基于人工智能动态CT心肌灌注成像分析技术的临床应用研究

目的 评价基于人工智能的动态CT心肌灌注（CTP）半自动分析软件Myocardiac Kit（MK）测量心肌血流参数的稳定性和准确性。 方法 前瞻性纳入接受负荷动态CTP联合冠状动脉CT血管成像（CCTA）的87例疑似冠心病病人,男67例,女20例,平均（60.98±0.78）岁。所有病人均在1周内接受有创冠状动脉造影（ICA）并测量血流储备分数（FFR）。由2名医师独立采用MK软件和CCTA工作站分析数据,计算心肌血流量（MBF）等动态CTP参数以及CCTA直径狭窄率,并记录软件分析数据所需时间。采用组内相关系数（ICC）评价观察者间在心肌节段和冠状动脉水平上参数测量的一致性。以在ICA检查中狭窄程度 ≥ 90%或FFR≤0.80作为心肌缺血的诊断标准,采用独立样本t检验比较缺血心肌和非缺血心肌MBF差异。绘制受试者操作特征（ROC）曲线,并计算曲线下面积。采用约登指数计算MBF判断心肌缺血的最佳临界值,分别计算CCTA上直径狭窄率 ≥ 50%、MBF以及两者联合诊断心肌缺血的敏感度、特异度及准确度。结果 研究共纳入261支冠状动脉和1 479个心肌节段。在心肌节段水平和血管水平上,2名医师对各参数测量较为一致（ICC≥0.60）。缺血心肌节段平均MBF低于对应非缺血节段的（123.14±41.83） mL·100 mL^-1·min^-1和（147.47±43.98） mL·100 mL^-1·min^-1,P<0.05],而缺血冠状动脉供血心肌节段的平均MBF亦低于非缺血冠状动脉供血的心肌节段 （124.34±42.86） mL·100 mL^-1·min^-1和148.68±44.49） mL·100 mL^-1·min^-1,P<0.05]。在血管水平,MBF最佳临界值为115.0 mL·100 mL^-1·min^-1,联合血管狭窄 ≥ 50%和MBF诊断心肌缺血的诊断效能最高,ROC曲线下面积为0.91（95%CI:0.87~0.95）]。MK软件平均数据处理时间为（10.51±1.95） min。结论 基于人工智能的动态CTP半自动分析软件具有稳定性好、结果准确、操作简便等优点,具有较好的临床应用推广价值。

Clinical application of dynamic CT myocardial perfusion imaging analysis technique based on artificial intelligence

Objective To evaluate the stability and accuracy of dynamic CT myocardial perfusion (CTP) semi-automatic analysis software Myocardiac Kit (MK) based on artificial intelligence in measuring myocardial blood flow parameters. Method This prospective study included 87 patients (mean age 60.98±0.78 years; 67 males and 20 females) with suspected coronary heart disease who underwent dynamic CTP combined with coronary CT angiography (CCTA), all of them underwent invasive coronary angiography (ICA) and measured fractional flow reserve (FFR) within one week. Two physicians independently used the MK and CCTA workstations to obtain the dynamic CTP parameters including myocardial blood flow (MBF) and the CCTA diameter stenosis, and recorded the time needed for analyzing the data. Intra-group correlation coefficient (ICC) was used to evaluate the consistency of dynamic CTP parameters between observers at segmental and vascular level. ICA stenosis ≥ 90% or FFR≤0.80 were considered as the diagnostic criteria of myocardial ischemia. Independent sample t-test was used to compare the difference in MBF between ischemic and non-ischemic myocardium. Receiver operating characteristic(ROC) curve analysis was performed to calculate the area under the curve. The maximum Youden index was used to define optimal cutoff values for MBF. The sensitivity, specificity and accuracy of stenosis ≥ 50% on CCTA, MBF and their combination in the diagnosis of myocardial ischemia were calculated respectively. Results A total of 261 coronary arteries and 1 479 myocardial segments were included in the study. At the myocardial segmental level and vascular level, the consistency of the parameters measured by the two physicians was well(ICC≥0.60). The average MBF of ischemic segments was significantly lower than that of non-ischemic segments (123.14±41.83 mL·100 mL^-1·min^-1 vs 147.47±43.98 mL·100 mL^-1·min^-1, P<0.05), and the average MBF of the segments dominated by the ischemic coronary artery was also significantly lower than that of the non-ischemic coronary (124.34±42.86 mL·100 mL^-1·min^-1 vs 148.68±44.49 mL·100 mL^-1·min^-1, P<0.05). At the vascular level, the best critical value of MBF was 115.0 mL·100 mL^-1·min^-1. The diagnostic of stenosis ≥ 50% combined with MBF had the highest efficacy in diagnosing myocardial ischemia, the area under the ROC curve was 0.91 (95%CI: 0.87-0.95)]. The average time of using the MK to analyze dynamic CTP data was 10.51±1.95 min. Conclusion The semi-automatic analysis software of dynamic myocardial perfusion CT data has the advantages of good stability, accurate results and simple operation, so it should be of great value in clinical application.