Наука

Поиск

Method to reduce the effect of miagrafic and sensory noise with isolating the isoline on ECG signal

Скачать в формате pdf

MATEC Web of Conferences 132, 05017 (2017) DOI: 10.1051/matecconf/201713205017

DTS-2017

License 4.0 (http://creativecommons.org/licenses/by/4.0/).

Corresponding author: sea.sea@mail.ru

Method to reduce the effect of miagrafic and sensory noise with

isolating the isoline on ECG signal

Evgeny Semenishchev

1,*

, Dmitry Chernyshov

, Ilya Svirin

Don state Technical University, 344000, Rostov-on-Don, Russian Federation

CJSC Nordavind, 117545, Moscow, Russian Federation

Abstract: This paper deals with an approach to the analysis of ECG data, which allows to remove the

noise component, to preserve the peaks characterizing the work of the heart, and to identificate floating

isoline.

1 Introduction

Processes automation is one of the priorities of the

modern world. Constant modernization of equipment

requires large investments, as well as retraining of

personnel, installation works and maintenance of new

systems. In this connection, the most promising direction

is the automation of data processing, which allows to

reduce the number of personnel, and to improve the

overall efficiency of the control system. One of the

leading areas of data analysis is the development of the

digital signals processing theory, based on the analysis

of initial data.

The use of digital signal processing methods has

found wide application [1]:

- in automation and control systems

- in modern antenna systems

- in the study of biomechanical parameters, biometric

data collection systems located directly on the object

under study

- in modern systems of computer vision and

automatic processing of two-dimensional signals

- in economics and sociology in the study of trends;

- in information-measuring systems;

- in computer technology to increase accuracy of

analog-digital transformation

Data, received with ECG, in addition to the useful

information on the propagation of electrical impulses in

the cardiac muscle, contains the noise component, which

has a different source. Analog-digital convertion also

adds random distortions to signal [2].

The main task in the signal processing is the

separation of the useful component and attenuation of

the noise component. In practice, a the mean-square

error minimization criterion or the criterion of mean-

absolute deviation is used to determine the processing

quality. Each of this methods has advantages and

limitations of use depending on the task and a priori

information on the components of the input signal. In

this connection, digital signal processing based on the

objective function of the combined criteria becomes

relevant objective. Of particular interest is the use of

multicriterial methods for processing digital signals, that

are represented by a single implementation with a

limited amount of a priori information about the useful

component function and statistical noise characteristics

[3].

Figure 1 contains an algorithm for obtaining

estimates using multi-criteria signal smoothing methods,

based on minimizing the objective functions:

 

















2)(),...,,(

kkk

kkn

sssYssss



when



– adjusting factor.

When implementing the considered smoothing

methods using machine modeling, in the form of

programs, regions of values of the coefficients were

obtained





. ECG data processing requires

real time analysis. It is done by finding estimates of the

objective function in window

, and applying sliding

window

for all values of the input signal. Here

size of the window and

is step of sliding window.

The process of obtaining estimates in a sliding

window is performed by parallel processing of the initial

values of the multicriteria objective function in the

window

, with various processing parameters



. The

transition between estimates obtained with different

parameters



is accomplished by the condition:

 

1 12

2 12

() () ()

k kk

s ss p

 

















Here

)(



)(



are input realization

estimates, received with parameters

MATEC Web of Conferences 132, 05017 (2017) DOI: 10.1051/matecconf/201713205017

DTS-2017

)(

1 ош





)(

2 ош





is a threshold value, determined

experimentally with the RMS deviation of the additive

noise component

2.0





and equals



[3].

Fig. 1. Algorithm for obtaining estimates using multi-criteria

signal smoothing methods.

The need to solve the isoline detection problem is

due to the requirement to adjust the monitoring system to

locate the report points recorded by the ECG device

within the minimum error range of the analog-to-digital

converter (ADC). To solve the problem of finding the

midline, the algorithm presented in Figure 2 was

developed.

Algorithm is shown on figure 2 and implemented as

follows:

At the first step, the noise component, possibly with

nonuniform distribution law is supressed. It is performed

in order to exclude its influence on the middle line

searching method.

In the second step, the R peak is suppressed using the

R-R processing window.

Fig. 2. Isoline searching algorithm.

At the final stage, parameters

)(





ош

are fixed, and

the entire set is processed. The choice of the filtering

method parameter is done using the results shown in

Figure 3.

Fig. 3. A graph of the choice of the smoothing parameter for

isoline selection method

)(





ош

Figure 3. - Example of the ECG isoline detection (1 -

normalized ECG model, 2 - normalized ECG model with

pathology, 3 - isoline).

The processing example shown in figure 3 is

obtained with fixed parameters of the middle line search

algorithm. The parameter of the processing window

is equal to the value of the R-R interval, the window step

is fixed and is equal to half the processing window



, the filtering parameter





, i.e. the priority

of the second summand of the objective function is eight

times larger than the first.

MATEC Web of Conferences 132, 05017 (2017) DOI: 10.1051/matecconf/201713205017

DTS-2017

)(

1 ош





)(

2 ош





is a threshold value, determined

experimentally with the RMS deviation of the additive

noise component

2.0





and equals



[3].

Fig. 1. Algorithm for obtaining estimates using multi-criteria

signal smoothing methods.

The need to solve the isoline detection problem is

due to the requirement to adjust the monitoring system to

locate the report points recorded by the ECG device

within the minimum error range of the analog-to-digital

converter (ADC). To solve the problem of finding the

midline, the algorithm presented in Figure 2 was

developed.

Algorithm is shown on figure 2 and implemented as

follows:

At the first step, the noise component, possibly with

nonuniform distribution law is supressed. It is performed

in order to exclude its influence on the middle line

searching method.

In the second step, the R peak is suppressed using the

R-R processing window.

Fig. 2. Isoline searching algorithm.

At the final stage, parameters

)(





ош

are fixed, and

the entire set is processed. The choice of the filtering

method parameter is done using the results shown in

Figure 3.

Fig. 3. A graph of the choice of the smoothing parameter for

isoline selection method

)(





ош

Figure 3. - Example of the ECG isoline detection (1 -

normalized ECG model, 2 - normalized ECG model with

pathology, 3 - isoline).

The processing example shown in figure 3 is

obtained with fixed parameters of the middle line search

algorithm. The parameter of the processing window

is equal to the value of the R-R interval, the window step

is fixed and is equal to half the processing window



, the filtering parameter





, i.e. the priority

of the second summand of the objective function is eight

times larger than the first.

2 Method of searching pathologies for

ECG analysis

For detect pathologies by analyzing the

electrocardiogram obtained by an automated mobile

system and stored on a mobile device. We will develop a

multi-criteria approach to the preliminary detection.

Diagnosis by the means of the developed mathematical

models of both a useful (ideal) ECG and possible

pathologies.

As the first criteria, we will use the square

measurement of the discrepancy between the developed

mathematical model

kmatkmat

sts



)(

and the

established cardiogram





kkmat

Ys 

As the second criteria, we will use a function

representing the normalized correlation coefficient:









 









kkkk

YYgg

where:

- model o.

Like a third criteria we will use resulting index,

frequency area.









ikfft



Resulting multi-criteria function will have the form:













































































































  

 







,...,

ikfft

kkmatn

Yssss







here:







- parameters that establish weight for each

of the criteria,

- set size,

fft

- frequency index,

- imaginary unit.

Each parameters of the criteria are established during

the preparation phase using the experience of specialists

in ECG analysis.

Figure 4 shows the pathology search algorithm for

ECG analysis.

Start

Searching the intersection

points of zero

Normalization of the

input values

Determining the range of

confidence intervals

Determining intervals and

detection areas

Upload

database

the pathology

Calculation of correlation

coefficients

Calculation of correlation

coefficients

Transition to the

frequency domain

Indicating parameters

calculation

Indicating parameters

calculation

Pathology detection

Pathology detection Pathology detection

Decision on the

pathology

Saving results

yes

End

Fig. 4. Pathology search algorithm for ECG analysis.

The algorithm presented in Figure 4 is implemented

as follows:

In the first step, the areas of intersection in

established points of the cardiogram report with the

isoline in the zones confined to confidence intervals are

searched. At the same time, the input values are

normalized in the range of R-R intervals.

In the next step, we make the determination of 5

peaks: P, Q, R, S, T and segments ST and R-R, the QRS

complex. The determination is made according to the

standard ranges [4] shown in Figure 5

In the third step, we use the ranges defined in the

previous step to select intervals. The values for each

selected range are normalized.

At the fifth step, the correlation coefficients are

calculated for each pathology and for each examination

area. Processing is performed in the sliding window. At

the same time, we perform the transition to the frequency

domain and the calculation of the main central moments

for the normalized indicating parameters. The

calculations of the central points are made using the

analysis of the differential coefficient.

MATEC Web of Conferences 132, 05017 (2017) DOI: 10.1051/matecconf/201713205017

DTS-2017

Fig. 5. The main complex of ECG.

On the sixth step the obtained frequency coefficients

are compared with a mathematical model of each of the

pathologies and the excess of a threshold value is

detected. The calculation of the established threshold is

also made by analyzing the divergence of the central

moments and correlation coefficients. Pathology is

considered detected if more than two indicators coincide

and in case of repeated confirmation notification

decision is made. In the case of coincidence of all

criteria for ECG analysis and their coincidence with the

investigated pathology, the patient is notified [5].

The implementation of the pathology

search methods for the analysis of

electrocardiograms

The first step is a searching for intersection of the fixed

points of the cardiogram report with a baseline in that

zones, which are limited by confidence intervals. We'll

use a value of two quantization levels as confidence limit

[6]. The device uses 10 bits analog-to-digital convertor

or ADC and the range of standard QRS complex

amplitudes for different areas of data sources should not

exceed 22 mm and 25 mm for adults. This limit specifies

the monitoring interval, which is limited in the range of

5mV. It means that the device indicates the intersection

of the baseline when fixed values exceed the confidence

intervals which are limited by two quantization levels

and it's equal to ±0,02 mV. We will impose restrictions

in the form of derivative of the change in direction for

the obtained values. It will be used the criterion of

exceeding the measurement sequence over the baseline

for more than 3 reports as a simple pike detector. The

restrictions on the excess of three reports are empirical,

they are based on the minimum possible duration of

restriction of the QRS complex. As a rule, this QRS

complex lasts 0,08 seconds. For children under the age

of 5 it lasts 0,09, and for children over 10 years – 0,10.

When the QRS complex lasts more than 0,11 for adults,

it is said of pathology and suggests ventricular

hypertrophy or ventricular blockade. We get the

minimum duration of QRS complex, which doesn’t

exceed 500 reports at a sampling frequency of the device

limited to 5 kHz. This article introduced a restriction of

the range of fixed changes, which is equal to 100 reports.

It will be made a decision about the intersection of the

baseline during these reports.

We will search for intervals of consecution of the

pikes R-R according to the following procedure: we will

use the complex approach in which the maximum value

will be determined as the main condition for detecting

the R wave, as secondary and confirming conditions we

will also use the relative time and sequence of

intersection intervals of the baselines corresponding to P

and Q and the following ST pikes, as well as their

relative intervals of consecution. We can see the ECG

duration model at the pic.6. Along with the R-R pikes, it

is also possible to detect the remaining pikes and their

ranges according to the data, which was obtained in the

previous stages.

Fig. 6. The duration model ECG within normal limits.

The input values are normalized in the range of P-P

intervals in the next step. The normalization process

consists of fixing the maximum values at a given interval

and dividing all values by a given value. Normalization

of the values allows not to consider the scaling factor for

the subsequent analysis and detection of pathologies. At

the third stage, we produce the allocation of pike and

segment intervals. We produce normalization of the

values in the area of each obtained intervals. We define

the group of central moments, such as mathematical

expectation:





YMv

)(

dispersion









)()(

kkk

YMYMYD 

skewness ratio









 

)(

YMYM







0,10

0,08

0,18

–

0,35

–

MATEC Web of Conferences 132, 05017 (2017) DOI: 10.1051/matecconf/201713205017

DTS-2017

Fig. 5. The main complex of ECG.

On the sixth step the obtained frequency coefficients

are compared with a mathematical model of each of the

pathologies and the excess of a threshold value is

detected. The calculation of the established threshold is

also made by analyzing the divergence of the central

moments and correlation coefficients. Pathology is

considered detected if more than two indicators coincide

and in case of repeated confirmation notification

decision is made. In the case of coincidence of all

criteria for ECG analysis and their coincidence with the

investigated pathology, the patient is notified [5].

The implementation of the pathology

search methods for the analysis of

electrocardiograms

The first step is a searching for intersection of the fixed

points of the cardiogram report with a baseline in that

zones, which are limited by confidence intervals. We'll

use a value of two quantization levels as confidence limit

[6]. The device uses 10 bits analog-to-digital convertor

or ADC and the range of standard QRS complex

amplitudes for different areas of data sources should not

exceed 22 mm and 25 mm for adults. This limit specifies

the monitoring interval, which is limited in the range of

5mV. It means that the device indicates the intersection

of the baseline when fixed values exceed the confidence

intervals which are limited by two quantization levels

and it's equal to ±0,02 mV. We will impose restrictions

in the form of derivative of the change in direction for

the obtained values. It will be used the criterion of

exceeding the measurement sequence over the baseline

for more than 3 reports as a simple pike detector. The

restrictions on the excess of three reports are empirical,

they are based on the minimum possible duration of

restriction of the QRS complex. As a rule, this QRS

complex lasts 0,08 seconds. For children under the age

of 5 it lasts 0,09, and for children over 10 years – 0,10.

When the QRS complex lasts more than 0,11 for adults,

it is said of pathology and suggests ventricular

hypertrophy or ventricular blockade. We get the

minimum duration of QRS complex, which doesn’t

exceed 500 reports at a sampling frequency of the device

limited to 5 kHz. This article introduced a restriction of

the range of fixed changes, which is equal to 100 reports.

It will be made a decision about the intersection of the

baseline during these reports.

We will search for intervals of consecution of the

pikes R-R according to the following procedure: we will

use the complex approach in which the maximum value

will be determined as the main condition for detecting

the R wave, as secondary and confirming conditions we

will also use the relative time and sequence of

intersection intervals of the baselines corresponding to P

and Q and the following ST pikes, as well as their

relative intervals of consecution. We can see the ECG

duration model at the pic.6. Along with the R-R pikes, it

is also possible to detect the remaining pikes and their

ranges according to the data, which was obtained in the

previous stages.

Fig. 6. The duration model ECG within normal limits.

The input values are normalized in the range of P-P

intervals in the next step. The normalization process

consists of fixing the maximum values at a given interval

and dividing all values by a given value. Normalization

of the values allows not to consider the scaling factor for

the subsequent analysis and detection of pathologies. At

the third stage, we produce the allocation of pike and

segment intervals. We produce normalization of the

values in the area of each obtained intervals. We define

the group of central moments, such as mathematical

expectation:





YMv

)(

dispersion









)()(

kkk

YMYMYD 

skewness ratio









 

)(

YMYM







0,10

0,08

0,18

–

0,35

–

index of kurtosis









 

)(





YMYM



The calculation is made for each segment defined at

this stage and the pathology specified in the

mathematical model. After that, a comparison is made

for a match in the parameters in the range defined by the

rule



. This calculation helps to define solution of the

first criterion of the function.

We should set a research window



for

realization of the second criterion of the function. The

search for correlation coefficients and their comparison

is performed in a sliding window in the range of the

interval P-P.



















































































  

 

cor

pat

when k - elements of observations in the range of the

processing window,

pat

cor

- correlation coefficient for

each pathology.

When the threshold value is exceeded, a decision is

made to detect the pathology, if it is confirmed for more

than 30% of the searches in the processing window.

We’ll make the transition to the frequency area using

discrete Fourier transform for realization of the third

criterion of the function.











ikfft

eYX



Comparison in the frequency area will be made

according to the shape of the envelope of the frequency

coefficients and the central moments calculated for them.

The envelope will be constructed using the method of

least squares, taking into account the criterion





min)(





xfY

. The function

)

(

is polynomial

with the size of a polynomial not exceeding the 4th

degree. The degree of the polynomial is bounded by the

fourth order because of the computational complexity of

the calculations of these indicators. In the case of

coincidence of polynomial functions, or their

discrepancy with an inaccuracy that does not exceed the

threshold value, it is decided to confirm the diagnosis.

Confirmation of the diagnosis in the frequency area is

made in case of coincidence of the group of central

moments with the central moments of the mathematical

model of the supposed pathologies.

The diagnosis is made after the analysis of the results

of each the components of the objective function.

4 Conclusion

As a result of the conducted researches the approach to

the analysis of electrocardiograms is received. The

proposed approach makes it possible to eliminate the

noise component while retaining the signal pikes.

According to the analysis of the received data, an

algorithm for finding the diagnosis is proposed.

References

1. Chung, S. H., and R. A. Kennedy, Journal of

neuroscience methods, 71-86 (1991)

2. Veeneman, D., and S. BeMent, IEEE transactions on

acoustics, speech, and signal processing, no.2, 369-

377 (1985)

3. Semenishchev, Evgeny, et al. In Proc. EWDTS, 444-

449 (2016)

4. Stadler, Robert, et al. Patent No. 6115628 A, United

States, A61B5/0452

5. Sameni, Reza, et al. Computers in Cardiology (2005)

6. Vorobyov, Sergiy, and Andrzej Cichocki, Biological

Cybernetics. No. 4, 293-303 (2002)