基于拓扑非线性动态建模的神经退行性疾病异常步态识别

Abnormality Detection in Neurodegenerative Disease Analysis Based on Topological Nonlinear and Dynamic Modelling of Gait

通过对步态节律变化的观察,可以实现对人体复杂系统的观测,在行走过程中,步幅间隔时间序列的动态特性,能有效反映人体系统的状态变化,可用于步态异常检测及相关疾病辨识。人体步态传感信息的时间序列相空间重建,是一种有效表征系统非线性动力学特性的建模方法。相空间的几何建模和统计分析,是异常步态识别典型的分析方法,被广泛应用于神经退行性疾病检测等临床研究中。该文从空间拓扑特性分析的角度出发,提出一种基于拓扑非线性动态建模的分析方法,用于神经退行性疾病的异常步态识别。该文首先采用延时嵌入的相空间重构方法,将步态的波动时间序列转化为抽象相空间的状态点云;其次,采用基于计算拓扑中的持续同调工具,提取状态点云所在空间的拓扑描述信息;再次,采用基于拓扑描述的持续态势图,构建时间序列的拓扑非线性动态特征;最后,融合步态周期中左右足的步幅间隔、站立间隔、摆动间隔时间序列的拓扑非线性动态特征,作为分类器输入,构建出异常步态的机器学习识别模型。对患有肌硬化症、亨廷顿病和帕金森病的神经退行性疾病患者,进行5 min异常步态的连续行走数据(50步滑动窗数据)测试,步态识别的准确率分别为0.8750(0.9146)、0.9406(0.9623)和0.9583(0.9614)。因此,拓扑非线性动态建模分析是一种有效的神经退行性疾病异常步态检测方法,为基于步态分析的神经退行性疾病检测和可穿戴数据分析提供了一种新的思路。

The complex system of human body can be observed by observing the change of gait rhythm.The dynamic characteristics of the time series of stride interval during walking can effectively reflect the state change of human system, which can be used for abnormal gait detection and related disease identification.Time series phase space reconstruction of human gait sensor information is an effective modeling method to characterize nonlinear dynamics of the system. Geometric modeling and statistical analysis of phase space are typical analysis methods for abnormal gait recognition, which are widely used in clinical research such as neurodegenerative disease detection. In this paper, a topological nonlinear dynamic modeling method is proposed to identify abnormal gait in neurodegenerative diseases from the perspective of spatial topological analysis. Firstly, a phase-space reconstruction method based on time-delay embedding is used to transform the wave time series of gait into an abstract phase-space state point cloud. Secondly, the continuous homology tool based on computational topology is used to extract the topology description information of the space where the state point cloud resides. Thirdly, the topological nonlinear dynamic characteristics of time series are constructed by using the continuous situation graph based on topological description. Finally, a machine learning recognition model of abnormal gait was constructed by integrating the topological nonlinear dynamic characteristics of the time series of stride interval, standing interval and swing interval of left and right feet in the gait cycle as the input of classifier. The results showed that the area under receive-operator-curve was 0.875 0(0.914 6), 0.940 6(0.962 3) and 0.958 3(0.961 4) in the 5-minute continuous walking data(50-step sliding window data) of abnormal gait in patients with neurodegenerative diseases of muscular sclerosis, Huntington’s disease and Parkinson’s disease, respectively. Therefore, topological nonlinear dynamic modeling analysis is an effective gait detection method for neurodegenerative diseases, and provides a new idea for neurodegenerative disease detection and wearable data analysis based on gait analysis.