuehua; LI Wenbin; LU Zhigang; WEI Meng; SHEN Chengxing; YAN Jing; ZHANG Jiayin

doi:10.19300/j.2018.L6085zt

Abstract：Objective To investigate the diagnostic performance of machine learning based FFRCTacross the different intervals of FFRCTvalue for identifying functionally significant stenosis,with reference to invasive FFR. Methods We retrospectively reviewed the data of one hundred and twenty-five patients[mean age:61.0±8.2(range,42.0-88.0)years;79 males and 46 females]with 162 lesions, who underwent both coronary computed tomography angiography(CCTA)and invasive coronary angiography(ICA)within 2 weeks interval.Diameter stenosis derived from CCTA were recorded and FFRCT were performed on routine CCTA datasets using Siemens prototype software(cFFR,version 3.0.0).Lesions with FFR≤0.8 were considered to be hemodynamically functional significant stenosis,receiver operating characteristic curve analysis was performed to calculate the area under the receiver operating characteristic curve(AUC). Sensitivity, specificity, positive predictive value,negative predictive value,and accuracy were recorded.Results The AUC values were 0.85 and 0.76 for FFRCTand CCTA,respectively(P<0.05),for diagnosing functionally significant stenosis.Per-patient sensitivity,specificity, and diagnostic accuracy to identify a significant functionally stenosis were 85.7%,78.9%,86.1%,respectively,for FFRCT;and 77.6%,69.7%,76.8%,respectively,for CCTA.For lesions with FFRCTvalues below 0.69,0.70 to 0.80,0.81 to 0.89,and above 0.90,the diagnostic accuracies of FFRCTwere 86.4%,61.2%,88.6%,98.2%,respectively.Conclusion Using FFR calculated from ICA as the garden standard,it can achieve a high diagnostic accuracy when the FFRCTvalue is less than 0.7 or more than 0.8,otherwise efficient diagnostic performance is not guaranteed.