基于机器学习方法心室舒张期和收缩期CT血流储备分数评估狭窄冠状动脉血流动力学  

Diagnostic accuracy of diastolic versus systolic CT-derived fractional flow reserve based on machine lear-ning for hemodynamic significance of coronary stenosis

作　　者：凡子榕 刘庆爽 王照谦[1] 贾崇富[1] 孙喜霞[1] 王浩[1] 潘双 杨志强[1] FAN Zi-rong;LIU Qin-shuang;WANG Zhao-qian(Department of Cardiovascular Radiology,the First Affiliated Hospital of Dalian Medical University,Liaoning 116011,China)

机构地区：[1]大连医科大学附属第一医院心血管放射科,辽宁大连116011 [2]昆明医科大学附属延安医院放射科,云南昆明650051

出　　处：《放射学实践》2025年第4期478-484,共7页Radiologic Practice

摘　　要：目的:以侵入性血管造影测得的血流储备分数(FFR ICA)为参考标准,探讨基于机器学习方法测得的心室舒张期和收缩期冠状动脉狭窄段CT血流储备分数(CTFFR)及跨狭窄段CTFFR差值(ΔCTFFR)对心肌缺血的评估价值。方法:使用第三代双源CT机对55例冠心病患者进行冠脉CTA检查,使用基于机器学习方法的CTFFR测量软件,分别基于心室舒张期和收缩期冠脉CTA数据测量冠脉狭窄病变处的CTFFR(分别称为dCTFFR和sCTFFR),并计算其近端与远端CTFFR的差值(ΔdCTFFR和ΔsCTFFR)。采用Pearson相关分析探讨dCTFFR和sCTFFR与FFR的关系。以同期测得的FFR ICA≤0.80作为参考标准,分别评估dCTFFR、sCTFFR、ΔdCTFFR和ΔsCTFFR对狭窄血管所致心肌缺血的诊断准确性。结果:55例患者共68支冠状动脉血管ICA诊断有狭窄病变(其中11例患者为多支血管狭窄),其中44支血管狭窄处FFR≤0.80、24支FFR>0.80。①dCTFFR与FFR的相关性良好(r=0.656,P<0.001),sCTFFR与FFR呈中度相关(r=0.457,P<0.001)。②在血管水平,以FFR≤0.80作为参考标准,dCTFFR的诊断效能优于sCTFFR(AUC分别为0.813和0.700,P=0.025),且ΔdCTFFR的诊断效能也优于ΔsCTFFR(AUC分别为0.833和0.742,P=0.027)。③在24支非缺血性狭窄病变(FFR>0.80)中,ΔdCTFFR和ΔsCTFFR假阳性诊断率(均为25.00%)均小于dCTFFR(54.17%)和sCTFFR(45.83%);在12支“灰区”狭窄病变(0.75≤FFR≤0.80)中,dCTFFR和ΔdCTFFR的假阴性诊断率(分别占16.67%和占25.00%)均小于sCTFFR(50.00%)和ΔsCTFFR(66.67%);对于32处严重缺血性狭窄病变(FFR<0.75),dCTFFR和ΔdCTFFR的假阴性诊断率分别为3.13%和12.50%,小于sCTFFR(15.63%)和ΔsCTFFR(21.88%)。结论:基于机器学习CTFFR测量技术获得的dCTFFR与侵入性FFR的相关性高于sCTFFR。与sCTFFR或ΔsCTFFR相比,dCTFFR和ΔdCTFFR的诊断效能更高,在评估血管狭窄所致心肌缺血时应作为首选指标。Objective:Taken the fractional flow reserve(FFR)measured by invasive coronary angiography(FFR ICA)as the reference standard,to investigated diagnostic accuracy of diastolic and systolic CT-derived fractional flow reserve(CTFFR)based on machine learning and the difference of CTFFR across stenosis(abbreviated asΔCTFFR)for specific ischemia of coronary stenosis.Methods:55 patients with coronary heart disease were examined using the third-generation dual-source CT.The CTFFR of the stenotic lesions on the diastolic and systolic CCTA images was measured using CTFFR software based on machine learning(abbreviated as dCTFFR and sCTFFR,respectively).The diffe-rence of CTFFR across stenotic lesions proximal to distal based on diastolic and systolic CCTA image data was abbreviated asΔdCTFFR andΔsCTFFR,respectively.The relationship between dCTFFR,sCTFFR and FFR was analyzed using Pearson correlation analysis.The diagnostic accuracy of dCTFFR,sCTFFR,ΔdCTFFR andΔsCTFFR were evaluated using FFR ICA≤0.80 as the reference standard for the diagnosis of specific ischemia of vascular stenosis.Results:68 coronary artery vessels of 55 patients were diagnosed as stenotic lesions with hemodynamic significance by ICA(including 11 patients with multiple vessel stenosis);68 vascular stenosis included 44 vessels with FFR ICA≤0.80 and 24 vessels with FFR ICA>0.80.①The dCTFFR correlated well with the FFR(r=0.656,P<0.001)and the sCTFFR was moderately correlated with the FFR(r=0.457,P<0.001).②At the vascular level,using FFR≤0.80 as the reference standard for the diagnosis of specific ischemia of vascular stenosis,The diagnostic performance of dCTFFR was superior to that of sCTFFR(AUCs were 0.813 and 0.700,respectively;both P=0.025)and the diagnostic performance ofΔdCTFFR was also superior to that of△sCTFFR(AUCs were 0.833 and 0.742,respectively;both P=0.027).③In 24 non-ischemic stenosis lesions(FFR>0.80),the false-positive diagnositic ratio ofΔdCTFFR orΔsCTFFR(both 25.00%)was less than that of dCTFFR(54.17%)or sCTFFR(45.83%

关 键 词：冠状动脉狭窄 血流储备分数 CT血管成像 机器学习 

分 类 号：R814.42[医药卫生—影像医学与核医学] R543.3[医药卫生—放射医学]