A wavelet based time frequency analysis of electromyograms to group steps of runners into clusters that contain similar muscle activation patterns

PurposeTo wavelet transform the electromyograms of the vastii muscles and generate wavelet intensity patterns (WIP) of runners. Test the hypotheses: 1) The WIP of the vastus medialis (VM) and vastus lateralis (VL) of one step are more similar than the WIPs of these two muscles, offset by one step. 2) The WIPs within one muscle differ by having maximal intensities in specific frequency bands and these intensities are not always occurring at the same time after heel strike. 3) The WIPs that were recorded form one muscle for all steps while running can be grouped into clusters with similar WIPs. It is expected that clusters might have distinctly different, cluster specific mean WIPs.MethodsThe EMG of the vastii muscles from at least 1000 steps from twelve runners were recorded using a bipolar current amplifier and yielded WIPs. Based on the weights obtained after a principal component analysis the dissimilarities (1-correlation) between the WIPs were computed. The dissimilarities were submitted to a hierarchical cluster analysis to search for groups of steps with similar WIPs. The clusters formed by random surrogate WIPs were used to determine whether the groups were likely to be created in a non-random manner.ResultsThe steps were grouped in clusters showing similar WIPs. The grouping was based on the frequency bands and their timing showing that they represented defining parts of the WIPs. The correlations between the WIPs of the vastii muscles that were recorded during the same step were higher than the correlations of WPIs that were recorded during consecutive steps, indicating the non-randomness of the WIPs.ConclusionsThe spectral power of EMGs while running varies during the stance phase in time and frequency, therefore a time averaged power spectrum cannot reflect the timing of events that occur while running. It seems likely that there might be a set of predefined patterns that are used upon demand to stabilize the movement.研究目的：对股肌（Vastii Muscles）的肌电图（Electromyography, EMG）进行小波变换，生成跑步者的小波强度模式（Wavelet Intensity Pattern, WIP）。验证以下三项研究假设：1）单一步幅中股内侧肌（Vastus Medialis, VM）与股外侧肌（Vastus Lateralis, VL）的小波强度模式，较二者错步（相差一个步幅）后的小波强度模式更为相似；2）单块肌肉的小波强度模式存在差异，表现为其峰值强度出现在特定频带，且这些峰值强度并非总在足跟触地后同一时刻出现；3）跑步过程中单块肌肉所有步幅的记录所得小波强度模式，可聚类为具有相似WIP的组别，且不同聚类可能拥有差异显著、聚类专属的平均WIP。 研究方法：采用双极电流放大器，记录12名跑步者至少1000步的股肌肌电信号，并生成对应的小波强度模式。基于主成分分析（Principal Component Analysis, PCA）得到的权重，计算小波强度模式间的相异性（1-相关系数）。将相异性矩阵用于分层聚类分析，以筛选出具有相似WIP的步幅组别。通过随机生成的替代WIP所形成的聚类，判断该分组是否为非随机产生。 研究结果：所有步幅被划分为具有相似WIP的聚类组别，分组依据为频带及其时序特征，该特征可体现WIP的核心构成要素。同一步幅中记录的股肌WIP间的相关系数，高于连续步幅记录的WIP间相关系数，证实了WIP分组的非随机性。 研究结论：跑步过程中肌电图的频谱功率在站立相的时域与频域均存在动态变化，因此时间平均功率谱无法反映跑步过程中事件发生的时序特征。推测机体可能存在一套预定义的运动模式，可根据需求调用以维持运动稳定性。