Application to biomedical signals - Practical Biomedical Signal Analysis Using MATLAB - page 211

tion linear discriminant analysis (LDA) and multilayer perceptron (MLP) were ap-
plied. The best performance was obtained for WP/PCA/LDA combination, yielding
a classification error of 6%. This result was an improvement in respect to procedures
based on time features of MUAPs, which for the same data gave an average error of
9%.
WT was also successfully applied for diagnosis and follow up of Parkinson dis-
ease [De Michele et al., 2003]. The sEMG for ballistic movement was recorded from
major pectoralis and posterior deltoid muscles. The Morlet complex wavelet was
used and cross-correlation was computed between continuous wavelet transforms
W f describing signals from the consid-

ered muscles. The wavelet cross-scalogram is given by:

( a , τ ) and W g ( a , τ ) of functions f ( t ) and g ( t )

W f ( W fg ( a , τ )= a , τ ) W g ( a , τ ) (4.44)

The cross-correlation spectrum is defined as:

2 2 2

| W fg ( a , τ ) | = | Re ( W fg ( a , τ )) | + | Im ( W fg ( a , τ )) | (4.45)

The integration of the local wavelet cross-correlation spectrum over τ gives the
global wavelet cross-spectrum. In distinction to FT here it is possible to select a
proper integration time interval. In [De Michele et al., 2003] the threshold was cho-
sen as 5% above maximum wavelet power peak. The time-frequency distributions
of cross-correlation power for the normal group was more concentrated in time and
frequency than for the Parkinsonian group, where a large spread was observed, espe-
cially in time. In order to quantify these differences a parameter called global power
was determined by integration of scalograms in frequency and time. This parameter
differentiated the Parkinsonian group from normal subjects better than conventional
medical measure. The global power index was proposed as a parameter useful in
estimation of the disease progress.

4.3.4.1 Surface EMG decomposition
Recently, the sEMG technique has drawn a lot of attention. One of the reasons is
the fear of infection by needle electrodes. The problem inherent to sEMG technique
is the influence of volume conduction and low-pass filtering effect of the tissues,
which causes an increase in MUAPs duration, smoothing of their shapes, and a de-
crease of the differences between them. For these reasons sEMG was initially treated
as an interference signal, from which only global properties of motor units and their
firing rates could be assessed.
Nevertheless, the application of surface EMG (sEMG) for detection of single MU
activity by means of small bipolar electrodes and their arrays was proposed by Gy-
dikov and coworkers in the early 1970s [Gydikov and Kosarov, 1972, Gydikov and
Gatev, 1982, Gydikov et al., 1986, Kosarov, 1974]. Further studies by these authors
provided information on the velocity of the activity propagation along muscle fibers
and on MU structure, in particular its size and location within the muscle, endplate
distribution, muscle fibers topography [Gydikov and Gantchev, 1989].