International Conference on Intelligent Computing and Control Systems

ICICCS 2017

ECG Based Algorithm For Detecting Ventricular

Arrhythmia And Atrial Fibrillation
Ann Brian, Sabna N., Greeshma Grace Paulson
Department of Electronics and Communication
Rajagiri School of Engineering and Technology
Kochi, India
[email protected]
[email protected]
[email protected]

Abstract—As per the report of statistics sudden cardiac

deaths affects about 30 percent of the population all around the
world. In most cases these deaths are due to arrhythmias One of the major causes of cardiac deaths is arrhythmias
including ventricular tachycardia or ventricular fibrillation and including ventricular tachycardia(VT) or ventricular
atrial fibrillation. Ventricular arrhythmia is an abnormal ECG fibrillation(VF) and atrial fibrillation(AF). Ventricular
rhythm that occurs in the ventricles of the heart and is arrhythmia is the occurrence of abnormal ECG rhythm and is
responsible for 75 to 85 percent of sudden deaths in persons with responsible for 75-85 percent death in persons with heart
heart problems unless treated within seconds. Similarly, atrial problems unless treated within seconds. Major reasons for the
fibrillation occurs due to abnormal ECG rhythm occurring in the occurrence of ventricular arrhythmias are coronary heart
upper chambers of the heart and causes the heart to quiver. disease, hypertensions, frequent heart attacks and other cardiac
These deaths occur due to hypertension and other cardiac related health issues. In other words, extra beats originating in the
issues. Sudden deaths due to arrhythmia also occur in individuals lower chambers of the heart are called ventricular arrhythmias.
who do not have high risk profiles. When a person is diagnosed These beats range from single extra beats to prolonged or
with arrhythmia, patient is left with only few seconds to die. Long sustained episodes as occurs in ventricular tachycardia or
term ECG monitoring is the standard criterion for the diagnosis fibrillation. VT is a condition in which the lower chambers of
of arrhythmia, which is time consuming. So, detection and the heart beat faster but the rate in the upper chamber remains
analysis of arrhythmia is vital as it has become one of the major constant. It acts as a complication of heart disease. VF results
causes of cardiac deaths. The project illustrates a simple
from very fast, erratic and uncontrolled electrical signals in the
algorithm for the detection of arrhythmia using a unique set of
ECG features. Databases of the ECG signal recordings from the
ventricles. This is very dangerous rhythm as it causes the heart
MIT physionet ATM, sampled at 250 Hz were used to evaluate to quiver rather than beat, leading to blackouts within seconds.
the performance of algorithm. The algorithm presented in this AF is similar to VF except that it occurs in the upper chambers
paper provides a computationally efficient solution for the of the heart. It also causes the heart to quiver. Major reasons
detection of ventricular arrhythmia and atrial fibrillation. This leading to AF are cardiac related issues, cholesterol and
makes it ideal for a wide range of machine diagnosis application hereditary reasons are also possible [2].
in biomedical field.
The implantable cardioverter defibrillator (ICD) has been
Keywords—Adaptive techniques, atrial fibrillation,
considered as the best protection against sudden death from
classification, electrocardiography(ECG), feature extraction, arrhythmias. The ECGs are obtained and analyzed, to detect
preprocessing, ventricular arrhythmia. any changes in the characteristics of the ECG signal. By
extracting information about intervals, amplitudes, and
waveform morphologies of the different P-QRS-T waves,
I. INTRODUCTION ventricular arrhythmia can be detected. Different algorithms
As per the statistics of World Health Organization(WHO), were developed to detect arrhythmias based on morphological,
cardiovascular diseases are the world’s biggest killers, spectral or mathematical features extracted from the ECG
accounting up to 9 million deaths till 2016. This has caused the signal. Each method has its own shortcomings. Some
medical researchers to give more importance on cardiac health algorithms are too difficult to implement, some not feasible
research. This gave rise to strong focus on preventive practically and some have poor detection time [1].
medicinal and technological advances. One such research lead
to the improvement of conventional cardiovascular diagnosis II. LITERATURE SURVEY
technologies in hospitals, clinics. The most common clinical
cardiac test is ECG analysis. It represents a useful screening Conventionally, doctors or human observers used to
tool for various cardiac abnormalities. It is also simple, risk monitor and diagnose arrhythmia based on particular signal
free and inexpensive [1]. features. Due to large number of patients in ICU and the need
for continuous observations of such conditions, several

978-1-5386-2745-7/17/$31.00 ©2017 IEEE 506

 International Conference on Intelligent Computing and Control Systems
ICICCS 2017

automated techniques have been developed in the past ten

years as an attempt to solve this problem [2].
A wide range of algorithms were developed to detect
arrhythmia based on autocorrelation function, frequency
domain features, time frequency analysis, wavelet transforms.
Other techniques used adaptive filtering, sequential hypothesis
[3] testing as well as morphological features [4]. Reference [4]
featured a study based on morphological features of ECG. It
includes a study of nonlinear dynamics of ECG signals for
arrhythmia characterization. Two factors such as correlation
function and largest Lyapunov exponent are used to classify
five different classes of ECG signals. The disadvantage of the
system is that discrimination between different arrhythmia
types is difficult using above mentioned two features.
Another method has been proposed in [5] is an ECG
classification algorithm based on spectral features. A real-time
discrimination of ventricular Tachy arrhythmia with Fourier
transform neural network. By observing QRS complexes in
each heartbeat, VA are distinguished from normal heartbeats.
The method achieved high sensitivity and specificity of greater
than 0.98. But the drawback of the system is the accuracy of Fig1: Schematic representation of arrhythmia detection algorithm.
distinguishing VA beats from normal beats is less when it In the first stage, the ECG preprocessing is responsible for
comes to real time implementation. 3 tasks i.e. ECG filtering, QRS complex detection and T and P
Yet another ECG cardiac arrhythmia classification based on wave delineation. The ECG filtering removes the noise coupled
higher order spectral techniques has been proposed in [6]. In with the ECG signal and prepares it for further analysis. After
this paper, a higher order spectral analysis technique is that QRS complex is detected using modified PAT algorithm.
suggested for quantitative analysis and classification of cardiac New techniques such as adaptive thresholding is introduced to
arrhythmias. The algorithm is based on bispectral analysis detect the fiducial points in each heartbeat. In second stage, six
technique. Further investigations and modification can improve features are extracted from the ECG signal. All features
the results of the technique. Several other algorithms based on represent different intervals and widths of the ECG signal.
neural networks and SVM [7] have also been implemented in They are RR, TP and PR intervals and P, QRS and T wave
literature. The disadvantage of these are either they are too widths. So far, the reported algorithms classify arrhythmia
complex for real time implementation or have less accuracy for based on a single feature such as heart rate interval, the
detection. variability of the timing delay of the ECG segments or the QT
interval variability. However, in this paper 6 unique set of
This paper presents an algorithm for detection of features are extracted that increases the robustness of the
ventricular arrhythmia and atrial fibrillation. Previous system. In the final stage, a single bit classifier signal is used to
algorithms related to arrhythmia detection had its own differentiate the abnormal ECG i.e. signals susceptible to VA
disadvantages while the algorithm presented here uses a unique or AF from normal ECG. This signal becomes high whenever a
feature extraction set that helps to achieve an efficient detection VA or VF is detected, otherwise it remains low for normal
of arrhythmia. The remaining part of the paper is arranged as ECG [1].
follows. In section III an overview of the algorithm is
presented. Section IV explains the algorithm in detail. The
results are discussed in section V. Conclusion is drawn in IV. METHODOLOGY
section VI. Finally, the future scope is mentioned in section The detailed description of algorithm is as follows. The
VII. data input is digitized. Data values are taken from MIT
physionet ATM which is sampled at 250 Hz. The readings
III. ALGORITHM OVERVIEW obtained from the MIT physionet ATM is taken from lead II.
Lead II is mostly preferred because it is most useful for
The algorithm is life savior for patients who have chances detecting cardiac arrhythmias as it lies close to the cardiac axis
of getting arrhythmia. The algorithm alerts patients who and also allows best view of P and T waves [8]. The digitized
are susceptible to VA and AF to take immediate attention to ECG data are applied in series from ROM. Each data value is
their medical condition [1]. The algorithm consists of three 8-bit size. Variable bit sizes are used in different stages to
main stages which are ECG preprocessing, feature extraction enhance accuracy and avoid truncation errors.
and classification as shown in figure 1.
A. ECG preprocessing
The block diagram of the ECG preprocessing stage is
shown in figure 2. The preprocessing stage has three main
tasks to complete.
Identify applicable sponsor/s here. If no sponsors, delete this text box
(sponsors).

978-1-5386-2745-7/17/$31.00 ©2017 IEEE 507

 International Conference on Intelligent Computing and Control Systems
ICICCS 2017

Fig2: Block diagram of preprocessing stage.

Bandpass filtering of the noisy ECG signal is the first step Fig3: Flow chart of modified PAT algorithm.
of ECG preprocessing. Firstly, a low pass filter is designed to
supress high frequency noise spikes such as because R peak has the highest amplitude among other
electromyographic(EMG) noise and power line interfrence. A waves. The 80 percent of the amplitude of the R peak value is
low pass FIR filter of order 6 is designed having a cut off taken as threshold. Next every incoming sample is compared
with this threshold. All the samples values greater than
frequency of 11 Hz. High pass filter is designed to suppres
threshold are given to output and the maximum among them is
low frequency noises i.e noises associated with the T and P chosen as R peak and so the threshold also updates. If none of
waves and baseline drift. A high pass FIR filter was designed the samples are greater than threshold then the level of the
with a cut off frequency of 5 Hz and the order of the filter is threshold is redefined. This process repeats itself for every
16. The role of differentiator is to further suppress the noise previously calculated RR interval [1].
and also gives information about slope of QRS complex,
thereby highlighting the R peak. Squaring is a non linear B. Feature Extraction
filtering stage to emphasize the R peak from T and P waves. While designing a detection system two main factors that
Moving window integral as the name suggests a window of must be considered are the complexity and accuracy of the
size 32 samples is moved one sample at a time and average is feature extraction stage to provide best outcome. The
taken. It gives information about the width of QRS complex. complexity of the system should be as simple as possible. More
The window size 32 is taken experimentally by trial and error complex system will increase the overall cost of the system.
method[1]. The extracted features must be apt and must help in producing
best results. Previous works were based on only a single
feature, which gave less accuracy for detection. While in this
The second task of the preprocessing stage is QRS
paper 6 unique feature extraction was done to increase the
detection. This detection is performed based on modified PAT robustness of the system. The extracted features are such that it
algorithm. The original PAT algorithm[9] was taken as maintains low complexity and high accuracy [1].
reference for modified PAT algorithm. The QRS detection is
important as it serves as the reference to T and P wave The result of such analysis lead to the extraction of a unique
detection[1]. So it must to done with great precision. The flow set of features. A set of six features were extracted, which
indicated great significance in the detection of arrhythmia. The
chart of the algorithm is shown in figure 3.
features are RR, TP, PR, T width, P width and QRS width.
These features were processed using two consecutive heart
The third task of the preprocessing stage is T and P wave beats, whereas previous algorithms used only single heartbeat
delineation. From the QRS offset to 2/3 of the RR interval a [1]. Figure 4 shows these intervals on ECG record.
window is considered and the value maximum within that
window is considered as T peak. From the QRS onset to 1/3 of C. Classification
the RR interval a window is considered and the value The feature set is given as input to the classifier stage.
maximum within the window is taken as P peak[1,10]. A single bit classifier was designed to detect arrhythmia.
When the extracted features and the heart rate crosses, the
The algorithm first reads 3000 values from the SRAM. standard values the signal becomes high. Heart rate is
SRAM stores the time averaged ECG samples. The maximum calculated based on equation 1.
value among these samples is found out and it is set as R peak,

978-1-5386-2745-7/17/$31.00 ©2017 IEEE 508

 International Conference on Intelligent Computing and Control Systems
ICICCS 2017

an abnormal VT detected ECG. Figure 6(c) shows the

simulation of an abnormal VF detected ECG. Figure 6(d)
shows the simulation of an abnormal AF detected ECG. For
figures 6b, 6c and 6d the Analysis out signal remains high.
Table III, IV and V compares the standard values with values
obtained after simulation. These two tables satisfy the criterion
mentioned in table I and II.

Fig4: ECG features extracted marked on waveform

Fig5(a): Original noisy signal.

Heart rate= 60/(count*time)

where count is the number of clock cycles and time is the
time period of one clock cycle which is 11.5 ms.

V. RESULTS AND DISSCUSSIONS

Fig5(b): Bandpass filtered Output.

This section illustrates the simulated results of the proposed
arrhythmia detection. Certain stages of the proposed algorithm
have been plotted on the MATLAB to get a clear picture of
the operations done on the ECG waveform. Xilinx ISE 14.2 is
used for functional verification and simulations.

A. Results of ECG preprocessing stage

The filtering of the ECG signal was done through a cascaded
LPF and HPF. The two filters have done a good job in
cleaning up the ECG signal from the coupled noise. The
original noisy signal is shown in figure 5a and the noise free
signal is shown in figure 5b. Both the signals have been
Fig5(c): Output after Moving Window Integral.
plotted in MATLAB. Also, the output after the moving
window integral stage has also been plotted in MATLAB
(figure 5c). As mentioned in theory after MWI stage only QRS
complexes will be emphasized, P and T waves will be VI. CONCLUSION
attenuated such that it is almost absent in the waveform [1].
As per the report of the statistics, cardiac deaths are the
major causes of death rates all around the world. As the death
B. Results of feature extraction and classification stage rates are increasing year by year the researches have been
probing to find a solution to this problem since last ten years.
Here the Verilog simulations of the normal ECG database
Many different algorithms have been implemented in literature.
and abnormal ECG databases are shown. The different
Arrhythmias are serious issues regarding health, and the patient
features which are extracted are also shown here. Figure 6(a) is left with only seconds to die. The conventional monitoring
shows the simulation of a normal ECG database. As seen in the ECG screen and then deciding the cause will take a long
the figure Analysis out is the single bit ECG classifier signal, time. This will cause the patient to die. So, a method for the
this signal remains low. Figure 6(b) shows the simulation of detection of ventricular arrhythmia and atrial fibrillation is

978-1-5386-2745-7/17/$31.00 ©2017 IEEE 509

 International Conference on Intelligent Computing and Control Systems
ICICCS 2017

presented in this paper. This paper is focuses on the detection

of arrhythmia. The algorithm presented in this paper provides a
low latency detection system. Also, a unique set of 6 new ECG
features helped in increasing the accuracy and robustness of the
system.

VII. FUTURE WORK

This algorithm can be further improvised to increase the
percentage of detection accuracy. Changes such as search-back
mechanisms to reduce the miss detections, better filtering Fig 6(a): Verilog simulation on AF detected database.
techniques can be employed to further reduce the noise and
more features can be extracted for even better accuracy. Also,
ECG signals can also be integrated with multiple biomedical
signals [1].

Fig 6(a): Verilog simulation tested on normal database.

Fig 6(b): Verilog simulation tested on VT detected database.

Fig 6(a): Verilog simulation tested on VF detected database.

978-1-5386-2745-7/17/$31.00 ©2017 IEEE 510

 International Conference on Intelligent Computing and Control Systems
ICICCS 2017

testing of binary sequences” ,in Proc. IEEE Comput. Cardiol., Sep./Oct.

2007, pp. 573_576..
[4] P. de Chazal, M. ODwyer, and R. B. Reilly, ”Automatic classification of
heartbeats using ECG morphology and heartbeat interval features”
IEEE Trans. Biomed. Eng., vol. 51, Issue no. 7, pp. 1196_1206, Jul.
2004.
[5] Kei-ichiro Minami, Hiroshi Nakajima and Takeshi Toyoshima, ”Real-
Time Discrimination of Ventricular Tachyarrhythmia with Fourier-
Transform Neural Network”,IEEE Trans. Biomedical Engineering,
vol.46 Issue no. 2, pp.179_185, Feb 1999.
[6] Labib Khadra, Amjed S. Al-Fahoum, and Saed Binajjaj, ”A Quantitative
Analysis Approach for Cardiac Arrhythmia Classification Using
Higher Order Spectral Techniques”,IEEE Trans. Biomedical
Engineering,vol.11, Issue no.11, pp. 1840 1845, Nov 2005
[7] Qibin Zhao, Liqing Zhang, ”ECG Feature Extraction and Classification
REFERENCES Using Wavelet Transform and Support Vector Machines” ,IEEE Trans.
Biomedical Engineering, pp.1089_1092, Feb 2005.
[8] http://www.ebme.co.uk/articles/clinical-engineering/20-
[1] Nourhan Bayasi, Temesghen Tekeste, Hani Saleh,Baker Mohammad,
cardiac_monitoring
Ahsan Khandoker and Mohammed Ismail, Low-Power ECG-Based
Processor for Predicting Ventricular Arrhythmia, IEEE Transactions on [9] J. Pan and W. J. Tompkins, ”A real-time QRS detection algorithm” ,
Very Large Scale Integration(VLSI)Systems, 2016. IEEE Trans. Biomed. Eng., vol. BME-32, no. 3, pp. 230_ 236,Mar.
1985.
[2] http://www.hopkinsmedicine.org/healthlibrary/conditions/cardiovascular
[10] N. Bayasi, T. Tekeste, H. Saleh, A. Khandoker, B. Mohammad, and M.
_diseases_anatomy_and_function_of_the_hearts_electrical_system_
Ismail, ”Adaptive technique for P and T wave delineation in
85, P00214
electrocardiogram signals” ,in Proc. IEEE 36th Annu. Int. Conf. Eng.
[3] J. Pardey, ”Detection of ventricular fibrillation by sequential hypothesis Med. Biol.Soc., Aug. 2014, pp. 90_93.

978-1-5386-2745-7/17/$31.00 ©2017 IEEE 511