Predicting Muscle Fatigue via Electromyography and a Comparative Analysis
General Problem Description
The main objective of the present research was to develop, test, and validate
models for the prediction of muscle fatigue associated with sustained muscle
contraction. An experimental study was conducted to study the effects of
heavy isometric loading (maximum and 80% of the maximum) on the recorded
EMG. Also, the effect of electrode orientation on the detection of muscle
fatigue during heavy isometric loading was investigated. The recorded EMG
signals were analyzed using MATLAB numeric computation software and its
signal processing toolbox developed by the MATH WORKS, Inc. Both time and
frequency domain analyses were conducted.
Time domain analysis
In this study, the full wave rectified integral (FWRI) and the root
mean square (RMS) values were estimated. For a discrete signal which consists
of N equispaced samples x(n), n=1 to N these measures are given algebraically
as
xFWRI = sum{ ( |x(n)| + |x(n+1)|) / 2),
n = 1..N-1} / (N-1),
and, xRMS =
sqrt(sum{x(n)2, n = 1..N}/N).
Frequency domain analysis
In the frequency domain analysis, the estimated power spectrum and
its characteristic fractile frequencies are calculated. The estimated power
spectrum , also known as the periodogram, is calculated as the modulus
squared Fourier transform.
The characteristic frequencies used in this study are 1, 5, 10, 25,
50, 75, 90, 95, 99 fractile frequency, and the peak frequency. The p-th
fractile frequency fp, analogous to the statistical definition
of fractile, is defined as the frequency for which
(please note that "int" indicates the integration symbol)
int{Gs(f)df, f = 0.. fp } / S = p
holds; where Gs(f) is the one-sided power spectrum of s(t)
and S is defined as
S = int{Gs(f)df, f = 0..infinity}.
Experimental Procedure
Eighteen healthy male subjects with no history of musculoskeletal injuries
participated in this study. All subjects were selected on a voluntary basis
from the student population or having sedentary life style. They represented
a wide spectrum of body weights, heights, age, and muscle strengths. They
ranged in age between 22 and 40 years with a mean value of 27.2 years.
Subjects weights ranged from 53.2 kg to 105.9 kg (117 to 233 lbs) with
a mean value of 75.86 kg (166.89 lbs). Heights ranged from 160 cm to 187.5
cm (5' 4" to 6' 3") with a mean value of 172.5 cm (5' 9"). All subjects
volunteered to participate in this study and were informed about the experi-ment
in advance.
All subjects were required
to perform a static muscle effort corresponding to a predetermined load
level. The load was applied to permit static contraction of the biceps
brachii muscle. The load was placed in the dominant hand of each subject
with the upper arm hanging freely in a neutral adducted position to the
side of the body. The forearm was flexed at 90o at the elbow
joint. The wrist was maintained in a straight neutral position with hand
supinated to support the load. The load consisted of a bar and two balanced
weights attached to both sides of the bar. Two levels of loading were studied.
These loads were set to the maximum amount of weight the individual can
hold for a few seconds (3-5 seconds) and 80% of the maximum weight. The
maximum weight was deter-mined on a separate day prior to the experimental
sessions. Subjects were instructed to hold the weight, as described earlier,
as long as possible. EMG was recorded from the biceps brachii muscle using
two sets of electrodes simultaneously. One set was placed along the muscle
fibers and the other across the muscle fibers. Electrodes used in this
experiment were Beckman type, 11 mm silver/silver chloride surface electrodes.
The EMG signal was recorded using an R611 Multichannel Sen-sormedics Dynograph
via a type 9853A voltage/pulse/pressure coupler. The amplifier gain was
adjusted to allow full utilization of the dynamic range of the A/D converter
(+ 10 volts). A sampling rate of 512 Hz
was used to digitize the EMG signal
using a 12 bit A/D converter model DT 2801-A.
EMG was recorded from the
onset of the load under investigation until the subject could not hold
the load anymore. Analysis of EMG was conducted as described earlier. The
window size used in the analysis is selected to be 512 m second based on
the results of the first experiment. EMG parameters were estimated for
the first and last window in the recorded signal. Also, EMG parameters
were calculated at fixed periods of time as a percentage of the total time
an individual was able to maintain the task of holding the load. The center
of the EMG window used in the analysis is set at the selected fixed periods
of time. These periods were selected at 5% through 95% of the total time
with an increment of 5%.
Analysis
Several statistical analyses were conducted to achieve the objectives of
the study. These analyses were performed in series to study the factors
1 and 2 below, leading up to the discriminant function developed for the
classification problem (3 below).
1. The effects of load, and electrode orientation
on the estimated EMG indices in both time and frequency domains.
The results obtained indicated that the time domain parameters did not
change significantly for the all interactions and main effects of the three
independent variables (load, electrode orientation and muscle condition
of rest or fatigue). The frequency domain parameters were significantly
affected by the main effects of electrode orientation and muscle condition.
The load had no significant effect on these parameters. The effect of electrode
orientation on the haracteristic frequencies used in this investigation
was more pronounced for the lower frequencies of the spectrum. Electrodes
placed across the muscle fibers showed lower fractile frequencies compared
to electrodes along the muscle fibers. The effect of electrode orientation
was only significant for the lower fractiles with the exception of the
99th fractile (peak frequency, 1, 5, 10, 25, and 99 fractile).
2. The effects of muscle condition,
load, and electrode orientation on the EMG indices.
The first and last window of the recorded EMG signals were used to represent
the muscle at a resting condition and the fatigue state respectively. The
EMG indices were the dependent variables. The independent variables were
the muscle condition (rest or fatigue), load, and electrode orientation.
Two separate statistical analyses were conducted to investigate the estimated
parameters in both time and frequency domains. The effect of muscle fatigue
was significant for all the characteristic frequencies used. A significant
shift toward lower frequencies was observed, however the amount of shift
in these frequencies was higher for the submaximum load. Also, it is worth
noting that the shift in these frequencies was not linear across the spectrum.
Therefore, monitoring a single characteristic frequency may not be adequate
for the quantification of the spectrum shift.
3. Classifying the
muscle state (either rest or fatigue) using estimated EMG parameters.
Based on the results obtained in steps #1 and #2, only frequency domain
parameters were used in the development of the discriminant function.
A Comparative Analysis of Our Method with
Some Other Methods
The Data Used in the Development and Validation of the Discriminant Functions
Related Publications by Our Group
To request reprints please send
email to: vangelis@imse.lsu.edu
For the descriptions of the abstracts of most of these and other papers,
please refer to our main home web page at:
http://www.imse.lsu.edu/vangelis.
Please note that once in a single paper page, to come back to this
section you need to hit the "BACK" key. Otherwise, you may access the refereed
journal papers page (of our central home page and not this one).
Torvik, V.I., E. Triantaphyllou, T.W. Liao, and S.M. Waly,
(1999),
"Predicting Muscle Fatique via Electromyography: A
Comparative Study,"
Proceedings of the 25-th Inter'l Conference
on Computers and Industrial Engineering,
New Orleans, LA, March, pp. 277-280.
Deshpande, A.S., and E. Triantaphyllou, (1998), "A Greedy Randomized
Adaptive Search Procedure (GRASP) for Inferring Logical Clauses from Examples
in Polynomial Time and some Extensions," Mathematical and Computer
Modelling, Vol. 27, No. 1, pp. 75-99.
Visit Dr. Triantaphyllou's homepage.