基于机器学习方法研究造成缺血特异性狭窄的斑块特征  

Machine learning based study for evaluating the plaques characteristics of ischemia-specific artery stenosis

作　　者：张晓蕾 张元秀 唐春香 侯阳[3] 郑敏文[4] 张佳胤[5] 张波 张代民[7] 徐磊[8] 胡秀华[9] 王怡宁[10] 杨健[11] 刘辉[12] 张龙江 ZHANG Xiaolei;ZHANG Yuanxiu;TANG Chunxiang;HOU Yang;ZHENG Minwen;ZHANG Jiayin;ZHANG Bo;ZHANG Daimin;XU Lei;HU Xiuhua;WANG Yining;YANG Jian;LIU Hui;ZHANG Longjiang(Department of Diagnostic Radiology,Jinling Hospital,Medical School of Nanjing University,General Hospital of Eastern Theater Command,Nanjing 210002,China;Laboratory of Image Science and Technology,School of Computer Science and Engineering,South East University;Department of Radiology,Shengjing Hospital of China Medical University;Department of Radiology,Xijing Hospital,Fourth Military Medical University;Institute of Diagnostic and Interventional Radiology,Shanghai Jiao Tong University Affiliated Sixth People’s Hospital;Department of Radiology,Jiangsu Taizhou People’s Hospital;Department of Cardiology,Nanjing First Hospital,Nanjing Medical University;Department of Radiology,Beijing Anzhen Hospital,Capital Medical University;Department of Radiology,Shao Yifu Hospital Affiliated to Medical College of Zhejiang University,Sir Run Run Show Hospital,Zhejiang University School of Medicine;Department of Radiology,Peking Union Medical College Hospital,Chinese Academy of Medical Sciences and Peking Union Medical College;Department of Radiology,The First Affiliated Hospital of Xi’an Jiaotong University;Department of Radiology,Guangdong Provincial People’s Hospital)

机构地区：[1]南京大学医学院附属金陵医院东部战区总医院放射诊断科,南京210002 [2]东南大学计算机科学与技术学院影像科学与技术实验室 [3]中国医科大学附属盛京医院放射科 [4]第四军医大学附属西京医院放射科 [5]上海交通大学附属第六医院放射介入科 [6]江苏泰州人民医院放射科 [7]南京医科大学附属南京市第一医院心脏科 [8]首都医科大学附属安贞医院放射科 [9]浙江大学医学院附属邵逸夫医院放射科 [10]中国医学科学院北京协和医学院附属北京协和医院放射科 [11]西安交通大学第一附属医院放射科 [12]广东省人民医院放射科

出　　处：《国际医学放射学杂志》2021年第5期523-528,共6页International Journal of Medical Radiology

基　　金：江苏省科技项目(BE2020699)。

摘　　要：目的利用机器学习(ML)方法探讨定量、定性的冠状动脉斑块特征以及血流动力学参数对缺血特异性狭窄血管的影响。材料与方法回顾性分析同时行冠状动脉CT血管成像(CCTA)、有创冠状动脉造影(ICA)及FFR测量的255例病人共328支血管的临床及影像资料。以FFR≤0.80作为提示病变特异性缺血的指标,依据FFR值将纳入血管分为非缺血组(FFR>0.80)和缺血组(FFR≤0.80)。测量所有纳入斑块的CCTA参数,包括斑块的定量、定性特征以及血流动力学参数。采用LogitBoost算法建立随机森林模型,通过信息增益排序方法自动选择特征。分类准确度、敏感度以及f1值(分类准确度与敏感度的调和平均值)用以评价随机森林模型对造成缺血特异性狭窄斑块的预测价值,并计算受试者操作特征(ROC)曲线下面积。采用十折分层交叉验证法计算模型的总体分类准确度。结果应用ML方法得出,血流动力学参数FFR_(CT)、ΔFFR_(CT)是预测缺血特异性狭窄最重要的2个特征,其次是斑块的定量、定性特征,包括脂质斑块体积、斑块弯曲、斑块不规则、非钙化斑块体积、狭窄程度、纤维斑块体积和管腔体积。在排序前10的特征中有9个是CCTA相关参数,只有1个临床参数。采用随机森林模型预测缺血特异性狭窄的分类准确度为0.940,敏感度为0.940,f1值为0.940;预测缺血特异性狭窄的ROC曲线下面积为0.992,模型的总体分类准确度为0.921±0.047。结论ML方法能够很好地预测引起心肌缺血的冠状动脉特异性狭窄病变的斑块特征。Objective To explore the influence of quantitative and qualitative plaque characteristics and the hemodynamic parameters of ischemia-specific stenosis by machine learning.Methods A retrospective analysis was performed in 255 patients with 328 vessels who underwent coronary CT angiography(CCTA),invasive coronary angiography(ICA)and flow fractional reserve(FFR).FFR≤0.80 was defined as lesion specific myocardial ischemia.The 328 vessels were divided into two groups by FFR(FFR>0.80 and FFR≤0.80).CCTA-derived parameters including quantitative and qualitative plaque characteristics and the hemodynamic parameters were evaluated.The random forest model was established by LogitBoost algorithm.Features were auto selected by information gain ranking.The precision rate,recall rate,f1 value(harmonic mean value of precision rate and recall rate)and area under the receiver operator characteristic(ROC)curve were studied by random forest model to predict the influence factors of plaques which caused ischemia-specific stenosis.The overall classification accuracy of the model was calculated by 10-fold stratified cross-validation.Results The hemodynamic parameters(FFR_(CT)andΔFFR_(CT))were the most important features of ischemia-specific stenosis prediction from machine learning,followed by quantitative and qualitative plaque characteristics,including lipid plaque volume,bending,irregularity,non-calcified plaque volume,stenosis grade,fibrotic plaque volume and lumen volume.Nine of the top 10 features were CCTA-derived parameter,while only one was clinical parameter.The influence factors of ischemia-specific stenosis were studied by random forest model.The influence factors of ischemia-specific stenosis were studied by random forest model.The precision rate,recall rate and f1 value of the model were 0.940,0.940 and 0.940,respectively.The area under ROC curve for predicting ischemia-specific stenosis was 0.992,and the overall classification accuracy of the model was 0.921±0.047.Conclusions Machine learning could predict coronary

关 键 词：机器学习 CT冠状动脉成像 血流储备分数 斑块特征 

分 类 号：R445.3[医药卫生—影像医学与核医学] R54[医药卫生—诊断学]