Self-organizing maps for time series analysis of electromyographic data

In the present work, time series analysis is used for simultaneous analysis of multiple channels of data, and to define complex inter- and intra-channel features of electromyographic (EMG) data for pattern classification. The ability to objectively quantify differences in complex patterns of EMG data has potential value for clinical and research applications. In this report, an unsupervised clustering neurocomputational approach, self-organizing maps (SOM), was applied to the problem of time series analysis of EMG data to provide a means to objectively quantify differences in muscle activity patterns related to differences in the underlying movement task, ambulation at different velocities and cadences on a treadmill. The SOM technique provided a means to discern differences between LEEMG data from the different ambulation tasks. In addition, observation of the weight vector associated with each SOM cluster in comparison to the ensemble averaged LEEMG for each condition helped determine underlying task-related changes in LEEMG patterns.