Multivariate analysis of surface electromyography signals

As the primary method of measuring muscle activation, the surface electromyography (sEMG) is of great importance in the study of motor deficits seen in patients with brain injuries and neuromuscular disorders. While clinicians have long intuitively understood that deficits in motor control are relat...

Full description

Bibliographic Details
Main Author: Chiang, Joyce Hsien-yin
Language:English
Published: University of British Columbia 2011
Online Access:http://hdl.handle.net/2429/31587
Description
Summary:As the primary method of measuring muscle activation, the surface electromyography (sEMG) is of great importance in the study of motor deficits seen in patients with brain injuries and neuromuscular disorders. While clinicians have long intuitively understood that deficits in motor control are related to inappropriate recruitment of muscle synergies across several muscles, sEMG recordings are still typically examined in a univariate fashion. However, most traditional univariate techniques are unable to quantitatively capture the complex interactions between muscles during natural movements. To address this issue, multivariate signal processing techniques are employed in this thesis to study muscle co-activation patterns in patient populations. A method for classification of multivariate sEMG recordings between stroke and healthy subjects is proposed. The proposed classification scheme utilizes the eigenspectra of time-varying covariance patterns between sEMG channels as feature vectors and the support vector machines (SVM) as classifiers. Despite the minimal differences in the RMS profiles of individual muscles, the proposed scheme is able to effectively differentiate between healthy and stroke subjects. Moreover, the classification rate is shown to be monotonically related to the severity of motor impairment. This simple, biologically-inspired approach is able to quantitatively capture the subtle differences in muscle recruitment patterns between two populations and appears to be a promising means to measure motor performance. The other approach to modeling multivariate sEMG utilizes the HMM-mAR framework, which combines hidden Markov models (HMMs] and multivariate autoregressive (mAR) models. Different forms of sEMG data are analyzed, including raw sEMG, amplitude sEMG and carrier sEMG. The classification between healthy and stroke subjects is performed using structural features derived from estimated model parameters. Both the raw and carrier data produce excellent classification performance. The proposed method represents a fundamental departure from most existing classification methods where only amplitude sEMG is analyzed or mAR coefficients are directly used as feature vectors. In contrast, our analysis shows that the structural features of the carrier sEMG can enhance the classification performance and provide additional insights into motor control. === Applied Science, Faculty of === Electrical and Computer Engineering, Department of === Graduate