Estimating different velocities in wrist movements from information contained in surface electromyography: Application of a machine learning technique

The study of surface electromyography (sEMG) has several approaches. It is used to classify upper and lower extremity movements by identifying the muscle groups that have been excited to generate movements. In general, movements have certain properties related to the type of movement, the force, and the speed at which they are performed. The hypothesis of this study is that information about different speeds is contained in sEMG signals. Participants performed wrist movements at different speeds, following verbal instructions to alternate between fast and slow movements. Our objective was to estimate whether there is information in the sEMG signal that can be associated with the different speed conditions; therefore, binary differencing (two classes) was chosen to test this. These two conditions (fast and slow) were used as classes for analysis and classification based on surface electromyography signals. The moving window method was used to extract sEMG envelopes at two different speeds performed by the test subjects. A linear discriminant analysis model was created to estimate the velocities with the resulting model. Finally, cross-validation was performed to estimate sensitivity (76.67%), specificity (91.2%), and accuracy (approximately 87%).