一种基于支持向量机的扩张型心肌病计算机辅助诊断方法  

A computer-aided diagnosis method of dilated cardiomyopathybased on SVM

作　　者：邵晓宁[1] 靳雅楠[1] 张文博[1] 程敬亮[1] SHAO Xiaoning;JIN Yanan;ZHANG Wenbo;CHENG Jingliang(Department of Magnetic Resonance,The First Affiliated Hospital of Zhengzhou University,Zhengzhou 450052)

机构地区：[1]郑州大学第一附属医院磁共振科,郑州450052

出　　处：《北京生物医学工程》2023年第5期503-506,512,共5页Beijing Biomedical Engineering

基　　金：河南省医学科技攻关计划(LH-GJ20210342)资助。

摘　　要：目的扩张型心肌病(dilated cardiomyopathy,DCM)的确诊过程相对繁琐,为帮助临床医生诊断DCM,本研究在分析DCM患者磁共振(magnetic resonance,MR)心肌短轴T1-mapping图像的直方图特征参数的基础上,提出了一种基于支持向量机(support vector machine,SVM)的半自动计算机辅助诊断DCM的方法。方法回顾性分析50例扩张型心肌病患者的临床资料和MR图像,并对24例健康志愿者进行前瞻性研究。在3T磁共振扫描仪上使用改进的Look-Locker反转恢复(modified Look-Locker inversion recovery,MOLLI)序列获得T1-mapping图像。由有经验的影像科医师在短轴层面T1-mapping图像上绘制心内膜和心外膜,从而提取出心肌区域并计算心肌T1值的直方图参数,包括均值、方差、最大值、最小值、众数、百分位数等。随机选取70%被试者的直方图参数作为SVM的训练集,余下30%为测试集。测试程序共运行了100次,平均准确度作为整体准确度。结果DCM组与健康对照组相比,均值、最大值、众数等直方图特征参数显著高于正常对照组。以12个直方图参数为分类依据的SVM分类器的诊断准确率为0.81±0.07。结论基于SVM的磁共振心肌短轴T1-mapping图像直方图分析是一种有效的半自动计算机辅助诊断DCM的方法。Objective The diagnosis process of dilated cardiomyopathy(DCM)is relatively cumbersome.In order to help clinicians diagnose DCM,this study aims to propose a semi-automatic computer-aided diagnosis method of DCM based on support vector machine(SVM)on the basis of analyzing the histogram parameter characteristics of MRI myocardial short axis T1 mapping images of DCM patients.Methods The clinical data and MR images of 50 patients with dilated cardiomyopathy were retrospectively analyzed,and 24 healthy volunteers were prospectively studied.T1-mapping images were obtained using the modified Look-Locker inversion recovery(MOLLI)sequence on a 3T MR magnetic resonance scanner.Experienced radiologists draw endocardium and epicardium on the short axis T1-mapping,so as to extract the myocardial region and calculate the histogram characteristic parameters of myocardial T1 value,including mean,variance,maximum,minimum,mode,percentile,etc.70%of the subjects’histogram parameters were randomly selected as the training set of SVM,and the remaining 30%were the test set.The program ran 100 times,and the average accuracy was considered as the overall accuracy.Results Compared with the healthy control group,the histogram characteristic parameters such as mean,maximum and mode in DCM group were significantly higher than those in normal control group.The diagnostic accuracy when using the SVM classifier with all twelve histogram parameters was 0.81±0.07.Conclusions The histogram analysis of MRI myocardial short axis T1 mapping image based on SVM was an effective semi-automatic computer-aided diagnosis method of DCM.

关 键 词：扩张型心肌病 心血管磁共振 T1-mapping 直方图分析 支持向量机 

分 类 号：R318.04[医药卫生—生物医学工程]