The development of a gait waveform classifier using electromyograpic signals

走路，或步态，是一种基本的人类运动。一个人行走技能的丧失或改变会极大地影响一个人的机能和生活质量。生物机械师使用最先进的运动捕捉系统来记录步态模式，并描述正常和异常运动的机制(步态分析)。定量步态分析通常检查运动学、运动学和肌电数据。由于这些数据的高维性、时间依赖性、高变异性和高度相关性，对这些机械和肌电数据的分析是具有挑战性的。因此，比较步态曲线(即正常和异常)是非常困难的。为了应对与大量机械数据相关的挑战，研究人员开发了各种数据约简和分类技术。然而，大多数人在检测异常运动模式的能力和/或其可解释性方面都受到限制。这项研究旨在进一步设计和实现算法，将大量的机械数据减少到一系列基于年龄匹配的标准数据的一维正态指数。这些指数或分类器使用自动系统(运动捕获和后续数据分析)来检测基于幅度、运动模式和步态曲线之间的相关性的运动异常。输出是一系列的数字分数，使机械和生理意义，并产生可解释的结果。很少存在能够满足合并相关多维输入数据和提供易于解释的输出的双重约束的分类器。提出的工作是新颖的，因为它结合了机械波形数据、多节段足部运动学和肌电小波数据。到目前为止，还没有其他分类器包括所有这些重要的衡量标准。该分类器的开发将为研究人员提供一种自动化工具，以减少机械和肌电数据，并量化步态波形的差异。步态分类器还将识别哪些测量是异常运动和正常运动的最佳判别器。这将有助于更好地理解人群或个体之间步态模式的差异。

Walking, or gait, is a fundamental human motion. The loss or alteration of a person's walking skills greatly affects one's ability to function and quality of life. Biomechanists use state of the art motion capture systems to record gait patterns and describe the mechanics of normal and abnormal movements (gait analysis). Quantitative gait analyses typically examine kinematic, kinetic, and electromyographic data. The analysis of this mechanical and myoelectric data is challenging due to the high-dimensionality, temporal dependence, high variability, and highly correlated nature of the data. As a result, comparison of gait curves (i.e. normal vs. abnormal) is very difficult. To address the challenges associated with the large volumes of mechanical data, researchers have developed various data reduction and classification techniques. However, most are limited in either their ability to detect abnormal movement patterns and/or their interpretability. This study aims to further my work on the design and implementation of algorithms that reduce the vast quantities of mechanical data to a series of one-dimensional indices of normality based on age-matched normative data. The indices, or classifiers, use an automatic system (motion capture and subsequent data analysis) to detect movement abnormalities based on magnitude, pattern of motion, and correlations between gait curves. The output is a series of numerical scores that make mechanical and physiological sense and yield interpretable results. Very few classifiers exist that can satisfy the two-fold constraint of incorporating correlated multi-dimensional input data and providing readily interpretable output. The proposed work is novel in that it incorporates mechanical waveform data, multisegment foot kinematics, and myoelectric wavelet data. To date, no other classifiers include all of these important measures. The development of the classifier will provide researchers with an automated tool to reduce mechanical and myoelectric data and quantify differences in gait waveforms. The gait classifier will also identify which measures are the best discriminators of abnormal and normal motion. This will lead to a greater understanding of the differences in gait patterns between populations or individuals.