Research On Decomposition Of Multi-channel Surface Electromyograms

The surface electromyogram (sEMG) signals, which in general are considered as the sum of motor unit action potential trains (MUAPt), can be recorded by electrodes on the surfaces of the contracting muscles. The characteristics of motor unit’s firing and recruitment could be obtained by decomposing the sEMG signals. The decomposing results can be used for investigating and diagnosing the neuromuscular system.In this doctoral dissertation, some novel approaches for decomposing multi-channel surface electromyogram signals are addressed. The main research works and achievements in this dissertation are stated below:1. A hybrid approach was successfully developed through combining the Convolution Kernel Compensation (CKC) method and fuzzy C means (FCM) clustering method for multi-channel surface electromyogram decomposition. The FCM method is used to process the initial innervation pulse trains (IPTs) of motor units (MUs) from a few channel’s surface EMG signals, then the CKC method is employed to estimate the final IPTs. Computer simulation results demonstrate the improved accuracy and efficiency of the proposed hybrid approach compared to the classic CKC method.2. A new method based on Convolution Kernel Compensation (CKC) to decompose multi-channel surface electromyogram (sEMG) signals is developed. The self-Organizing Map (SOM) neural network with the function of unsupervised learning and clustering is employed in this method. An initial IPT is firstly estimated, some time instants associated with the highest peaks from the initial IPT are clustered via SOM Neural Network. Then the final IPT can be obtained from the measurements associated with these time instants. The proposed method was tested through simulated signal. The effects of number of groups clustered by SOM, the signal to noise ratio (SNR) and the number of highest peaks selected from the initial pulse train on the pulse accuracy and the number of reconstructed IPTs were investigated. The results show that the proposed approach is effective in decomposing multi-channel sEMG signals. 3. A new approach that is able to approach the results of linear minimum mean square error (LMMSE) estimator has been developed for multi-channel surface electromyograms (EMG) decomposition. The K-means clustering (KMC) method was first employed to cluster vectors of measurements at time instants and then to estimate the initial IPT. A uniqiue multi-step iterative process was then performed to update the estimated IPT. The performance of the proposed approach in reconstructing IPTs was evaluated with both simulated and experimental surface EMG signals. It can reconstruct all the10IPTs with the true positive rates (TPR) being greater than90%from the simulated signals with a low SNR of-lOdB. More than10motor units were successfully extracted using this approach from the64-channel real sEMG signals of the first dorsal interosseous (FDI) muscles. A’two-source’test was further conducted. The high percentage of common pulses (over92%) between the IPTs and common MUs obtained from the two independent groups of sEMG signals indicates the reliability and capability of the proposed method in multi-channel sEMG decomposition.4. Two methods based on measurement correlation are proposed for decomposing multi-channel sEMG signals. One is to use Moore-Penrose pseudo-inverse to reconstruct a correlation matrix of the measurement matrix; the other is to use the singular value decomposition (SVD) to reconstruct a correlation matrix of the measurement matrix. These two methods gradually and iteratively increases the correlation between an optimized vector and the reconstructed matrix correlated with the measurement matrix based on the feature that the measurement vectors associated with the firing times generated by the same motor unit (MU) have a certain degree of similarity. The performances of the proposed methods were evaluated with both simulated and experimental sEMG signals. Simulation results show that both methods can successfully reconstruct over48IPTs with a true positive rate (TPR) greater than95%. Over15MUs were successfully extracted by using these two methods from real ulti-channel sEMG signals. The performance of the methods with experimental electrode array surface EMG was further validated by the "two sources" method, the results demonstrate the reability of these two methods in decomposing sEMG signals. This work was supported in part by NIH K99DK082644, NIH R00DK082644of NIH, USA.