基于深度学习的帕金森症步态障碍分析与研究

Analysis and Research on Gait Disorder in Parkinson's Disease Based on Deep Learning

步行作为人类的基本活动之一,对其进行分析在临床研究中有重要意义。通过对公开帕金森症足底压力数据集Gait in Parkinson’s Disease进行分析,设计划分步态周期的方法,并提取步态特征参数。应用一种混合神经网络(GRU-DNN),将门控循环单元(GRU)与深度神经网络(DNN)相结合对帕金森症病情诊断进行分类。数据分析为临床诊断提供更多客观依据,从而辅助医生进行病情诊断。为验证方法的有效性,使用该网络对数据集中具有病情标签的步态信息进行分类预测,结果显示:在帕金森症诊断实验中,该网络的识别准确率为98.7%;在帕金森症严重程度诊断实验中,该网络对于严重程度为2级的识别准确率达到100%;对于其余严重程度的识别准确率达到98%。

As one of the basic activities of human beings,the analysis of walking is of great significance in clinical research.In this paper,we analyzed the published plantar pressure dataset Gait in Parkinson's Disease,designed a method to divide the gait period,and extracted the characteristic parameters of gait.A hybrid neural network(GRU-DNN)was applied to classify Parkinson's disease by combining the gated circulation unit(GRU)and the deep neural network(DNN).Data analysis provided more objective basis for clinical diagnosis,thereby assisting doctors in diagnosing the disease.To verify the effectiveness of the method,the network was used to classify and predict gait information with disease labels in the dataset.In Parkinson's disease diagnosis experiment,the recognition accuracy of this network was 98.7%.In the Parkinson's severity diagnosis experiment,the network achieved 100%recognition accuracy for severity level 2,and 98%for the rest of the severity level.