动态收敛微分神经网络帕金森病诊断系统

Dynamic convergent differential neural network for Parkinson’s disease diagnosis system

帕金森病是一种在老年人中常见的神经系统疾病,它影响患者的说话、写字和平衡能力.根据声学特征对帕金森病进行诊断可以尽早发现病情,有利于早日介入治疗.基于神经动力学方法,本文提出了一种动态收敛微分神经网络(dynamic convergent differential neural network,DCDNN),利用帕金森病的声学特征进行识别诊断.此前,对样本进行标准化、统计池特征增强和主成分分析.数值仿真实验表明所提出方法获得97.22%的准确率,比现存最好方法高斯支持向量机(Gaussian SVM)高约6个百分点.

Parkinson’s disease is a common nervous system disease existing in the elderly.It affects a patient’s ability to speak,write and balance.The diagnosis of Parkinson’s disease based on its acoustic features can help to find the disease as early as possible,which is conducive to early intervention treatment.In this paper,based on the neural dynamics method,a novel dynamic convergent differential neural network(DCDNN)is proposed.It uses the acoustic features of Parkinson’s disease for recognition and diagnosis.Specifically,the samples of the Parkinson’s disease data set are standardized and the features are enhanced by the statistical pools.Furthermore,the principal component analysis is applied for feature extraction and the proposed DCDNN is applied for classification.The computer simulation results verify that the proposed framework has achieved the classification accuracy of 97.22%on the Parkinson’s disease data set,which is about 6 percentage points higher than the Gaussian SVM.