目的 初步探索基于心脏磁共振(cardiovascular magnetic resonance,CMR)影像组学揭示肥厚型心肌病(hypertrophic cardiomyopathy,HCM)患者心肌病变影像组学特征的临床应用价值.材料与方法 回顾性分析宁夏医科大学总医院2018年1月1日至2022年12月31日确诊为HCM的患者142例及健康对照组72例的CMR影像及临床资料.选择CMR未增强亮血电影序列左心室长轴位两腔心、四腔心二尖瓣口水平切面,以舒张末期壁厚(end-diastolic wall thickness,EDWT)为基础,将HCM患者心肌分为肥厚区域(无HCM家族史,EDWT≥15 mm;有HCM家族史,EDWT≥13 mm)与非肥厚区域(无HCM家族史,EDWT<15 mm;有HCM家族史,EDWT<13 mm),和健康对照组在室间隔、心尖、左心室侧壁、左心室前壁及左心室下壁等相同解剖部位室壁分别进行感兴趣区(region of interest,ROI)勾画及影像组学特征提取,分组后行影像组学特征组间比较.采用Mann-Whitney U检验、递归特征消除法(recursive feature elimination,RFE)及最小绝对收缩与选择算法(least absolute shrinkage and selection operator,LASSO)进行影像组学特征筛选并使用支持向量机(supportvector machine,SVM)算法建立影像组学模型,最后计算准确度、敏感度、特异度、阳性预测值(positive predictive value,PPV)、阴性预测值(negative predictive value,NPV)受试者操作特征曲线下面积(area under the curve,AUC),分别评价各模型效能;并统计分析HCM患者及健康对照组临床资料.结果 (1)HCM组的左心室收缩末期容积(left ventricular end systolic volume,LVESV)、LVESV指数(LVESV index,LVESVI)、左心室舒张末期室壁质量(left ventricular end diastolic wall mass,LVED wall mass)、左心室舒张末期室壁+乳头肌质量(left ventricular end diastolic wall+papillary mass,LVED wall+papillary mass)、左心室收缩末期室壁质量(left ventricular end systolic wall mass,LVES wall mass)、左心室收缩末期室壁+乳头肌质量(left ventricular end systolic wall+papillary mass,LVES wall+papillary mass)、舒张末期左室壁最大厚度(left ventricular maximum wall thickness,LVMWT)均高于健康对照组(P<0.05).(2)HCM患者肥厚区域、非肥厚区域与健康对照组左心室相同解剖部位室壁影像组学鉴别模型鉴别效果良好;HCM非肥厚区域组(左心室前壁)与HCM肥厚区域组(左心室前壁)影像组学鉴别模型为本研究的最优模型,其AUC值、准确度、敏感度、特异度、PPV及NPV在训练集中为0.98、93%、94%、92%、89%、96%;在测试集中为0.97、94%、92%、96%、92%、96%.结论 影像组学可通过深入全面的定量分析HCM患者不同病变状态心肌组织,为评估HCM患者心肌组织学改变提供客观敏感的新方法,有助于更准确地评估HCM患者病情、提高CMR检查效率及减少检查费用,有望为探寻方便快捷检测心肌组织学状态的人工智能技术开发及临床应用提供新思路.
Study of the value of radiomics based on cardiac magnetic resonance in hypertrophic cardiomyopathy
Objective:To extract and analyze the myocardial radiomics features of patients with hypertrophic cardiomyopathy(HCM)and healthy controls,and to explore the clinical application value of radiomics features of HCM based on cardiac magnetic resonance(CMR)radiomics.Materials and Methods:The CMR imaging and clinical data of 142 patients with HCM and 72 healthy controls from the General Hospital of Ningxia Medical University from January 1,2018 to December 31,2022 were retrospectively collected and analyzed.The CMR unenhanced bright-blood cine sequence was used to select the left ventricular long axis two-chamber and four-chamber mitral valve level view.Based on the end-diastolic wall thickness(EDWT),the myocardium of HCM patients was divided into hypertrophic region(no family history of HCM,EDWT≥15 mm;family history of HCM,EDWT≥13 mm)and non-hypertrophic region(no family history of HCM,EDWT<15 mm;family history of HCM,EDWT<13 mm),the region of interest(ROI)was delineated and radiomics features were extracted from the ventricular wall of the same anatomical part of the healthy control group,such as ventricular septum,apex of heart,lateral wall of left ventricle,anterior wall of left ventricle and inferior wall of left ventricle,and the radiomics features were compared between the two groups after grouping.Mann-Whitney U test,recursive feature elimination(RFE),and least absolute shrinkage and selection operator(LASSO)were used to select the radiomics features and establish the radiomics model.Finally,the accuracy,sensitivity,specificity,positive predictive value(PPV)and negative predictive value(NPV)and area under the curve(AUC)were used to evaluate the performance of each model.The clinical data of HCM patients and healthy controls were statistically analyzed.Results:(1)In the HCM group,left ventricular end systolic volume(LVESV),LVESV index(LVESVI),left ventricular end diastolic wall mass(LVED wall mass),left ventricular end diastolic wall+papillary mass(LVED wall+papillary mas),left ventricular end systolic wall mass(LVES wall mass),left ventricular end systolic wall+papillary mass(LVES wall+papillary mass),left ventricular maximum end-diastolic wall thickness(LVMWT)were higher than those of the healthy control group(P<0.05).(2)The radiomics features of hypertrophic regions and non-hypertrophic regions in HCM patients and the healthy control group had a good discrimination effect on the same anatomical part of left ventricular wall.The radiomics differential model between HCM non-hypertrophic region group(left ventricular anterior wall)and HCM hypertrophic region group(left ventricular anterior wall)was the optimal model in this study,and its AUC value,accuracy,sensitivity,specificity,PPV and NPV were 0.98,93%,94%,92%,89%and 96%in the training set.In the test set,the values were 0.97,94%,92%,96%,92%,96%.Conclusions:The results show that the radiomics model not only has good efficacy in the identification of hypertrophic and non-hypertrophic regions of the myocardium in HCM and healthy people,but also can well distinguish hypertrophic and non-hypertrophic regions of the myocardium in HCM,indicating that the radiomics method can provide a potential and sensitive new technology for the evaluation of myocardial histological changes through in-depth and comprehensive quantitative analysis of myocardial tissue in HCM patients with different pathological states.It provides a kind of potential and sensitive new technique for evaluation of myocardial histological changes.It is helpful to evaluate the condition of HCM patients more accurately,improve the efficiency of CMR examination and reduce the cost of examination,and provide a new idea for the development and clinical application of artificial intelligence technology to detect myocardial histological status easily and quickly in the future.
hypertrophic cardiomyopathycardiac magnetic resonanceradiomicsmagnetic resonance imagingcine sequencesupportvector machine
万方数据
维普
