CT影像组学在非瓣膜性心房颤动患者左心耳血栓识别中的价值  

Value of CT-based Radiomics in Identifying Left Atrial Appendage Thrombus in Patients With Nonvalvular Atrial Fibrillation

作　　者：辛然 窦冠华 李超 荆晶 陈韬 穆洋 王晶 杨俊杰 郭军 陈韵岱 XIN Ran;DOU Guanhua;LI Chao;JING Jing;CHEN Tao;MU Yang;WANG Jing;YANG Junjie;GUO Jun;CHEN Yundai(Senior Department of Cardiology,The Sixth Medical Center,Chinese People's Liberation Army General Hospital,Beijing 100048,China;Department of Cardiology,The Second Medical Center and National Clinical Research Center for Geriatric Diseases,Chinese People's Liberation Army General Hospital,Beijing 100853,China)

机构地区：[1]中国人民解放军总医院第六医学中心心血管病医学部,北京100048 [2]中国人民解放军总医院第二医学中心心内科国家老年疾病临床医学研究中心,北京100853

出　　处：《中国循环杂志》2023年第4期448-455,共8页Chinese Circulation Journal

基　　金：国家重点研发计划项目课题(2021YFC2500505、2016YFC1300304)。

摘　　要：目的:探究基于CT的影像组学在非瓣膜性心房颤动患者左心耳血栓检出中的价值。方法:回顾性纳入2015年5月至2020年5月在我院先后接受冠状动脉CT血管造影(CCTA)和经食道超声心动图(TEE)检查的非瓣膜性心房颤动患者364例,以TEE诊断左心耳血栓为金标准分为血栓组(n=33)和非血栓组(n=331)。分别测量左心耳感兴趣区(ROI)内最低CT密度值(LAAmin)与同一轴面的升主动脉根部ROI内最低CT密度值(AAmin),并计算二者比值(LAAmin/AAmin)。自动提取左心耳ROI影像组学特征并采用随机森林算法进行特征筛选,利用RMS包构建影像组学评分。基于CHA_(2)DS_(2)-VASc评分、CT参数和影像组学评分运用多元Logistic回归构建临床模型、影像学模型、影像组学模型和复合模型。绘制ROC曲线、校准曲线、决策曲线进行区分度、校准度和临床净获益等效能评价。结果:两组间左心耳开口直径、左心耳长径、LAAmin和LAAmin/AAmin的差异均有统计学意义(P均<0.05)。本研究提取1232个影像组学特征,并运用随机森林算法筛选出23个特征。模型效能评价方面,复合模型的AUC高于临床模型和影像学模型(0.922 vs.0.742,P<0.001;0.922 vs.0.856,P<0.05),差异有统计学意义,与影像组学模型间的差异无统计学意义(0.922 vs.0.905,P>0.05)。与其余三个模型相比,复合模型的准确度(ACC)、特异度(SPE)、阳性似然比(PLR)、阳性预测值(PPV)和Kappa值更高(ACC:89.0%;SPE:88.8%;PLR:8.133;PPV:44.8%;Kappa值:0.545)。校准曲线和决策曲线提示复合模型具有更高的校准度和临床净获益。结论:影像组学技术可以提高CCTA对心房颤动患者左心耳血栓的诊断效能。Objectives:To explore the performance of radiomics technology in improving the efficiency of identifying left atrial appendage(LAA)thrombus by coronary CT angiography(CCTA)in patients with non-valvular atrial fibrillation(NVAF).Methods:364 patients with NVAF who underwent CCTA and transesophageal echocardiography(TEE)from May 2015 to May 2020 in our hospital were retrospectively enrolled and divided into the thrombus group(n=33)and the non-thrombus group(n=331)based on TEE results.The ratio of minimum Hounsfield units(HU)in the left atrial appendage(LAAmin)to minimum HU in the ascending aorta(AAmin)was calculated(LAAmin/AAmin).Radiomics features were automatically extracted from the matrix of the region of interest(ROI)of LAA.Random forest algorithm was used for feature selection and the Rad score was established using the RMS package.Four models(clinical model,image model,radiomics model and complex model)were constructed by multiple logistic regression based on CHA2DS2-VASc,CT parameters and Rad score.ROC curve,calibration curve and decision curve were plotted,and the performance parameters were calculated to evaluate the diagnosis efficacy of these models in terms of discrimination,calibration and clinical net benefit.Results:The LAA orifice diameter,the length of LAA,LAAmin,and LAAmin/AAmin were significantly different between the two groups(P<0.05).1232 radiomics features were extracted and 23 target features were selected.The AUC of the complex model was significantly higher than those of the clinical model and the image model(0.922 vs.0.742,P<0.001;0.922 vs.0.856,P<0.05),but not significantly different from that of the radiomics model(0.922 vs.0.905,P>0.05).The accuracy(ACC),specificity(SPE),positive likelihood ratio(PLR),positive prospective value(PPV),and kappa value of the complex model were significantly higher(ACC:89.0%;SPE:88.8%;PLR:8.133;PPV:44.8%;kappa value:0.545)than those of the other three models.The calibration curves and decision curves showed that the complex model demonstrated better calibr

关 键 词：心脏计算机体层摄影 左心耳血栓 非瓣膜性心房颤动 影像组学 

分 类 号：R54[医药卫生—心血管疾病]