EXPLORING THE DIFFERENCES IN SURFACE

ELECTROMYOGRAPHIC SIGNAL BETWEEN

MYOFASCIAL-PAIN AND NORMAL GROUPS

Feature Extraction through Wavelet Denoising and Decomposition

Ching-Fen Jiang

, Nan-Ying Yu

Department of Biomedical Engineering, I-Shou University, Kaohsiung, Taiwan

Department of Physical Therapy, I-Shou University, Kaohsiung, Taiwan

Yu Ching Lin

Department of Physical Medicine and Rehabilitation, National Cheng Kung University, Tainan, Taiwan

Keywords: Myofascial pain, Surface electromyography, Wavelet energy.

Abstract: Upper-back myofascial pain is an increasingly significant syndrome associated with frequent computer

using. However, the changes in neuromuscular functions incurred by myofascial pain are still under-

discovered. This study aims to discover the changes in neuromuscular function on the taut band through

signal analysis of surface electromyography. We first developed a fully automatic algorithm to detect the

duration of an epoch of muscle contraction. Following that, the features of epochs in both time-domain and

frequency-domain were extracted from the 13 patients to compare with the measurement from 13 normal

subjects. The higher contraction strength with lower median frequency found in the patient group is similar

to the reported changes with muscle fatigue. The signal was further analyzed by wavelet energy of 17 levels.

The result shows that the energy measured from the patients exceeds that from the normal group at the low

frequency band, suggesting that an increasing synchronization level of motor unit recruitment may cause the

drop in the median frequency and the increase in contraction strength.

1 INTRODUCTION

Nowadays due to the popularity of using computer

and increasing working stress, myofascial pain

(MFP) has been a common occupational hazard. The

number of people with this syndrome seeking

medical treatment is increasing abruptly. Although

there are various inferences for the etiology of MFP,

the investigation into any induced changes in

neuromuscular functions is rare.

The detected signal form surface

electromyography (SEMG) is called the interference

pattern (IP), which provides considerably more

diagnostic information than that of the motor unit

action potential (MUAP) along. The IP is commonly

used to predict the muscle force and evaluate the

muscular motor functions in several fields such as

rehabilitation, and sport and geriatric medicine. The

popularity of application of SEMG in clinics is due

to non-invasiveness. Past efforts to analyze SEMG

signals were mainly based on the feature extractions

in time (Fricton et al., 1985) or frequency domain

(Hagberg and Kvarnstrom, 1984) separately. These

methods do not take both time and frequency

variation into account in an optimal sense. However,

since the IP is comprised of the summation of

MUAP trains from all active motor units within the

surface electrode recording range; as a result of that,

the variations in MUAP shapes and sizes are

averaged. In addition, the SEMG signal is non-

stationary as its statistical properties change over

time and usually contaminated with random noises.

All these factors can lead to a loss of key motor

control information contained in the signal.

Therefore, the non-stationery nature of SEMG signal

associated with the large subject-dependent

variances in its parametric measures hinder

practitioners from interpreting their clinical findings.

203

Jiang C., Yu N. and Ching Lin Y..

EXPLORING THE DIFFERENCES IN SURFACE ELECTROMYOGRAPHIC SIGNAL BETWEEN MYOFASCIAL-PAIN AND NORMAL GROUPS -

Feature Extraction through Wavelet Denoising and Decomposition.

DOI: 10.5220/0003515402030206

In Proceedings of the International Conference on Signal Processing and Multimedia Applications (SIGMAP-2011), pages 203-206

ISBN: 978-989-8425-72-0

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

The wavelet transform (WT) is an efficient tool

for multi-resolution analysis of non-stationary and

fast transient signals. These properties make it

especially suitable to study the neurophysiological

signals. Numerous WT applications in biosignal

analysis have been proposed, including for EMG

analysis (Arikidis et al., 2002, Kumar et al., 2003).

In our previous study (Jiang and Kuo, 2008), we

have developed a wavelet denosing method that can

automatically detect the occurrence of SEMG

epochs and render more consistent and stable epoch

strength. Based on this denoising method, this study

further applies feature extraction and analysis of

SEMG signal in both time and frequency domain to

explore the changes in neuromuscular function with

MPF.

2 MATERIALS AND METHODS

2.1 SEMG Measurement

We recruited two groups of participants with the age

ranged from 30 to 50 years old. One was the patient

group and the other is normal group. Table 1

provides the descriptive information of the

participants. In order to make a consistent condition,

only the right-handed participants were selected.

The active electrodes MA-411 were attached on

the taut-band loci at the right side of the upper back

to measure the SEMG signal. The analogue signal

was amplified up to 3800 times and band-passed

(20Hz to 3,000Hz) by MA-411 and then digitized

with 5000 kHz sampling rate by instruNet 100 data

acquisition card and transferred to computer for

further analysis.

The participant conducted only five repetitive

trials. For each trial, participants lay down on their

stomach steadily for standby at the first three beats

and lifted both their arms toward the ceiling on the

4th beat with their maximal force, released the hold

on the 5th beat. One beat last one second, so each

trial last for 5 seconds.

Table 1: Descriptive information about participants.

Variable patient group normal group

Number 13 13

age (years) 41.2±6.2 39.5±7.3

weight (kg) 58.6±7.5 59.3±8.7

2.2 Denoising and Epoch Detection

The overall procedure to detect the occurrences of 1-

sec epoch for each contraction is summarized in

Figure 1. According to our previous comparative

study, we first applied Universal-soft denoising

method to yield the reconstructed signal with the

best signal quality. Following that, the 1

differentiation of the denoised signal was calculated

to detect the abrupt spikes, contained in each epoch.

Finally, the dominate spike, indicating the central

location of the epoch, was detected by using our

developed multi-resolution thresholding algorithm

based on statistical clustering process. The principle

of the algorithm is based on the existence of an

optimal threshold that should separate the population

into two groups with maximal between-class

variance. This concept was originally proposed by

Otsu for image segmentation (Otsu, 1979).

However, the distribution of the SEMG signal

derivatives is not like an image histogram with

definite discrete levels. To circumvent this problem

of indefinite derivatives, we developed a novel auto-

thresholding algorithm with multi-resolution

concept. The algorithm is described in detail in the

reference(Jiang and Kuo, 2008).

Once the central location of the epoch is

determined, then the duration can be spread

symmetrically toward both side of it within 1 sec.

(each side contains 2500 sampling points). The

following feature analysis was subject to the

detected epochs.

Figure 1: Wavelet processing.

2.3 SEMG Analysis

Three features were extracted from the SEMG

activity for group comparisons to examine the

SIGMAP 2011 - International Conference on Signal Processing and Multimedia Applications

204

Figure 2: Top panel shows the original SEMG signal from the muscle contraction at a constant pace and the corresponding

signal after denoising shown in the bottom panel.

difference between normal and MFP contraction.

They are described separately in detail as follows.

 Root mean square (RMS) of a SEMG epoch

is an index to evaluate the strength of the

corresponding muscle contraction. It can be

calculated as the summation of amplitude

square within an epoch according to

∫

dttmTtmRMS

2/12

])([/1})({

(1)

 Median frequency (MDF) is a common

parameter derived from the power spectrum

density (PSD) of SEMG signal to evaluate

the activity of muscle fiber recruitment and

conduction. The definition of MDF is the

frequency where the area of the PSD is

exactly half of total area of the PSD as

∫∫

∞

)(

)( dffPSDdffPSD

MDF

(2)

 Wavelet energy is a quantifier commonly

used to evaluate signal strength in a specific

frequency band through wavelet

decomposition. The wavelet energy (Ej) at

level j is defined as the summation of the

power of the “detail” coefficient (dj[m]).

])[( mdE

∑

(3)

We further made a plot of wavelet energy (E

)

versus decomposition level (j) and fitted the plot into

a curve using a non-linear regression model.

3 RESULTS AND DISCUSSIONS

The results are described from two aspects as

follows.

3.1 Denoised Signal and Detected

Epochs

The effect of the denoising process for the SEMG

signal is illustrated in Figure 2. It can be found that

the denoised signal still keeps the epoch location and

preserves the features.

An example of the detection of the occurrences

of SEMG epoch is given in Figure 3.

3.2 Group Comparison of SEMG

Features

The analysis in the time domain shows that the RMS

from the patient group is slightly higher than that of

normal group. The power-spectrum analysis shows

that the MDF from the patient group is lower than

that of normal group. Table 2 summarizes the results

of those analyses.

From the results of the comparative study, it is

evident that the SEMG signal from the MPF

participants tends to have greater RMS amplitudes

than that from the normal subjects. This result agrees

with the finding of the previous study (Fricton et al.,

1985, Hagberg and Kvarnstrom, 1984). The

decreasing median frequency observed in the power

spectral density of the SEMG from MPF subjects is

similar to a fall in median frequency during muscle

fatigue (Hagberg and Kvarnstrom, 1984).

The result of wavelet analysis (Figure 4) shows

an increasing energy gap between patient and

normal groups as the decomposition level increases,

especially beyond level 8.

Table 2: Comparison of the SEMG features between two

groups.

Variable patient group normal group

RMS (volt) 0.98±0.07 0.94±0.13

MDF (Hz) 966.6±28.7 975.8±20.7

EXPLORING THE DIFFERENCES IN SURFACE ELECTROMYOGRAPHIC SIGNAL BETWEEN

MYOFASCIAL-PAIN AND NORMAL GROUPS - Feature Extraction through Wavelet Denoising and Decomposition

205

Our finding in increasing energy difference in

the lower-frequency-band is in accord with the

finding of previous approach regarding muscle

fatigue (Kumar et al., 2003). The WT of a signal

provides a multiresolutional decomposition of the

signal for the analysis of signal components at

different scales in the time domain. The energy in

the higher decomposition level is contributed by the

signal in the coarser resolution or within the lower

frequency-band. In addition, the greater wavelet

energy derived from MPF SEMG signal than that

from the normal SEMG signal in the lower-

frequency-band could be a reflection of the drop in

the median frequency of the spectral analysis.

Figure 3: The detected center of the epoch for each

contraction is indicated as the green bar.

Figure 4: The trend of difference in the wavelet energy

between the MDF patients and the normal subjects across

the 17 frequency levels.

4 CONCLUSIONS

In recent years, Upper-back myofascial pain (MP) is

increasingly found to be associated with consistent

computer using. We postulate that MP could be due

to muscle fatigue caused by long-lasting computer

usage. However, the effect of MP on the muscle

function is unclear. Therefore in this study we tried

to use the wavelet energy to analyze differences in

SEMG signal between MP and normal groups.

Results either in the time domain or in the

frequency domain show similar changes found in

muscle fatigue. Wavelet analysis further explores

that these changes may be attributed to the

increasing difference towards the lower frequency-

band, as the result of the increasing synchronization

level of motor units recruitment. These finding may

suggest that the changes in neuromuscular function

associated with myofascial pain can be induced by a

long-term muscle fatigue.

REFERENCES

Arikidis, N. S., Abel, E. W. & Forster, A. 2002. Interscale

wavelet maximum-a fine to coarse algorithm for

wavelet analysis of the EMG interference pattern.

Biomedical Engineering, IEEE Transactions on, 49,

337-344.

Fricton, J., Auvinen, M., Dykstra, D. & Schiffman, E.

1985. Myofascial pain syndrome: electromyographic

changes associated with local twitch response.

Archives of physical medicine and rehabilitation, 66,

314-317.

Hagberg, M. & Kvarnstrom, S. 1984. Muscular endurance

and electromyographic fatigue in myofascial shoulder

pain. Archives of physical medicine and rehabilitation,

65, 522-525.

Jiang, C. F. & Kuo, S. L. 2008. Detection of occurrence of

motor unit action potential in the surface

electromyographic signal based on wavelet denoising.

International Journal of Electrical Engineering, 15,

161-168.

Kumar, D. K., Pah, N. D. & Bradley, A. 2003. Wavelet

analysis of surface electromyography to determine

muscle fatigue. IEEE Trans Neural Syst Rehabil Eng,

11, 400-406.

Otsu, N. 1979. A Threshold Selection Method from Gray-

Level Histograms. IEEE Transactions on Systems,

Man and Cybernetics,, SMC-9, 62-66.

SIGMAP 2011 - International Conference on Signal Processing and Multimedia Applications

206