0% found this document useful (0 votes)
60 views16 pages

Fusion Based Feature Extraction Analysis of ECG Signal Interpretation - A Systematic Approach

This document is a research article from the Journal of Artificial Intelligence and Capsule Networks that examines fusion-based feature extraction for electrocardiogram (ECG) signal interpretation using a systematic machine learning approach. The authors propose a framework involving preprocessing with de-noising filters, fusion-based feature extraction, and predictive modeling to classify ECG signals. They achieved high accuracy of 96.5% in classifying normal, abnormal, and irregular heartbeats for cardiovascular disease detection. The article reviews related work on ECG classification and supervised/unsupervised learning methods before detailing the proposed methodology and discussing results.

Uploaded by

Jati Warama
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
60 views16 pages

Fusion Based Feature Extraction Analysis of ECG Signal Interpretation - A Systematic Approach

This document is a research article from the Journal of Artificial Intelligence and Capsule Networks that examines fusion-based feature extraction for electrocardiogram (ECG) signal interpretation using a systematic machine learning approach. The authors propose a framework involving preprocessing with de-noising filters, fusion-based feature extraction, and predictive modeling to classify ECG signals. They achieved high accuracy of 96.5% in classifying normal, abnormal, and irregular heartbeats for cardiovascular disease detection. The article reviews related work on ECG classification and supervised/unsupervised learning methods before detailing the proposed methodology and discussing results.

Uploaded by

Jati Warama
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 16

Journal of Artificial Intelligence and Capsule Networks (2021)

Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

Fusion based Feature Extraction Analysis of ECG


Signal Interpretation – A Systematic Approach
Dr. T. Vijayakumar,
Professor,
Department of ECE,
GNIT, Hyderabad, India.
Email id: [email protected]

Mr. R. Vinothkanna,
Department of ECE, Vivekanandha College of Technology for Women,
Namakal, India.
Email: [email protected]

Dr. M. Duraipandian,
Head of Department, Department of Computer Science and Engineering,
Vivekanandha College of Technology for Women,
Namakal, India.
Email: [email protected]

Abstract- Our human heart is classified into four sections called the left side and right side of the
atrium and ventricle accordingly. Monitoring and taking care of the heart of every human is the
very essential part. Therefore, the early prediction is essential to save and give awareness to
humans about diet plan, lifestyle schedule. Also, this is used to improve the clinical diagnosis and
treatment of any patients. To predict or identifying any cardiovascular problems, Electro Cardio
Gram (ECG) is used to record the electrical signal of the heart from the body surface of humans.
The algorithm learns the dataset from before cluster is called supervised; The algorithm learns to
train the data from the set of a dataset is called unsupervised. Then the classification of more
amount of heartbeat for different category of normal, abnormal, irregular heartbeats to detect
cardiovascular diseases. In this research article, a comparison of various methods to classify the
dataset with a fusion-based feature extraction method. Besides, our research work consists of a de-
noising filter to reconstruct the raw data from the original input. Our proposed framework
performing preprocessing that consists of a filtering approach to remove noises from the raw data

1
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

set. The signal is affected by thermal noise and instrumentation noise, calibration noise due to
power line fluctuation. This interference is high in many handheld devices which can be eliminated
by de-noising filters. The output of the de-noising filter is input for fusion-based feature extraction
and prediction model construction. This workflow progress has given good results of classifier
effectiveness and imbalance arrangement conditions. We achieved good accuracy 96.5% and
minimum computation time for classification of ECG signal.

Keywords: ECG classification, Random Forest, Feature Extraction

1. INTRODUCTION
In the world, 30% of death is due to cardiovascular problems disease according to World
Health Organization (WHO). The disease mechanism is very crucial during the death of any human
in dangerous circumstances. Therefore, any strange activity or abnormal condition of the heart
routine process can be determined by the radiologist to predict the heart disease [1]. There are lots
of ECG formats to understand the normal and abnormal state of the heartbeat in the body. The
probe leads fix temporarily in the body and record reading from the heartbeat in the body surface
[2]. Based on the number of leads the information’s are vast that both of them is depending on
each other. Recently this ECG recording includes standard 12 lead probe’s that processing by deep
learning technique to improve the prediction rate with high accuracy. This large amount of probe
leads provides more information about heartbeat for getting higher accuracy [3]. Deal this amount
of data, we need to go for a powerful tool to compute that information. The ECG formats are
needed powerful computational methods to extract information from the complete dataset. This
research article is focusing to describe the machine learning method to analyze the ECG dataset
for the diagnosis of all cases of the patient. We are trying to extend our proposed framework for
this analysis. This analysis method is used to learn and train networks to make an accurate
prediction of diagnosis [4].

2
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

Figure 1 Cross sectional view of human heart picture

There are two type of learning methods as follows;

1. Supervised
2. Unsupervised

The machine learning methods are a capable approach to deal with medical ECG signals with
various directions of simulation results. Besides, the lack of a database is one of the great
challenges in the medical field to solve the disease interpretation issues [5]. For training and
validation process of the dataset is interpreting the ECG signal and it depends on the generation of
synthetic data. This simulation results have several points and task which includes the deal with
noise in the signal for QRS detection, P waves, and T waves. The noise reduction can be done in
this article for getting clear signals for training and learning further [6]. Many transform and filter
techniques are carried out to deal with this noise reduction in a signal. The wavelet transform
method increases a good prediction and detection rate over 99%. To the analysis of ECG signals,
the position and shape of the signal are very important components [7]. The advanced method of
Kalman filtering for detection is reaching good accuracy results in ECG signal processing. The
classification of ECG signal ensembles many class testing and training process in pattern
recognition [8]. The simplified block diagram of the proposed architecture framework is showing
in figure 2.

3
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

Figure 2 Block diagram of overview of proposed architecture construction

The changes in shape and position of ECG signal are very tough to analyze with a simple
and single algorithm. The threshold-based analysis is suffering from slight changes of ECG
recording that the new person’s recorded signal delay or early start of the heartbeat signal [9]. Also
if changes of new devices provide different threshold values based on the system specification.
Therefore, the data-driven or adaptive method is essential to handle this analysis with a hybrid
type of machine learning approach. The right side of the heart is containing a tricuspid valve used
to control the motion of the blood from the atrium to the ventricles which are showing in figure 1.
ECG is the process of recording the electrical signal from the heart based on the body surface to
monitor the activities [10]. The electrodes are called probe lead fixing in the body near to the heart
to get the information of heartbeat. This fixing of probe electrodes is near to atrioventricular and
ventricle nodes. The healthy heart has a specific order of division and intervals between each
heartbeat. If there are any deviations in the heartbeat will be noted as an abnormal condition [11].
The normal condition of the heartbeat is showing in figure 3.

Figure 3 Normal condition of heartbeat

4
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

The figure shows the voltage and time sequence of a signal produced in ECG. The sequence
of ECG indicates our heartbeat health and its conditions. The periodic PQRST signal is indicated
in the heartbeat activities graph. An among these PQRST signals consist of; the P indicates atrial
contraction, QRS indicates a contraction of ventricles, T indicates an expansion of the ventricle
action respectively [12]. This wave signal is depending on the heartbeat rate. The typical heartbeat
points are showing in figure 4.

Figure 4 Typical ECG complexes

The interval measures method is measuring the distance between each complex. Also, the
irregular and regular rhythm is classifying based on heartbeat rate. The heartbeat rate is in between
60 to 90 bpm is called the normal heartbeat rate [13]. The many ECG complexes classification
problems are solving by the hybrid computational algorithm with neural network solution. Because
of its strong robustness, the prediction rate is high. Also, this type of solution can solve problems
of the complex nonlinear structure easily [14].

2. ORGANIZATION OF THE RESEARCH


The structure of the research article is organized as follows; Section 3 gives related work
of ECG classification based on a machine learning algorithm. Section 4 discusses proposed
methodologies for ECG classification. Section 5 consists of a result discussion from the obtained
output. The conclusion and the future task delivers in section 6.

5
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

3. PRELIMINARIES
Pinho Andre et al propose artificial intelligence analysis for ECG signal classification.
They tested with SVM of a different set of extracted features. They compared with artificial
Intelligence Neural Network (ANN) and proven the performance is better than SVM [15].
Moavenian et al introduces Multi-Layered Perceptron (MLP) and compared it with SVM and
proving higher performance in the speed process. But the accuracy of the process is slower than
the existing method [16]. Zhang et al propose SVM with the genetic method is performing for the
classification and achieved 93% true positive with moderate data set around 750 ECG signal record
[17]. Chauhan et al tested a deep LSTM network with the unknown size of the dataset for signal
classification. The measuring metric is calculated as 97.5% precision and 96.45% F-score.
Leutheuser et al classifies the abnormal and normal condition of heartbeat with the MIT-BIH label
method and obtained 89.6% accuracy which is validated on the dependent dataset [18]. Oresko et
al compute the normal and abnormal condition detection of the heartbeat by many morphological
filter patterns and get 99% accuracy for the normal method [19]. J.Wang et al introduce the RR
interval feature normalization method to finding the distance between the PQRST complexes. The
measuring metric is computed for various classification methods. They achieved good accuracy,
specificity with a moderate amount of dataset [20]. A.Dallali et al review the RR interval feature
normalization with many metric measures. They introduce the removing noise in the raw electrical
signal from the ECG device. They obtained high accuracy and precision [21]. M.Vijayavanan et
al calculate intervals between PQRST duration segments from signal feature extraction. They
trained MIT-BIH arrhythmias for forwarding classifier technique and tested with high accuracy
[22]. V.Srivastava et al implement with same arrhythmias feed-forward transformation method
incorporated with fuzzy logic method provide good specificity rate [23]. M.Korurek et al propose
5R peak with noise removing algorithm of linear phase filtering techniques. They constructed the
framework with baseline correction for signal extraction and its classification; Therefore, they
achieved good higher sensitive and specificity rate [24].

6
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

4. METHODOLOGIES
4.1 Filtering Approach
Usually, ECG signal is affected by noise due to various sources. The variation of the signal
is depending on the changes in the position of electrodes. These small changes in the signal create
shifting the signal from the baseline which interprets the wrong diagnosis. These small signal
changes will not be in the specified range of frequencies. Therefore, elimination is a big challenge
here. The construction of the bandpass filtering approach is performed here to eliminate the noise
from the signal. The proposed model is based on many features that are computing for the
prediction level. To reduce the noise in the data, the time domain co-efficient has been taken to
get a higher accuracy level of the data set [25]. The model has developed with many features based
on fusion technique is showing a good prediction rate. The analysis of noisy datasets is performing
with feature coefficients for many decomposition iterations of bandpass filtering. To achieve
higher accuracy, rate the bandpass filtering techniques is one of the great choices for clean data
classification. The heartbeat variability measure also considering in the construction of the
framework. The filter decomposition level is variable based on the raw data input ECG signal
variation. It is an adaptive function approach transformation can provide better visibility after de-
noising the signal. The interval is computed includes time-domain components and frequency
domain components for good visibility structure. This all processing flow is showing in figure 5.

4.2 Fusion based Feature Selection


In this research article, feature selection has been included for detection. Mainly QRS
detection is a challenging task in this extraction process. Figure 6 shows the feature extraction
methods in our proposed architecture. To find R peaks in the electrical heartbeat signal of ECG,
our proposed algorithm is appropriate for a higher accuracy detection rate. The main standard
algorithm is the Pan-Tomkins algorithm which is used to detect QRS complexes [26]. This
heartbeat detection should be near to the ideal perfection of our proposed algorithm. Our research
work concentrates on filtering processing for raw data of input ECG signal in our algorithm. The
data is filtered with a range of frequencies is mentioned as equation below,

7
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

1 𝑡2
𝑊𝑏𝑝 (𝑡) = 𝑒𝑥𝑝 (− ) {exp[𝑗2𝜋𝑓𝐿 𝑡] + ⋯ + exp[𝑗2𝜋(𝑓𝐿 + 𝑘∆f)t] + ⋯ + exp[j2π𝑓𝐻 𝑡]}
√2𝜋𝑎 2𝑎2
Where k=0,1,2…N and 𝑓𝐻 = 𝑓𝐿 + 𝑁∆𝑓
This can help to obtain information about the slope of the QRS. The classification method
here is used to classify the data based on the training set and class labels in the proposed structure
and it helps to predicts the class label also. Finally, our proposed system construct with a prediction
segment to predict the structured models with continuous-valued function [27]. Therefore, the
missing or unlabeled values can be re-calculated for medical diagnosis.

Figure 5 Preprocessing Segment of Proposed Work

Besides, to put a boundary condition for minima and maxima for ECG signal to feature
extraction vector. The feature extraction computes the R peaks appearing in the length of the signal
sequences. The standard deviation of the R peak value is directly proportional to obtained and
predicted R-Peak values with a minus sign [28]. Based on our approach the heartbeat finds the best
representation of heartbeat from the input signal. Our filtering approach is also more resistant to
the noise signal.

Figure 6 Feature fusion technique of Proposed Work

8
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

This feature selection is one of the essential parts of our proposed architecture to select the
subset from the raw feature set. This selection procedure can be efficient and given data makes
weak cross-correlation function among all the features. Therefore, the categorizing of feature
selection made it easier to find appropriate standards algorithm. The evaluation measure metric
gives better results compared to the existing method. Besides, our proposed architecture is
reducing overfitting problems in a model generation to improve our algorithm. This filtering
approach, feature, and fusion selection technique is used to provide better metric values compared
to another existing method [29]. Figure 7 is showing the final stage of our proposed architecture.

Figure 7 Final stage of Our Proposed Framework

The Random Forest algorithm is incorporated with our proposed structure based on
decision trees. This approach will be appropriated for our structure model to train the data very
fast. This is a very fast computation algorithm consist of more than 60 decision trees which is
shown in figure 2. Also, this approach classification process is used to construct the framework
with an optimized structure. Our proposed architecture is constructed based on the decision tree
method. This supervised learning algorithm is used to make an ensemble process for regression
[30]. This technique is used to combine the classification and predictions from the multiple input
machine learning algorithm for making accurate predictions by any single model. Feature vector
transformation is showing in figure 8.

9
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

Figure 8 Decision Tree at Proposed Algorithm

5. RESULTS DISCUSSION
Our proposed model compares with 350 samples to analyze for training and testing the
dataset. The measurement of class balance, classifier completeness, decision optimization, and
model estimation. Our proposed research work is splitting the data into train and test for
minimizing the overlap problem in classes. This segmentation approach provides highly
interpretable and classes are balanced with high accuracy [31]. The formula for measurement
metrics is given below as an equation description for our proposed model. Figure 9 shows the
distance calculation method at each interval of ECG signal for classification.

Figure 9 Distance calculation at interval of ECG Signal

Our framework procedure is an independent and data classification from 5 lead probe
sensor output as raw data. This structure performs on two decision trees with the help of the
Random Forest method to fuse the final classification decision. We achieved higher accuracy,
sensitivity, and minimum execution time of our algorithm. Except for computation time other
metric parameters are a bit lower than the hybrid version of the learning algorithm. But the

10
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

application of the proposed method is directly the medical field diagnosis sector [32][33].
Therefore, this sector should be provided the result with time constraints. The overall performance
of the model is improved and compared to the existing method for optimization purposes. These
measuring metrics are performing to find the effectiveness of class with the help of labels. The
result is calculated and tabulated in Table 1 below.

Table 1 Comparison of Our techniques

𝑇𝑃 + 𝑇𝑁
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 =
𝑇𝑃 + 𝑇𝑁 + 𝐹𝑃 + 𝐹𝑁
𝑇𝑃
𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦 =
𝑇𝑃 + 𝐹𝑃
𝑇𝑁
𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑖𝑡𝑦 =
𝑇𝑁 + 𝐹𝑃
The frequency power calculation of our proposed architecture is based on an estimation of
power spectrum performance. The interpolation of intervals between in the ECG signal is
calculated by integration of composite rule-based method. Our proposed architecture is very fast
as well as computation time is very less compared to the hybrid learning method. Our proposed
method is good in the overall performance chart with execution time. Because the medical field
needs to record results immediately. Therefore, the time constraint is very important. As per the
perception of point of view, our proposed architecture is fulfilling those tasks very successfully
compared to the hybrid version of the learning method. The comparison of performance metric
measures is showing in figure 10.

11
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

Figure 10 Comparison chart of Performance measure

6. CONCLUSION
Our proposed framework is concentrating on the dataset class imbalance problem which
arises in many algorithms during classification. The random forest is a more efficient method to
classify class-appropriate balancing segments. Our research article is showing ECG signal analysis
with an RF machine learning algorithm. This approach has processed including imbalanced class
and overfitting problem with suitable decision tree approach. Our research article is detected QRS
complexes from the input ECG signal to verify the specified features for classification. The authors
believe strongly this research article can help for a better understanding of medical field challenges
especially cardiovascular research. Besides the machine learning approach has given better clinical
advancement to the analysis of ECG signals.

Future work
The proposed architecture is used to help upgrade the automatic disease diagnosis
monitoring and predicting based on ECG recording further. The manual handling of the ECG
recording segment is a very large computation and more time expensive. One of the problems of
our proposed work occupies a huge amount of storage to record ECG signal and its processing
also. The higher number of data gives more accuracy leads to high dimension data storage. The
classifying of the huge amount of irrelevant data or redundant data set is a challenging task in our

12
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

proposed algorithm. We can be achieved better accuracy and computation time result by removing
those features and correlated vector function features.

REFERENCES
[1] Emelia, J.; Benjamin, S.S.V.; Clifton, W.; Callaway, A.M.C.; Alexander, R.; Chang, S.C.;
Stephanie, E.; Chiuve, M.C.; Francesca, N.; Delling, R.D.; et al. Heart Disease and Stroke
Statistics—2018 Update: A Report from the American Heart Association. Circulation 2018, 137,
e67–e492.
[2] Hannun, A.Y.; Rajpurkar, P.; Haghpanahi, M.; Tison, G.H.; Bourn, C.; Turakhia, M.P.; Ng,
A.Y. Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms
using a deep neural network. Nat. Med. 2019, 25, 65–69. [CrossRef]
[3]. Yang, C.; Veiga, C.; Rodriguez-Andina, J.J.; Farina, J.; Iniguez, A.; Yin, S. Using PPG Signals
and Wearable Devices for Atrial Fibrillation Screening. IEEE Trans. Ind. Electron. 2019, 66,
8832–8842. [CrossRef]
[4]. Lyon, A.; Mincholé, A.; Martínez, J.P.; Laguna, P.; Rodriguez, B. Computational techniques
for ECG analysis and interpretation in light of their contribution to medical advances. J. R. Soc.
Interface 2018, 15, 20170821. [CrossRef]
[5]. Singh, B.N.; Tiwari, A.K. Optimal selection of wavelet basis function applied to ECG signal
denoising. Digit. Signal Process. 2006, 16, 275–287. [CrossRef]
[6]. Friesen, G.M.; Jannett, T.C.; Jadallah, M.A.; Yates, S.L.; Quint, S.R.; Nagle, H.T. A
comparison of the noise sensitivity of nine QRS detection algorithms. IEEE Trans. Biomed. Eng.
1990, 37, 85–98. [CrossRef]
[7]. Van Alste and, T.S.; Schilder, J.A. Removal of base-line wander and power-line interference
from the ECG by an efficient FIR filter with a reduced number of taps. IEEE Trans. Biomed. Eng.
1985, BME-32, 1052–1060. [CrossRef]
[8]. Oster, J.; Behar, J.; Sayadi, O.; Nemati, S.; Johnson, A.E.W.; Clifford, G.D. Semisupervised
ECG Ventricular Beat Classification with Novelty Detection Based on Switching Kalman Filters.
IEEE Trans. Biomed. Eng. 2015, 62, 2125–2134. [CrossRef]

13
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

[9]. Dalin Tang, Z.T.; Canton, G.; Hatsukami, T.S.; Dong, L.; Yuan, X.H.A.C. Local critical stress
correlates better than global maximum stress with plaque morphological features linked to
atherosclerotic plaque vulnerability: An in vivo multi-patient study. Biomed. Eng. Online 2009, 8,
15. [CrossRef]
[10]. Frolich, L.; Dowding, I. Removal of muscular artifacts in EEG signals: A comparison of
linear decomposition methods. Brain Inf. 2018, 5, 13–22. [CrossRef] [PubMed]
[11]. Blanco-Velasco, M.; Weng, B.; Barner, K.E. ECG signal denoising and baseline wander
correction based on the empirical mode decomposition. Comput. Biol. Med. 2008, 38, 1–13.
[CrossRef]
[12]. Alfaouri, M.; Daqrouq, K. ECG Signal Denoising By Wavelet Transform Thresholding. Am.
J. Appl. Sci. 2008, 5, 276–281. [CrossRef]
[13]. Xu, Y.; Luo, M.; Li, T.; Song, G. ECG Signal De-noising and Baseline Wander Correction
Based on CEEMDAN and Wavelet Threshold. Sensors 2017, 17, 2754. [CrossRef] [PubMed]
[14]. Han, G.; Xu, Z. Electrocardiogram signal denoising based on a new improved wavelet
thresholding. Rev. Sci. Instrum. 2016, 87, 084303. [CrossRef]
[15]. Pinho, A.; Pombo, N.; Silva, B.M.C.; Bousson, K.; Garcia, N. Towards an accurate sleep
apnea detection based on ECG signal: The quintessential of a wise feature selection. Appl. Soft
Comput. 2019, 83, 105568. [CrossRef]
[16] Moavenian, M.; Khorrami, H. A qualitative comparison of Artificial Neural Networks and
Support Vector Machines in ECG arrhythmias classification. Expert Syst. Appl. 2010, 37, 3088–
3093. [CrossRef]
[17]. Appathurai, A.; Jerusalin Carol, J.; Raja, C.; Kumar, S.N.; Daniel, A.V.; Jasmine Gnana
Malar, A.; Fred, A.L.; Krishnamoorthy, S. A study on ECG signal characterization and practical
implementation of some ECG characterization techniques. Measurement 2019, 147, 106384.
[CrossRef]
[18]. Chauhan S, Vig L. 2015 Anomaly detection in ECG time signals via deep long short-term
memory networks. In 2015 IEEE International Conference on Data Science and Advanced
Analytics (DSAA), pp. 1 – 7. 19– 21 October, Paris, France. Piscataway, NJ: IEEE.
(doi:10.1109/DSAA.2015.7344872).

14
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

[19]. Oresko JJ, Jin Z, Cheng J, Huang S, Sun Y, Duschl H, Cheng AC. 2010 A wearable
smartphone-based platform for real-time cardiovascular disease detection via electrocardiogram
processing. IEEE Trans. Inf. Technol. Biomed. 14, 734– 740. (doi:10.1109/TITB.2010.2047865)
[20] J. S. Wang, W. C. Chiang, Y. T. Yang, and Y. L. Hsu, "An effective ECG arrhythmia
classification algorithm," Bio-Inspired Computing and Applicat, Springer Berlin Heidelberg, pp.
545-550, 2012.
[21] A. Dallali, A. Kachouri, and M. Samet, "Classification of Cardiac Arrhythmia Using WT,
HRV, and Fuzzy C-Means Clustering," Signal Processing: An Int. J. (SPJI), vol. 5, no. 3, pp. 101-
109, 2011.
[22] M. Vijayavanan, V. Rathikarani, and P. Dhanalakshmi, “Automatic Classification of ECG
Signal for Heart Disease Diagnosis using morphological features,” Int. J. of Comput. Sci. and Eng.
Technology (IJCSET), vol. 5, no. 4, pp. 449-455, 2014.
[23] V. K. Srivastava and D. Prasad, "Dwt-Based Feature Extraction from ecg Signal," American
J. of Eng. Research (AJER), vol. 2, no. 3, pp. 44- 50, 2013.
[24] M. Korurek and B. Dogan, "ECG beat classification using particle swarm optimization and
radial basis function neural network," Expert syst. with Applicat., vol. 37, no. 12, pp. 7563-7569,
2010.
[25]. Üstünda ˘g, M.; Gökbulut, M.; ¸Sengür, A.; Ata, F. Denoising of weak ECG signals by using
wavelet analysis and fuzzy thresholding. Netw. Modeling Anal. Health Inform. Bioinform. 2012,
1, 135–140. [CrossRef]
[26] A. T. Sadiq and N. H. Shukr, "Classification of Cardiac Arrhythmia using ID3 Classifier
Based on Wavelet Transform," Iraqi J. of Sci., vol. 54, no. 4, pp. 1167-1175, 2013.
[27] N. P. Joshi and P. S. Topannavar, “Support vector machine based heartbeat classification,”
Proc. of 4th IRF Int. Conf., pp. 140-144, 2014.
[28] E. Zeraatkar et al., "Arrhythmia detection based on Morphological and time-frequency
Features of t-wave in Electrocardiogram," J. of medical signals and sensors, vol. 1, no. 2, pp. 99-
106, 2011.
[29] S. N. Yu and K. T. Chou, "Integration of independent component analysis and neural networks
for ECG beat classification," Expert Syst. with Applicat., vol. 34, no. 4, pp. 2841-2846, 2008.

15
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021
Journal of Artificial Intelligence and Capsule Networks (2021)
Vol.03/ No.01
Pages: 1-16
http://irojournals.com/aicn/
DOI: https://doi.org/10.36548/jaicn.2021.1.001

[30] D. Joshi and R. Ghongade, “Performance analysis of feature extraction schemes for ECG
signal classification,” Int. J. of Elect., Electron. and Data Commun., vol. 1, pp. 45-51, 2013.
[31] M. K. Das and S. Ari, "ECG Beats Classification Using Mixture of Features" Int. Scholarly
Research Notices, 2014.
[32] X. Tang and L. Shu, "Classification of Electrocardiogram Signals with RS and Quantum
Neural Networks," Int. J. of Multimedia and Ubiquitous Eng., vol. 9, no. 2, pp. 363-372, 2014.
[33] Wang, Zh., Yan, W., Oates, T. Time Series Classification from Scratch with Deep Neural
Networks: A Strong Baseline // arXiv preprint. arXiv:1611.06455, 2016.

16
ISSN: 2582-2012 (online)
Submitted: 29.12.2020
Revised: 2.02.2021
Accepted: 25.02.2021
Published: 11.03.2021

You might also like