Application to biomedical signals - Practical Biomedical Signal Analysis Using MATLAB - page 215

In the system proposed by Kleine et al. [Kleine et al., 2007] a 10x13 grid with
inter-electrode distance 5 mm was used. Bipolar and Laplacian reference systems
were considered ( Figure 4.56) . The MUAPs were isolated from the background ac-
tivity by peak detection. In the clustering procedure both the spatial (waveform and
amplitude difference between channels) and the temporal (time-course of the poten-
tial in each channel) information were taken into account. Additional information
was provided by the firing statistics. The clusters were inspected and adjusted inter-
actively; that is, the operator could split or merge clusters. Then the MUAP templates
were subtracted from continuous sEMG. In the decomposition procedure a single
row (perpendicular to muscle fibers) of bipolar derivations was taken into account.
The template matching procedure was repeated with different subsets of available
data. It was confirmed that two bipolar montages, separated by a few millimeters in
the direction parallel to the fibers will record almost the same waveform, delayed by
a few milliseconds. On the other hand in the direction perpendicular to the muscle
fibers the amplitude of bipolar MUAP is largely attenuated (see Figure 4.54 and Fig-
ure 4.55) . In this case the muscle fibers of a particular MU cover only a small part
of muscle fibers from another MU, so a single row of channels gives a reasonable
separation of MUAPs.
The limitation of the method is the restriction to low force contractions, since
at higher forces two or more MUAPs may have almost identical shapes and also
the superimposed potentials become a problem. It was suggested that sEMG may
be useful for quantification of neurogenic diseases, where the number of remaining
MUs and MUAP interference are low.
In general, sEMG may be helpful in case of disturbances of the neuromuscular
system leading to decrease of firing rates of MUs. An example may be the automatic
classification of MUAPs in sEMG recorded from muscles paralyzed by spinal cord
injury, where involuntary EMG activity occurs [Winslow et al., 2009]. However, it
should be mentioned that changes in firing rates are not specific to any particular
neuromuscular disorder.
The application of the blind source separation method for identification of motor
unit potential trains in sEMG was proposed by Nakamura and coworkers [Nakamura
et al., 2004]. The signals from eight electrodes placed perpendicularly to the mus-
cle fibers were processed. Principal component analysis and independent component
analysis were applied. The ICA approach performed better than PCA in separating
groups of similar MUAP waveforms, however single MUAPs were not separated
completely into independent components. In case of delays between potentials com-
ing from the same motor unit ICA regards them as several different sources. The
limitation of the method is connected also with the fact that the procedure does not
take into account the firing rates of the motor units, which limits useful information
available for classification. The BSS methods assume independence of the underly-
ing sources which hampers their application for higher force level when the synchro-
nization of the motor units is high. The advantages of the BSS are connected with
the fact that the method does not rely on prior estimation of the shapes of MUAPs. It
seems that the approach might be promising, for example, as a preprocessing step in
the procedure of sEMG decomposition.