Application to biomedical signals - Practical Biomedical Signal Analysis Using MATLAB - page 209

forms. It provides a graphical interface for displaying and editing results and al-
gorithms for template matching and resolving superpositions. The software may be
found at: http://www.emglab.net . At this location also MTLEMG [Florestal et al.,
2006], a MATLAB function for multichannel decomposition of EMG, including a
genetic algorithm for resolving superpositions may be found. EMG simulator avail-
able at http://www.emglab.net is a package for simulating normal and patholog-
ical EMG signals designed by [Hamilton-Wright and Stashuk, 2005]. The package
contains executable code for PCs and Macintosh and a user interface written in MAT-
LAB.
An automatic algorithm for decomposition of multichannel EMG recorded by wire
electrodes (including signals from widely separated recording sites) into MUAPs was
proposed by Florestal and coworkers [Florestal et al., 2009]. The program uses the
multichannel information, looking for the MUAPs with the largest single component
first, and then applies matches obtained in one channel to guide the search in other
channels. Each identification is confirmed or refuted on the grounds of information
from other channels. The program identified 75% of 176 MUAPs trains with accu-
racy of 95%. It is available at no cost at: http://emglab.stanford.edu .
As was mentioned above, both time and frequency features are useful for EMG
quantitative analysis [Pattichis and Elia, 1999]. Wavelet transform, which provides
description of signals in time and frequency domain, was applied for analysis of
EMG for normal, myogenic and neurogenic groups [Pattichis and Pattichis, 1999].
Four kinds of wavelets were tested including: Daubechies4, Daubechies20, Chui
(linear spline), and Battle-Lemarie (cubic spline) wavelets [Daubechies, 1992]. The
scalogram was inspected for different wavelet types and in respect of characteristics
of investigated groups of patients. Scalograms obtained by means of Daubechies4
wavelets had the shortest time spread in each band and they detected sharp spikes
with good accuracy. Daubechies20 wavelets had lower time resolution, however they
provided higher frequency resolution. The linear-spline scalograms showed a large
energy concentration in the lowest energy bands; on the contrary scalograms for cu-
bic spline wavelet were spread in the upper frequency band.
As might have been expected, scalograms for the neurogenic group had larger
time spreads and those for the myogenic group had shorter time-domain spreads
in respect to the normal group. Also shifts in frequency similar to these found in
[Pattichis and Elia, 1999] were reported in pathological groups. The reduction of
information was provided, since MUAP signals were described by a small number
of wavelet coefficients located around the main peak. High frequency coefficients
were well localized in time and captured the MUAP spike changes, whereas low
frequency coefficients described the time span and average behavior of the signal.
Wave l e t c o e fficients characterizing MUAPs were used as the input parameters in
classification procedures performed by ANN. The best diagnostic yield was obtained
for Daubechies4 wavelets. Although the 16 wavelet coefficients captured most of
the energy of the MUAP signals the results of classification were worse than those
provided by the time domain parameters (listed above) defined in [Pattichis and Elia,
1999] ( Sect. 4.3.2) .
The EMG decomposition based on wavelet transform was proposed in [Zennaro