Proceedings of the IEEE 2017 International Conference on Computing Methodologies and Communication

(ICCMC)

Design and develop an algorithm for a

diabetic detection using ECG signal

Reena Musale A. N. Paithane

Student of Electronics &Telecommunication
Engineering,
Student RSCOE
of Electronics Pune, India
&Telecommunication Faculty of Electronics & Telecommunication
Engineering, RSCOE Pune, India Engineering, RSCOE Pune, India
[email protected]
[email protected] [email protected]

Abstract--- Huge number of individuals are influenced characterized by high blood glucose and either
by Diabetes Mellitus (DM) which is hard to cure inefficient or ineffective insulin. There are four types
because of its endless nature and hereditary connection. of diabetes, Type I, Type II, gestational diabetes and
The uncontrolled diabetes may prompt heart related other specific type but main are type I and type II.
issues. Along these lines, the diagnosis and checking of
Type II diabetes is most common form of diabetes,
diabetes is of extraordinary significance. By utilizing
RR-interval signals, the automatic detection of diabetes with about 90% of diabetes falling into this category.
can be performed. The nature of RR-interval signals is It usually occurs in individuals who are over 40 years
nonlinear and non-stationary. Subsequently linear of age who are overweight [12] [13]. Diabetes can
strategies will be unable to catch the hidden data lead to risk of long-term complications. One of the
introduce in the signal. Empirical Mode Decomposition long-term complications is the damage of blood
(EMD) is recently used in a broad range of applications vessels. The diabetes can increase the risk of
for extracting signals from data generated in nonlinear Cardiovascular Disease (CVD).
and non-stationary processes. However, it has an
Intrinsic Mode Function (IMF) consisting of signals of
Globally, an approximate population of 382
widely disparate scales. a new nonlinear method based
on empirical mode decomposition (EMD) is proposed in million are having diabetes worldwide as of 2013.
this paper to discriminate between normal RR-interval Among them, 90% of the cases belong to type 2
signals and diabetic. By using IMFs of RR-interval diabetes which is equal to 8.3% of population of the
signals the different parameters are extracted namely, total adult subjects. According to recent World
the mean frequency parameter using Fourier -Bessel Health Organisation (WHO) findings diabetes may
series expansion , amplitude modulation double the risk of death. The expected number of
bandwidth , frequency modulation bandwidth people with diabetes is 592 million by 2035 [5] [10]
and area computed from second order difference plot [8] [9]. Currently there are many techniques that are
, Mean average value, First difference, ZCR.
applied to diagnose diabetes problems such as
SVM (support vector machine) classifier has been
utilized to measure the discrimination ability of the
Wavelet transformation, frequency domain analysis,
proposed features for detection of diabetic RR interval time domain analysis etc. The main drawback of
signals. The Results achieved from this proposed these approaches is that the basic functions are fixed
methodology indicate that these features provide the and do not match with varying nature of signal [1]
important difference between normal classes and [4].
diabetic. By using developed algorithm, the accuracy of
analysis is 95%. In past, several methods have been proposed for
diabetic detection using physiological signal. Heart
Keywords—Diabetes; ECG; EMD; IMF;
rate variability(HRV) is a signal which shows
Nonlinear
difference between two consecutives hear beats. This
I. INTRODUCTION signal contains information related to heart condition.
Therefore, analysed HRV in time and frequency
The analysis of Electrocardiogram (ECG) signal as a domain using non-linear parameter is useful in
diagnostic tool is a relatively old field and it is diagnostics, but time domain is unable to differentiate
therefore often assumed that the ECG is a simple between sympathetic and parasympathetic
signal that has been fully analysed. ECG give contributions of HRV [1]. In [3] empirical mode
valuable information about the heart’s electrical decomposition(EMD) decomposes centre of pressure
activity. diagnostic tool that measures and records the signal into a finite set of band limited signal.
electrical activity of the heart in exquisite detail. Detection of QRS-complexes plays an important role
Interpretation of these details allows diagnosis of in the analysis of ECG signal. This system used as
diabetes [2] [4] [10]. Diabetes Mellitus (DM) or algorithm based on discrete wavelet transformation
diabetes, is a group of metabolic disorder that is [8]. The outcomes of HRV applied to Poin-care plot

978-1-5090-4890-8/17/$31.00 ©2017 IEEE 961

Authorized licensed use limited to: COLLEGE OF ENGINEERING - Pune. Downloaded on March 18,2024 at 09:56:50 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the IEEE 2017 International Conference on Computing Methodologies and Communication
(ICCMC)

analysis and calculate indices such as SD1, SD2 and band pass filter for the QRS detection algorithm
CCM for diabetic patient [9]. Empirical mode reduces noise in the ECG signal by matching the
decomposition method is used to decompose ECG spectrum of the average QRS complex.
signal and then analysed normal and diabetic RR This attenuates noise due to muscle noise, power line
interval signal using Kruskal - Wallis statistical test interference, baseline wander, T wave interference.
[10]. The pass band that maximizes the QRS energy is in
the 5Hz-35Hz range [10]. The filter implemented in
In this paper, we have used a new technique, this algorithm is composed of cascaded high pass and
called the empirical mode decomposition (EMD) for low pass filters. Then, the Pan-Tompkins algorithm is
analysis of non-linear and non-stationary ECG signal. used to find the QRS complexes from the processed
This technique adaptively decomposes a signal into ECG signal. The RR-intervals are calculated as the
Intrinsic Mode Function (IMFs) which are symmetric time between two consecutive QRS complexes.
and band-limited in nature and keep the signal in its
own domain [4] [14] [11]. The parameters obtained
from the ECG signals can be used as feature set for Band-Pass
SVM classifier as an input to discriminate diabetic ECG Filtering
and normal class. The remaining part of the paper is Signal
organized as follows: In Section II, the dataset, data
preprocessing, EMD method are explained. Section
III includes results and discussion. The conclusion is
provided in Section IV. R- peak Detection
(Pan & Tompkins
Algorithm) and Pre-
II. METHODOLOGY processed
Generation of RR-
interval Signal ECG
A. Block Diagram

ECG Pre- Feature

signal processing Extraction RR-interval
using EMD Signal

Training Phase

Fig. 2. Processing steps for extracting RR-interval signal from the

ECG signal.
Classification Results

Pan & Tompkins Algorithm

Testing Phase

Feature The works of Pan Tompkins greatly influence

ECG Pre- Extraction
the R-R interval detection as compared to others. A
signal processing using EMD
survey of literatures signifies this approach as one of
important algorithm in detecting QRS peak [4].
Wherever The accuracy of any Electrocardiogram
Fig. 1. Block diagram of diabetic class detection. (ECG) waveform extraction plays a vital role in
helping a better diagnosis on diabetes. Normal ECG
B. Data set should consist of several parts include P wave, QRS
complex and T wave [5].
The ECG signal dataset used in this work is
publicly available online. In this study, a total 50 These waves reflect the heart’s activity such as P
Electrocardiogram (ECG) signals from 33 healthy or wave produced by muscle contraction of Atria and its
normal persons and 17 from diabetic patients were duration indicates the Atrial enlargement. Q wave
acquired from Physio bank [5] [6]. gives the first negative value and typically supposed
C. Data Pre-processing to be 25% less than the R wave value.

Following steps performed on the ECG signals to

obtain the RR-interval signals are presented in the
Fig. 2. The unwanted noise is filtered by passing the
original ECG signals through a band pass filter. The

978-1-5090-4890-8/17/$31.00 ©2017 IEEE 962

Authorized licensed use limited to: COLLEGE OF ENGINEERING - Pune. Downloaded on March 18,2024 at 09:56:50 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the IEEE 2017 International Conference on Computing Methodologies and Communication
(ICCMC)

condition is similar to the narrow-band requirement

for a stationary Gaussian process and the second
condition is a local requirement induced from the
global one, and necessary to ensure that the
instantaneous frequency will not have redundant
fluctuations as induced by asymmetric waveforms.
The non-stationary signal x(t) is then represented as a
linear sum of IMFs and the residual component:

x(t) =∑ ( )+ ( ) (1)

where M is the number of IMFs and ( ) is the

final residue.
Fig. 3. ECG signal.
F. Classification Technique
Normal Values of Amplitude and Duration of ECG
Here we use SVM as a classifier. as we know that
Parameters: SVM is globally used training algorithm used for
many applications [15], so the extracted parameters
Amplitude-: from ECG signal are applied to the selected SVM
classifier as a input to train the classifier. support
• P wave -: 0.25mV vector machines are supervised learning models with
associated learning algorithms that analyse data used
• R wave -: 1.6 mV for classification. output state of the selected
classifier is to assign to a certain class either normal
• Q wave -:25 percent of R wave or diabetic class.
• T wave -:0.1 to 0.5 mV

Duration -: III. RESULTS AND DISCUSSION

• P-R Interval -:0.12 to 0.20 sec The algorithm was implemented for several
nonlinear and non-stationary ECG signals. These
• Q-T Interval -:0.35 to0.44 sec signals are decomposed into IMFs by using EMD
which are symmetric, band-limited and oscillatory in
• S-T Interval -:0.05 to0.15 sec
nature. In order to extract features from IMFs, the
• P wave Interval -:0.11 sec EMD method is used to decompose the normal and
diabetic RR-Interval signals which are used in
The “Pan and Tompkins” QRS detection algorithm dataset. To extract useful information from the IMFs,
identifies the QRS complexes based upon digital time-domain and frequency-domain parameters are
analysis of slope, amplitude, and width of the ECG computed from IMFs. These extracted features are
data. The algorithm implements a special digital band area computed from ( ) SODP of IMFs, mean
pass filter. It can reduce false detection caused by the frequency of IMFs computed using Fourier–Bessel
various types of interference present in the ECG series expansion ( ), AM bandwidth ( ) and
signal. The algorithm automatically adjusts the FM bandwidth ( ) of IMFs, ZCR, Mean average
thresholds and parameters periodically to adapt the value, Mean frequency, first difference etc. These
changes in QRS complexes. parameters are helpful to discriminate the diabetic
and normal class using support vector machine
E. Empirical mode decomposition (SVM) Classifier. The EMD is a nonlinear method,
hence it is applied for analysis of RR-interval signals.
The principle of the EMD technique is to The RR-interval signals can be considered as the sum
decompose a signal x(t) automatically into a set of of the fast and slow oscillatory components. This
the band-limited functions ( ) called IMFs [7] EMD method is able to separate these fast
[11]. Each IMF satisfies following two conditions. oscillations from slow oscillations present in the
(a) Number of extrema (maxima and minima) and signal in the form of IMFs. Fig 4-8 shows the results
number of zero-crossings in a signal should be either respectively. The proposed approach yielded
equal or differ by one. reasonable classification accuracy between normal
and diabetic class.
(b) The mean value of two envelopes, one formed by
connecting local maxima and other formed by local
minima, at any point should be zero. The first

978-1-5090-4890-8/17/$31.00 ©2017 IEEE 963

Authorized licensed use limited to: COLLEGE OF ENGINEERING - Pune. Downloaded on March 18,2024 at 09:56:50 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the IEEE 2017 International Conference on Computing Methodologies and Communication
(ICCMC)

(a)
(b)

Fig. 5. Filtered ECG signal: (a) normal (b) diabetic

Fig .5 (a) and (b) shows Filtered ECG signal of

normal and diabetic subject, here unwanted noise
removed by using band pass filter.

(b)
Fig. 4. Original ECG signal: (a) normal (b) diabetic

Fig .4 (a) and (b) shows original ECG signal of

normal and diabetic subject.
(a)

(b)

Fig. 6. RR-interval signal: (a) normal (b) diabetic

(a)

978-1-5090-4890-8/17/$31.00 ©2017 IEEE 964

Authorized licensed use limited to: COLLEGE OF ENGINEERING - Pune. Downloaded on March 18,2024 at 09:56:50 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the IEEE 2017 International Conference on Computing Methodologies and Communication
(ICCMC)

Plots of RR- intervals of normal and diabetic subjects Table 1. Feature extraction
are shown in Fig.6(a) and (b). which shows
difference between QRS duration.

FEATURES NORMAL ABNORMAL

Root mean square 36.6749 32.5035

Mean average 17.3084 14.3121

value

First difference 390 338

Zero crossing 0.053889 0.046944

rates

Mean square 30.4882 23.929

Centre frequency -0.00127366 -0.00700556

Amplitude -0.403853, 0.3846, 18.3731,

modulation 26.6861, 57.1272, 48.3942, 1.53749
0.393928

Frequency 0.0108525, -0.012524,

(a) modulation 0.708538, 0.705504, 1.5044,
1.49201, 0.026246
0.0136358

All the parameters extracted from IMFs of ECG

signal shown in Table 1. these features used as input
to SVM classifier.

(b)

Fig. 7. IMFs of RR-interval signal: (a) normal (b) diabetic

The first nine IMFs of normal and diabetic RR-

interval signals shown in Fig. 7(a) and (b). These
(a)
IMFs can be considered as narrow- band signals.

978-1-5090-4890-8/17/$31.00 ©2017 IEEE 965

Authorized licensed use limited to: COLLEGE OF ENGINEERING - Pune. Downloaded on March 18,2024 at 09:56:50 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the IEEE 2017 International Conference on Computing Methodologies and Communication
(ICCMC)

very serious social problem since diabetes is one of

the major cause of cardiovascular diseases. so early
detection and monitoring can help people to save
their life.

REFERENCE
.
1. Acharya, Rajendra, N. Kannathal, and S. M. Krishnan.
"Comprehensive analysis of cardiac health using heart
rate signals." Physiological measurement 25.5 (2004):
1139
2. Mohamed Omar, Abdalla Sayd, et al. "Application of
the Empirical Mode Decomposition to ECG and HRV
Signals for Congestive Heart Failure Classification."
IEEE, 2011.
3. Pachori, Ram Bilas, et al. "Analysis of centre of
pressure signals using empirical mode decomposition
and Fourier-Bessel expansion." TENCON 2008-2008
IEEE Region 10 Conference. IEEE, 2008.
4. A. N. Paithane, D. S. Bormane, “Human Emotion
Recognition using Electrocardiogram Signals”,
(b) International Journal on Recent and Innovation Trends
in Computing and Communication ISSN: 2321-8169
Fig. 8. Results: (a) normal (b) diabetic. 5. Vanisree K, Jyothi Singaraju, “Automatic Detection of
ECG R-R Interval using Discrete Wavelet
Transformation”, International Journal on Computer
Fig 8. shows final classification results (a) indicates Science and Engineering (IJCSE), 2011.
normal class and (b) indicates abnormal class. 6. Paithane A. N., and D. S. Bormane. "Analysis of
nonlinear and non-stationary signal to extract the
features using Hilbert Huang transform."
Computational Intelligence and Computing Research
(ICCIC), 2014 IEEE International Conference on.
100 IEEE, 2014.
90 7. Shah, Meet, et al. "Analysis of epileptic seizure EEG
80 signals using reconstructed phase space of intrinsic
70 mode functions." 2014 9th International Conference on
60 Industrial and Information Systems (ICIIS). IEEE,
50 2014.
40 8. Acharya, U. Rajendra, et al. "An integrated diabetic
30
20 index using heart rate variability signal features for
10 diagnosis of diabetes." Computer methods in
0 biomechanics and biomedical engineering 16.2 (2013):
222-234.
9. Abubaker, Hanin B., et al. "Poincaré plot analysis of
heart rate variability in the diabetic patients in the
UAE." 2nd Middle East Conference on Biomedical
Engineering. IEEE, 2014.
10. Pachori, R. B., Avinash, P., Shashank, K., Sharma, R.,
& Acharya, U. R. (2015). "Application of empirical
mode decomposition for analysis of normal and
diabetic RR-interval signals". Expert Systems with
Applications, 42(9), 4567-4581.
Fig. 9. Performance analysis of different techniques 11. Paithane A. N., and D. S. Bormane.
"Electrocardiogram signal analysis using empirical
In Fig.9 horizontal axis shows different techniques mode decomposition and Hilbert spectrum." Pervasive
Computing (ICPC), 2015 International Conference on.
and vertical axis represents accuracy in percentage. IEEE,2015.
12. http://en.wikipedia.org/wiki/Diabetes_mellitus.
IV. CONCLUSION 13. https://www.idf.org
14. Paithane A. N., D. S. Bormane, and Ujwala Patil.
From this work, we conclude that ECG signal’s plays "Novel Algorithm for Feature Extraction and Feature
Selection from Electrocardiogram Signal."
an important role in diabetic class International Journal of Computer Applications 134.9
detection. The various features’ values calculated in (2016).
the training phase, are used to train the SVM 15. Ms. Suruchi Pimple "Application of Support Vector
classifier and further for determining the diabetic Machine for Diagnosis of Diabetes: A Systematic
review. "IOSR Journal of Computer Engineering
class accurately with the help of corresponding ECG (IOSR-JCE), (2016).
signal. As diabetic is metabolic disorder affecting
majority of population worldwide and very difficult
to cure due to its chronic nature and genetic link, it is

978-1-5090-4890-8/17/$31.00 ©2017 IEEE 966

Authorized licensed use limited to: COLLEGE OF ENGINEERING - Pune. Downloaded on March 18,2024 at 09:56:50 UTC from IEEE Xplore. Restrictions apply.