姿势步态障碍是帕金森病主要临床表现中致伤、致残最严重、最频繁的一种症状，严重影响了患者的正常生活。目前对于帕金森患者步态障碍的研究与治疗主要聚焦于冻结步态症状，对于其他较轻程度的步态障碍表现缺乏关注。此外，当前研究侧重于运用信号特征预测患者的患病情况、服药状态以及刺激状态，对步态障碍的精确诊治没有直接帮助。本文基于对运动学和表面肌电信号的分析，探究了帕金森患者轻度步态障碍的生理特征与治疗方法，并结合信号特征与量表评分为步态障碍的诊治提供参考。本文完成的主要工作与创新点包括：首先，本文提出了基于运动学信号提取步态试验直行过程的批处理方法，并进一步利用陀螺仪实现直行过程中的步态周期划分。在此基础上，采用时域分析提取了11个运动学特征与4个表面肌电特征，用于探究信号特征随步态障碍进展的变化趋势。结果显示，共有14个特征与步态评分呈现出显著的单调关系。其次，本文对比了低频、高频以及变频刺激对帕金森患者步态障碍的改善效果。按照刺激状态对所有特征进行统计，有12个特征表现出了显著的组间差异。结合特征随步态障碍进展的变化趋势，发现三种刺激频率均能有效改善步态障碍，其中高频刺激治疗效果最好，低频刺激次之，变频刺激再次之。此外，本文分析了慌张步态患者的步态行为，并基于支持向量机（support vector machines, SVM）完成了慌张步态识别实验。慌张步态样本数显著少于非慌张步态样本，采用BC（box constraints）法与SMOTE（synthetic minority oversampling technique）法对不平衡数据集进行处理。结果显示，BC-SVM分类器的敏感性为100%，特异性为98%；SMOTE-SVM分类器的敏感性与特异性均为100%，两者分类效果相当。最后，本文完成了非慌张步态患者的步态识别实验。同样采用BC法与SMOTE法处理不平衡数据集，结果显示SMOTE-SVM分类器的效果明显优于BC-SVM，在各二分类问题中的分类准确度均高于98%。在此基础上，对比了运动学特征与表面肌电特征各自对非慌张步态患者步态识别的效果。结果显示，以运动学特征作为输入的SMOTE-SVM分类器效果更好，分类准确率高于97%。
Postural instability and gait disorders are problematic symptoms for advanced Parkinson's disease (PD) patients. Currently, treatment for gait disorders in PD patients mainly focuses on freezing of gait, while other gait disorders are less well studied. These studies have focused on the use of signal features for predicting prevalence, medication and/ or stimulation status, which does not directly aid accurate diagnosis and treatment of gait disorders. In this study, we explored the physiological features and treatment of mild gait disorders PD patients based on the analysis of kinematics and surface EMG signals. The main objective of this investigation was to provide reference for the diagnosis and treatment of gait disorders from signal features and scale scores. In the first study, a batch processing method was proposed based on the kinematics signal to extract walking processes in the gait test, with the use of gyroscope to characterize gait cycles. On this basis, 11 kinematic features and 4 surface EMG features were extracted using time-domain analysis. These were then used to investigate the changing trend of signal features as the gait disorder deteriorated. The results showed that there were significant monotonous relationships between 14 features and gait scores respectively.In the second study, the effects of low frequency, high frequency, and variable frequency stimulation on gait disorders were investigated in PD patients receiving deep brain stimulation (DBS) therapy. Following specific stimuli, all gait features were counted and there were a total of 12 features that showed significant differences between groups. It was found that the three types of stimulation can effectively improve gait disorders, with greatest effects seen from high frequency stimulation, followed by low frequency stimulation and lastly by variable frequency stimulation.Analysis of gait behavior in PD patients with festinating gait was also conducted. Festinating gait recognition was based on support vector machines (SVM). The number of festinating gait samples was significantly less than that of non-festinating gait samples. This imbalanced data set was processed using BC (box constraints) and SMOTE (synthetic minority oversampling technique) methods. The results showed that the BC-SVM classifier had a sensitivity of 100% and specificity of 98%, while the SMOTE-SVM classifier had a sensitivity and specificity of 100%. These results demonstrated equivalent performance of the two classifiers.In the final part of this study, gait recognition experiment for PD patients with non-festinating gait was analyzed. The BC and SMOTE methods were used for imbalanced data sets. The results showed that the SMOTE-SVM classifier was significantly better than the BC-SVM, and the classification accuracy was higher than 98% in each of the binary classification problems. On this basis, we compared the effects of kinematics and surface EMG on gait recognition for non-festinating gait PD patients. The results showed that the SMOTE-SVM classifier with kinematics as input had better performance and the classification accuracy was 97%.
