基于迁移学习和支持向量机的胎心率分类方法  

Fetal Heart Rate Classification Using Transfer Learning and SVM

作　　者：叶明珠 赵治栋[1] YE Mingzhu;ZHAO Zhidong(School of Electronic Information,Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China)

机构地区：[1]杭州电子科技大学电子信息学院,浙江杭州310018

出　　处：《杭州电子科技大学学报（自然科学版）》2020年第3期14-18,43,共6页Journal of Hangzhou Dianzi University：Natural Sciences

基　　金：浙江省公益性技术应用研究计划资助项目(2017C31046)。

摘　　要：胎心率(Fetal Heart Rate,FHR)作为电子胎心宫缩监护(Cardiotocography,CTG)的重要生理参数,是目前临床评估胎儿宫内健康状况最为重要的安全指标。为降低主观分析的误诊率,简化传统胎儿监护模型,提出一种基于迁移学习和支持向量机(Support Vector Machine,SVM)的胎儿状态评估方法。首先,采用盲分割的方法对原始信号进行分割,通过插值法对截取的信号进行降噪预处理;然后,采用广义S变换来捕捉FHR信号在各个时间点的瞬时频率特征轨迹图,并从FHR信号的一维轨迹图中提取频域特征;最后,引入卷积神经网络AlexNet模型,学习得到的特征向量作为SVM输入进行胎心率分类。实验结果表明:该方法可以快速准确地对胎心率进行分类,采用预训练的AlexNet卷积神经网络和SVM,准确率达到97.90%。Fetal Heart Rate(FHR)is an important physiological parameter in Cardiotocography(CTG),which is the most important clinical safety index for assessing intrauterine health.To reduce the misdiagnosis rate of subjective analysis,and to simplify traditional fetal monitoring model,a fetal state assessment method based on transfer learning and Support Vector Machine(SVM)is proposed.Firstly,the original signal is segmented by blind segmentation method and preprocessed by interpolation method.Secondly,the generalized S transform is used to capture the instantaneous frequency characteristic trajectory of the FHR signal at each time point,and the features in frequency domain are extracted from the one-dimensional trajectory of the FHR signal.Finally,the features of the large number of input images learned from convolution neural network AlexNet model can serve as the input of the SVM.In this paper,the classification accuracy of 97.90% is achieved via the pre-trained AlexNet convolutional neural network and support vector machine,which indicates that this approach can effectively and accurately classify the fetal heart rate.

关 键 词：胎心宫缩图 胎心率 迁移学习 支持向量机 AlexNet 

分 类 号：TN911.7[电子电信—通信与信息系统]