Comparative Study of Compression Techniques Applied in Different

Biomedical Signals

A. A. Saraiva

2,3 a

, F. M. J. Castro

1 b

, Nator Junior C. Costa

1 c

, Jose Vigno M. Sousa

1 d

N. M. Fonseca Ferreira

4,5 e

, Antonio Valente

3,6 f

, Salviano Soares

7 g

UESPI-University of State Piaui, Piripiri, Brazil

School of Science and Technology, University of Tr

as-os-Montes and Alto Douro, Vila Real, Portugal

Department of Electrical Engineering, Institute of Engineering of Coimbra, Coimbra, Portugal

Knowledge Engineering and Decision-Support Research Center (GECAD) of the Institute of Engineering,

Polytechnic Institute of Porto, Portugal

NESC-TEC Technology and Science, Campus da FEUP, Rua Dr. Roberto Frias, 378,4200 - 465 Porto, Portugal

IEETA – UA and University of Tras-os-Montes and Alto Douro, Vila Real, Portugal

Keywords:

Compression and Signal and Electrooculography and Algorithms.

Abstract:

This paper aims to compare the compression of electro-oculographic signals, based on the (EOG) from MIT

/ BIH database, and the electromyographic signals, based on the (EMG) from MIT / BIH database, for that

purpose, two compression techniques that can be used in electro-oculograms and electromyograms was ap-

proached, the two techniques mentioned above, were, the discrete cosine transform and Fast Walsh Hadamard

Transform. For statistic the methods used was, the Mean squared error, mean absolute error, signal-to-noise

ratio and peak signal-to-noise ratio as well, and for results, the techniques and they performance on each tested

signal.

1 INTRODUCTION

With the constant development of computing, the

amount of information needed for humans inevitably

grows.

The volume of information carriers and the ca-

pacity of communication channels increase, but the

amount of information grows faster.

In this way the compression and decompression

procedure is a solution for a more rational use of the

storage and data transfer devices.

One of the medical applications is related to the

bio-medical signs, among them the ocular ones from

the electrooculography (EOG), and the muscular ones

from the electromyography (EMG).

https://orcid.org/0000-0002-3960-697X

https://orcid.org/0000-0002-7751-9455

https://orcid.org/0000-0001-5636-424X

https://orcid.org/0000-0002-5164-360X

https://orcid.org/0000-0002-2204-6339

https://orcid.org/0000-0002-5798-1298

https://orcid.org/0000-0001-5862-5706

Most modern EMG systems for algorithms for

data compression, capabilities for storing and trans-

mitting large data equipment via GSM / GPRS, Inter-

net and other communication channels for complete

telemedicine systems.

The problems about expansion of the signal

storage resources are being solved at the present mo-

ment, mainly due to the use of the technique of ”com-

pression” with losses. At the same time, the ability

to analyze data in ”full report” mode, which is acces-

sible to analog methods and in some cases important

information about rhythm disturbances and changes

not recognized by the microprocessor are lost.

Data compression is the process of detecting and

eliminating redundancies of a data set. In terms of

signal compression techniques, direct or transformed

methods were found in most of the literature. They

are characterized by not reaching the highest level of

compression, have no control over the quality of the

recovered signal.

In this paper, the efﬁciency of two compression

and decompression methods will be studied from

132

Saraiva, A., Castro, F., Costa, N., Sousa, J., Ferreira, N., Valente, A. and Soares, S.

Comparative Study of Compression Techniques Applied in Different Biomedical Signals.

DOI: 10.5220/0007350401320138

In Proceedings of the 12th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2019), pages 132-138

ISBN: 978-989-758-353-7

EOG, and EMG signs: FWHT and DCT. The data are

analyzed in terms of signal quality and compression

level and they come from the MIT/BIH database.

The document is divided into 7 sections, in which

section 2 and 3 is characterized by explanation of met-

hods used, followed by section 4 in which it shows the

metric comparison of the method. It follows the for-

mulation of the central algorithm applied in section 5

and the statistical method to verify reliability. The re-

sults after application of the proposal are presented in

section 6 and the conclusion in section 7.

2 ELECTROOCULOGRAPHY

Electrooculography (EOG) is a technique for measu-

ring the corneo-retinal standing potential that exists

between the front and the back of the human eye. The

resulting signal is called the electrooculogram. Pri-

mary applications are in ophthalmological diagnosis

and in recording eye movements. Unlike the electr-

oretinogram, the EOG does not measure response to

individual visual stimuli.

To measure eye movement, pairs of electrodes are

typically placed either above and below the eye or to

the left and right of the eye. If the eye moves from

center position toward one of the two electrodes, this

electrode ”sees” the positive side of the retina and the

opposite electrode ”sees” the negative side of the re-

tina. Consequently, a potential difference occurs bet-

ween the electrodes. Assuming that the resting poten-

tial is constant, the recorded potential is a measure of

the eye’s position (Brown et al., 2006).

Electro-oculography is used to record eye mo-

vements during electronystagmographic testing. It

is based on the corneoretinal potential (difference in

electrical charge between the cornea and the retina),

with the long axis of the eye acting as a dipole. Mo-

vements of the eye relative to the surface electrodes

placed around the eye produce an electrical signal that

corresponds to eye position. Recordings of eye mo-

vement are accurate to about 0.5 degree, but it is still

less sensitive than visual inspection, which can per-

ceive movements of about 0.1 degree. Therefore, vi-

sual inspection with Frenzel lenses is sometimes still

necessary to document nystagmus of low amplitude.

Another limitation of electro-oculography is that tor-

sional eye movements cannot be monitored. Again,

visual inspection with Frenzel lenses is sometimes

necessary to document torsional nystagmus(Saraiva

et al., 2018b).

Fortunately, new techniques have been developed

to provide greater accuracy and breadth for oculo-

motor testing. The most clinically useful technique

that has been developed is the infrared video elec-

tronystagmographic system. Here, the patient wears

goggles that illuminate the eyes with infrared light

(invisible to the patient), allowing a small video ca-

mera to pick up and project an image of the eyes onto

a monitor. This can also assess eye movement in ho-

rizontal, vertical, and torsional directions and is more

accurate than electro-oculography(das Chagas Fonte-

nele Marques Junior et al., 2018) (Schapira, 2006).

3 ELECTROMYOGRAPHY

The EMG is the process by which an examiner puts a

needle into a particular muscle and study the electrical

activity of that muscle, this electrical activities come

from the muscle itself no shocks are used to stimulate

the muscle, by that is possible to ﬁnd a muscle who

present a particular problem or disease(Weiss et al.,

2015).

The excitability of a muscle ﬁbers through neural

control represents a major factor in muscle physio-

logy, for that the EMG is a technique concerned with

the development recording and analysis of myoeletric

signals, the signals taken by this process are formed

by physiological variations in the state of muscle ﬁ-

ber membranes, by that some diseases and problems

can be detected if the variation don’t follow the nor-

mal patterns under minor exceptions(Konrad, 2005),

there’s another way to measure the EMG,is the neu-

rological EMG, by this way electrical shocks are used

to stimuli the muscle, but on this work, the kinesio-

logic EMG will be approached, on this type only the

natural response of the muscle are taken as object of

study then for that are used to take the signal.

4 DCT - DISCRETE COSINE

TRANSFORM

The DCT is very related to the Discrete Fourier Trans-

form (DFT), it can often reconstruct a precise se-

quence of only a few DCT coefﬁcients, this property

is very useful for applications that require data re-

duction, precisely the purpose of this work, to ex-

plore the reduction of data use in electrocardiogram,

(Nguyen et al., 2017). The DCT has four standard va-

riants, for an x-signal of size N and with the kronecker

δ, the transformations are deﬁned by the equations1,

2, 3 and 4 respectively.

y(k)=

√

∑

N−1

x(n)

√

1+ delta

+δ

√

1+δ

cos

(

N−1

(n−1)(k−1)

)

(1)

y(k)=

∑

n−1

x(n)

√

1=δ

cos

(

(2n−1)(k−1)

)

(2)

Comparative Study of Compression Techniques Applied in Different Biomedical Signals

133

y(k)=

∑

n−1

x(n)

√

1+δ

cos

(

(n−1)(2k−1)

)

(3)

y(k) =

∑

n=1

x(n)cos



(2n −1)(2k −1)



(4)

The series are indexed with n = 1 and k = 1 instead

of the usual n = 0 and k = 0.

On the equations, x is meaning the input array, y

are the DCT itself and n is equal to the length of the

transform, a positive integer scalar, with x and y being

vectors (they can be matrices), (Nguyen et al., 2017).

In his work, Swarnkar using the standlet transform

achieved better results compared to DCT and Wave-

let transform, being able to illustrate well its results

using data like SNR, also used in this work, CR and

Price Related Differential (PRD), (A.Swarnkar et al.,

2017).

A DCT expresses a series of ﬁnitely many data

points in terms of a sum of cosine functions. Oscil-

late at different frequencies. DCT has the applicati-

ons of solving partial differential equations, Chebys-

hev approximation, audio compression, (Raj and Ray,

2017).

5 FWHT- FAST WALSH

HADAMARD TRANSFORM

The WHT is a non-sinusoidal, orthogonal transfor-

mation technique that decomposes the signal into a

series of base functions, these base functions are cal-

led Walsh functions, which are rectangular and square

waves with values of -1 and 1. They are also known

as Hadamard, Walsh, or Walsh Fourier transform.

They are very useful in reducing the requirements

of storage, bandwidth and spectrum analysis. Like

Fast Fourier Transform (FFT) the WHT has a faster

version to Fast Walsh Hadamard Transform (FWHT),

which compared to FFT requires less storage space

and is faster to calculate, since it uses only real additi-

ons and subtractions, whereas the FFT uses complex

values.

Both the FWHT and the Inverse Fast Hadamard

Transform (IFWHT) are symmetric to each other and

use identical calculation processes (Saka et al., 2016).

For a signal x(t) of size N the FWHT and IFWHT

are deﬁned as follows:

y =

n−1

∑

i=0

xWAL(n, i) (5)

x =

n−1

∑

i=0

yWAL(n, i) (6)

Where i=0,1...,N-1 and WAL(n,i) are the Walsh

functions. Similar to the Cooley-Tukey algorithm for

the FFT, the N elements are decomposed into two sets

of N/2 elements, which are then combined using a

butterﬂy structure to form the FWHT, (Saka et al.,

2016).

6 STATISTICS

For the achievement of the quality of compressed and

reconstructed signal classiﬁcation, compared with the

original signal was used the Mean Squared Error

(MSE), MAE, SNR, PSNR.

6.1 MSE

The MSE is a signal ﬁdelity meter. The purpose of a

ﬁdelity meter is to compare two signals and provide

a quantitative score that describes the degree of simi-

larity or ﬁdelity and the level of error or distortion

between them, assuming that one of the signals is pri-

mitive and error-free while the other is distorted and

contaminated by errors (Saraiva et al., 2018c). The

MSE can be calculated as the equation 7, shows.

MSE =

m−1

∑

x=0

n−1

∑

y=0

I −K

(7)

Taking I, as a n predictions vector and K, as a vec-

tor of observed values of the variable being predicted.

6.2 MAE

The MAE is a ”scaled” measure, in this, it expresses

a precise prediction model of errors in units of the

variable of interest, as well as the MSE, the smaller

the value, the greater the ﬁdelity signal.

MAE =

∑

i=1

| (8)

Reconstructed to the original signal, the MAE can be

calculated assuming that there are n sample model er-

rors e calculated as (e

,i=1,2,...,n), (Willmott and Mat-

suura, 2008).

6.3 SNR

SNR is the rate between signal and noise, in engi-

neering and science the SNR is the measurement that

compares the level of the desired signal with the back-

ground noise level.

Mathematically the SNR is the intensity quotient

of a signal measured in a Region of Interest (ROI) and

BIOSIGNALS 2019 - 12th International Conference on Bio-inspired Systems and Signal Processing

134

the standard deviation of the signal intensity in an area

outside the imaged object’s anatomy.

SNR = Log

∑

n=0

(n)

∑

n=0

(n)

(9)

The SNR can be calculated assuming V

(n) as the re-

constructed signal, V (n) as the original Signal and the

(n) as the deformation of the reconstructed Signal,

(Princy et al., 2015).

6.4 PSNR

The PSNR is a parameter used to quantify the signal

quality, it is also used as a benchmark, of the level

of similarity between the reconstructed signal and the

original signal.

The higher the PSNR value, the better the signal

quality. And can be calculated as the Equation 10,

shows.

PSNR = 10 log



MAX

MSE



= 20 log

MAX

MSE

(10)

7 METHODOLOGY

In the Fig.1, was exempliﬁed the step by step of this

work in a simple block diagram, the example shows

the process of comparison between the two compres-

sion techniques, FWHT and DCT.

Figure 1: Methodology Block diagram.

First, the signal is taken from a source in this case

the MIT-Database Chosen from the database of EOG

and EMG signal, sometimes the signal taken has noise

so the ﬁltering was necessary, then, the signal who

was input is loaded to the software , then the data is

taken from it the data will be used to a comparison

later, before that, the compression technique is app-

lied to the signal, if the FWHT is applied, the signal

is repeated 8 times, and then random noise is added

before the compression, on the DCT case it won’t be

necessary to repeat the signal.

After compressed the signal, we get the data from

the input signal, and for that the inverse transform

of the respective compression technique is added,

getting with that, a reconstructed signal, the recon-

structed signals may vary depending of the compres-

sion technique, in this case, the FWHT has as the re-

constructed signal 8x bigger than DCT reconstructed

signal because of the repetition used early.

Before that, the data of the reconstructed signal

was compared to the size of the original signal of the

respective technique, then after the comparison bet-

ween the original and the reconstructed signal in mind

that the FWHT reconstructed signal must be compa-

red with the 8x repeated Original signal because of it

size, and the DCT reconstructed signal must be com-

pared with the original signal as well because of it

size keeping in mind that the signal behave on dif-

ferent ways according to the compression and recon-

struction technique(Saraiva et al., 2018d).

To do so, the statistic used are the MSE, MAE,

SNR and PSNR with them, the signal quality and

similarity can be tested, according to the results gi-

ven by the mentioned statistic methods the MSE and

MAE statistic methods, test the signal ﬁdelity of the

reconstructed signal compared to the original signal

for both FWHT or DCT keeping in mind that each

technique has its own size of original and recon-

structed signal, on the FWHT the signal was 8 times

repeated to get the coefﬁcients WHT, also the PSNR

and the SNR test the quality of the reconstructed sig-

nal compared to the original signal having as well the

cautions to each Technique used on this work as men-

tioned above (Saraiva et al., 2018a) (Saraiva et al., ).

8 RESULTS

After the methodology used the signals were graphi-

cally represented to increase the understanding of this

work the next images will show the comparison of the

reconstructed signals and the original ones then show

the statistic results to discuss the work before the con-

clusion.

In the Fig 2(a), are showed a 8x repeated EOG

and the WHT coefﬁcients of it, with that its possible

to obtains the Walsh functions using the FWHT, each

sample uses on the EOG example, was taken from the

right eye readings the same is shown on the Fig 2(b),

but with the EMG, also is showed its WHT coefﬁ-

cients.

In the Fig 3(c) and Fig 3(d), is showed graphi-

cally the original and the reconstructed by IDCT sig-

Comparative Study of Compression Techniques Applied in Different Biomedical Signals

135

(a) Repeated EOG and WHT coefﬁ-

cients

(b) Repeated EMG and WHT coefﬁ-

cients

Figure 2.

(d) IDCT reconstructed EMG signal

Figure 3.

nals, showed in the graphic in different colors, exem-

pliﬁed on the legend each sample the original EOG,

original EMG and the reconstructed are nearly simi-

lar, of course, taking the graphic as basis, that occurs

because the DCT have a small loss rate, keeping in

mind, the low difference seen in the graphic above.

In the Fig 4(e) and Fig 4(f), are showed the result

obtained after the signals passes through the FWHT,

being converted into WHT coefﬁcients, or Walsh

(e) IFWHT reconstructed EOG sig-

nal

(f) IFWHT reconstructed EOG sig-

nal

Figure 4.

functions, with that its possible to use the IFWHT to

obtain the reconstructed signals.

Table 1: Discrete cosine Transform.

Transform Test Value

DCT PSNR 24.876553

DCT SNR 19.1117

DCT MAE 0.005583

DCT MSE 0.000043

On the table 1, are exempliﬁed the results of the

statistic methods applied on the DCT EOG recon-

structed signal.

Table 2: Fast Walsh Hadamard Transform.

Transform Test Value

FWHT PSNR 13.90701

FWHT SNR -4.374118

FWHT MAE 0.078406

FWHT MSE 0.009601

On the table 2, are showed the results of the sta-

tistic methods applied on IFWHT EMG reconstructed

signal.

On the table 1, are exempliﬁed the results of the

statistic methods applied on the DCT reconstructed

EMG signal.

On the table 2, are showed the results of the sta-

tistic methods applied on IFWHT EMG reconstructed

signal.

BIOSIGNALS 2019 - 12th International Conference on Bio-inspired Systems and Signal Processing

136

Table 3: Discrete cosine Transform.

Transform Test Value

DCT PSNR 39.214499

DCT SNR 27.158767

DCT MAE 0.000299

DCT MSE 0.000000

Table 4: Fast Walsh Hadamard Transform.

Transform Test Value

FWHT PSNR -5.723057

FWHT SNR -17.778790

FWHT MAE 0.052583

FWHT MSE 0.004576

9 CONCLUSION

In a direct comparison between the above-mentioned

transforms, the FWHT obtained advantage because

the reconstructed signal approached the original sig-

nal and its compression was much more efﬁcient.

DCT has proven itself to be effective with a very

precise reconstruction of the compressed EOG in ad-

dition to the need for signal repetition. as seen in the

images above, the graph of the DCT is relatively close

to that of the original EOG taking as example the er-

rors that were the lowest compared to the FWHT.

The DCT was proven itself most effective on a di-

rect comparison in the EMG case for that the DCT is

the best transform between the analyzed ones to EMG

With greater advantage than on the EOG.

keeping in mind the results obtained, we see that

the compression techniques discussed have their dis-

tinct particulars in certain aspects, therefore, we must

always take into account that in some cases the results

may not be identical.

for future work, it is interesting to classify a lar-

ger variety of transforms and their performances as

the DWT (Discrete Wavelet Transform) into a wider

range of medical signals, such as ECG (Electrocardi-

ogram) and EEG (Electroencephalogram).

ACKNOWLEDGMENTS

This work is ﬁnanced by National Funds through the

FCT - Fundac¸

ao para a Ci

encia e a Tecnologia (Por-

tuguese Foundation for Science and Technology) as

part of project UID/EEA/00760/2019.

REFERENCES

A.Swarnkar, Kumar, R., Kumar, A., and Khanna, P. (2017).

Performance of different threshold function for ecg

compression using slantlet transform. in signal pro-

cessing and integrated networks (spin). volume 37,

pages 375–379.

Brown, M., Marmor, M., Zrenner, E., Brigell, M., Bach,

M., et al. (2006). Iscev standard for clinical electro-

oculography (eog) 2006. Documenta ophthalmolo-

gica, 113(3):205–212.

das Chagas Fontenele Marques Junior, F., Saraiva, A. A.,

Sousa, J. V. M., Ferreira, N. F., and Valente, A. (2018).

Manipulation of bioinspiration robot with gesture re-

cognition through fractional calculus. In IEEE LARS

2018 – 15th Latin American Robotics Symposium,

SBR 2018 – 6th Brazilian Robotics Symposium, 2018.

IEEE.

Konrad, P. (2005). The abc of emg. A practical introduction

to kinesiological electromyography, 1:30–35.

Nguyen, B., Nguyen, D., Ma, W., and Tran, D. (2017). In-

vestigating the possibility of applying eeg lossy com-

pression to eeg-based user authentication. In Neural

Networks (IJCNN), 2017 International Joint Confe-

rence on, pages 79–85. IEEE.

Princy, R., Thamarai, P., and Karthik, B. (2015). Denoi-

sing eeg signal using wavelet transform. International

Journal of Advanced Research in Computer Engineer-

ing & Technology, 3.

Raj, S. and Ray, K. C. (2017). Ecg signal analysis using dct-

based dost and pso optimized svm. volume 66, pages

470–478.

Saka, K., Aydemir,

O., and

Ozt

urk, M. (2016). Classiﬁ-

cation of eeg signals recorded during right/left hand

movement imagery using fast walsh hadamard trans-

form based features. In Telecommunications and Sig-

nal Processing (TSP), 2016 39th International Confe-

rence on, pages 413–416. IEEE.

Saraiva, A. A., Castro, F. M. J., Sousa, J. V. M., Valente,

A., and N M Fonseca, F. (2018a). Comparative study

between the walsh hadamard transform and discrete

cosine transform. In 7th International Conference on

Advanced Technologies.

Saraiva, A. A., Costa, N. J. C., Sousa, J. M., Araujo, T. P. D.,

Ferreira, N. M. F., and Valente, A. (2018b). Scalable

task cleanup assignment for multi-agents. Memorias

de Congresos UTP, 1(1):439–446.

Saraiva, A. A., Ferreira, N. M. F., Soares, S. F., Reis, M.

J. C. S., and Valente, A. Filtering of cardiac sig-

nals with mathematical morphology for qrs detection.

In Proceedings of ICAT’18, 7th International Confe-

rence on Advanced Technologies, pp. 1008-1017, 28

April-1 May, Antalya, Turkey, ISBN 978-605-68537-

1-5, 2018. IEEE.

Saraiva, A. A., Ferreira, N. M. F., and Valente, A. (2018c).

New bioinspired ﬁlter of dicom images. In Procee-

dings of the 11th International Joint Conference on

Biomedical Engineering Systems and Technologies -

Volume 1: BIODEVICES,, pages 258–265. INSTICC,

SciTePress.

Comparative Study of Compression Techniques Applied in Different Biomedical Signals

137

Saraiva, A. A., Nogueira, A. T., Ferreira, N. F., and Valente,

A. (2018d). Application of virtual reality for the tre-

atment of strabismus and amblyopia. In 2018 IEEE

6th International Conference on Serious Games and

Applications for Health (SeGAH), pages 1–7. IEEE.

Schapira, A. H. (2006). Neurology and clinical neuros-

cience, volume 1. Elsevier Health Sciences.

Weiss, L. D., Weiss, J. M., and Silver, J. K. (2015). Easy

EMG E-Book: A Guide to Performing Nerve Con-

duction Studies and Electromyography. Elsevier He-

alth Sciences.

Willmott, C. J. and Matsuura, K. (2008). Advantages of the

mean absolute error (mae) over the root mean square

error (rmse) in assessing average model performance.

In Climate research. JSTOR.

BIOSIGNALS 2019 - 12th International Conference on Bio-inspired Systems and Signal Processing

138