基于PCNN-LSTM神经网络的膝关节摆动信号分类识别  

Classification of Vibroarthrographic Signals Based on PCNN-LSTM Neural Network

作　　者：杨佳 邱天爽 刘宇鹏[2] Yang Jia;Qiu Tianshuang;Liu Yupeng(Faculty of Electronic Information and Electrical Engineering,Dalian University of Technology,Dalian 116024,Liaoning,China;Zhongshan Hospital,Dalian University,Dalian 116001,Liaoning,China)

机构地区：[1]大连理工大学电子信息与电气工程学部,辽宁大连116024 [2]大连大学附属中山医院,辽宁大连116001

出　　处：《中国生物医学工程学报》2021年第2期129-136,共8页Chinese Journal of Biomedical Engineering

基　　金：国家自然科学基金(61671105,61172108,61139001,81241059)。

摘　　要：膝关节摆动(VAG)信号是指膝关节屈曲或伸展时发出的声音或振动信号,可灵敏、客观地描述膝关节的健康状态,在膝关节疾病的无创检测中具有重要作用。现有的对VAG信号正常和异常分类方法自动化程度低,且分类准确度较低,总体性能有待进一步提升。因此,提出一种基于改进卷积神经循环网络(PCNN-LSTM)的VAG信号分类算法。首先,利用经验模式分解(EMD)和小波变换,将一维VAG信号变换为二维时频特征谱图,并将其用作数据集;然后,在串行神经网络的基础上融合并行卷积神经网络结构,再与LSTM神经网络相结合构成改进的PCNN-LSTM模型,以此区分正常或异常的VAG信号,实现对膝关节健康状态的自动检测。采用由加速度传感器(181A02)和USB采集仪(FSC812)所采集的真实VAG信号,构建数据集对所提出算法性能进行验证。数据集由654例样本构成,其中包括健康数据222例和患有膝关节疾病的数据432例。实验表明,所提出算法的分类正确率为96.93%,灵敏度为100%,特异性为95.56%,相比其他算法可得到较好的分类识别效果,对于膝关节疾病的无创检测和辅助诊断具有重要意义。The vibroarthrographic(VAG)signal is a sound of the knee joint during flexion and extension.VAG signal can be used to describe the health status of knee joint sensitively and objectively.Hence,it is often used in the detection of knee joint diseases.However,at present,classification accuracy of the normal and abnormal classification method of VAG signal is still low and not automated.The performance needs to be further improved.To solve this problem,in this paper,a classification algorithm of VAG signal based on improved convolutional neural network(PCNN-LSTM)was proposed.First,empirical mode decomposition(EMD)and wavelet transform are used to transform one-dimensional VAG signal into two-dimensional time-frequency characteristic spectrum,which was used as data set.Second,on the basis of CNN,the parallel CNN network structure was combined with LSTM neural network to form the PCNN-LSTM model,which could classify normal or abnormal VAG signals and realizd the automatic detection of knee joint health status.In this paper,the performances of the proposed algorithm were verified by the data set that composed of the real VAG signals collected by the acceleration sensor(181 A02)and USB acquisition instrument(FSC812).The data set consisted of 654 samples,including 222 health data and 432 data of patients with knee diseases.Results showed that the classification accuracy of the proposed algorithm was 96.93%,the sensitivity was 100%,and the specificity was 95.56%.Compared with other algorithms,the proposed algorithm achieved better results,and realized the classification and recognition of VAG signals,which was of great significance for non-invasive detection and auxiliary diagnosis of knee joint diseases.

关 键 词：膝关节摆动信号 经验模态分解 小波变换 卷积神经网络 

分 类 号：R318[医药卫生—生物医学工程]