基于支持向量机多模态超声模型诊断肾疾病  

Multi-modal ultrasound in diagnosis of renal diseases based on support vector machine

作　　者：李广涵 刘建 武敬平[1] 田艳[1] 刘将 马立勇[3] 刘跃军[4] 张波[1] 郑敏[1] LI Guanghan;LIU Jian;WU Jingping;TIAN Yan;LIU Jiang;MA Liyong;LIU Yuejun;ZHANG Bo;ZHENG Min(Department of Ultrasound Medical,China Japan Friendship Hospital,Beijing 100029,China;China Japan Friendship Clinical College of Peking University,Beijing 100029,China;School of Information Science and Engineering,Harbin University of Science and Technology,Weihai 264209,China;School of Automation,Harbin University of Science and Technology,Harbin 150080,China)

机构地区：[1]中日友好医院超声医学科,北京100029 [2]北京大学中日友好临床学院,北京100029 [3]哈尔滨工业大学(威海)信息科学与工程学院,山东威海264209 [4]哈尔滨理工大学自动化学院,黑龙江哈尔滨150080

出　　处：《中国医学影像技术》2020年第6期898-902,共5页Chinese Journal of Medical Imaging Technology

基　　金：国家政府间国际科技创新合作重点专项(2017YFE0110500);山东省自然科学基金(ZR2018MF026)。

摘　　要：目的比较基于支持向量机(SVM)和传统Logistic回归法基于常规超声、彩色多普勒超声和弹性成像参数构建的多模态超声模型诊断肾脏疾病的效能。方法收集94例肾脏疾病患者(肾病组)及无肾脏疾病的对照组患者109名,分别进行常规超声、彩色超声和剪切波弹性检查。采用Logistic回归法和SVM构建模型。利用随机数字法将全部201例患者按照3∶1分为2组,以其中153例为训练样本,进行单因素变量判断和建立SVM模型;以50例为验证样本,评价SVM模型的预测效果。结果Logistic回归方程纳入左肾皮质弹性硬度和右肾宽度。Logistic回归模型预测肾脏疾病的准确率为83.74%,SVM模型为85.10%(χ^2=0.21,P=0.65)。结论多模态超声对于肾脏疾病具有较高诊断效能;SVM和Logistic模型的诊断效能相似。Objective To compare the effectiveness of multi-modal ultrasound,including conventional ultrasound,color Doppler ultrasound and shear wave elastic imaging in diagnosis of renal diseases based on support vector machine support vector machine(SVM)and traditional Logistic regression.Methods Totally 94 patients with pathologically proved renal diseases(RD group)and 109 patients without renal diseases(control group)were collected and examined with conventional ultrasound,color Doppler ultrasound and shear wave elastic imaging,respectively.SVM and Logistic regression were used for modeling.Then all 203 patients were divided into 2 groups according to 3∶1,then 153 cases were used as SVM's training samples for single factor variable judgment and model establishment,the other 50 cases were used as validation samples to evaluate the prediction effect of SVM model.Results The elastic hardness of left renal cortex and the width of right kidney entered the regression equation in Logistic regression.The accuracy of Logistic regression model for diagnosis of renal diseases was 83.74%,of SVM model was 85.10%(χ^2=0.17,P=0.68).Conclusion Multimodal ultrasound has high effectiveness for diagnosis of renal diseases.SVM and Logistic models have similar diagnostic effectiveness.

关 键 词：肾疾病 超声检查 弹性成像技术 支持向量机 LOGISTIC模型 

分 类 号：R692[医药卫生—泌尿科学] R445.1[医药卫生—外科学]