REVIEW PAPER ON

ARRHYTHMIA CLASSIFICATION USING

DEEP-LEARNING
 JSS MAHAVIDYAPEETHA
JSS ACADEMY OF TECHNICAL EDUCATION
JSS Campus Dr. Vishnuvardhana Road, Bengaluru – 560 060
DEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERING

Arrhythmia classification using Deep-Learning

Batch No:24
NAME USN
Bhargav Kodali 1JS17EC025

Amol Mishra 1JS16EC005

Aravindh kumar V 1JS17EC118
Ajay kumar V 1JS17EC402

 UNDER THE GUIDANCE OF

Dr. S.T.Veerebhadrappa
Associate Professor
&
Ms. Suguna G.C
Asst. Professor
DEPT. OF E&C
JSSATE - BANGALORE - 60

2
 OVERVIEW
3

1. Introduction
2. Literature Survey
3. Objective
4. Back propagation
5. Introduction to CNN
6. Methodology
7. Reference
 INTRODUCTION
4

 According to World Health Organization (WHO), Cardio Vascular

Diseases are the main reasons which causes death world wide
 The electrical activity of heart is represented as Electrocardiograph (ECG) and is
measured using Electrocardiogram .
 The ECG with irregular rhythm and rate of heart beats is known as Arrhythmia .
 Ex : Atrial Fibrillation , Ventricular Fibrillation , Ventricular Tachycardia .
 Using the ECG waveform we can diagnose various abnormalities in our heart .
 Any variation in the waveform such as increase or decrease in values of amplitude
and time period it show that there is a presence of abnormality in the heart .
 Arrhythmia mainly occurs due to 2 reasons i.e.,
physical & emotional stress.
 5

 Since ECG signal is structured data in normal Deep-learning(DL) plays an important

role in classification of arrhythmia.
 Now a days, Data science becoming popular in all fields such as automobiles , medial
fields etc., so here we are using DL to classify the arrhythmias .
 This include DL basic algorithms like convolutional neural network (CNN),
Recurrent Neural Network (RNN) which is more efficient in classifying ECG signals .
 We build an ensemble model which combines the single lead and twelve (12) lead
ECG model.
 Both the models uses CNN and RNN because the features are learnt automatically
from the ECG signals.
 The data is collected from china physiological challenge (CPSC) 2018 DataSet .
 PHYSIOLOGY OF HEART
6

Source : Internet
 ECG WAVEFORM
7

Source : Internet
 LITERATURE SURVEY
8

Sl.no Researchers Title Proposed model Data base Accuracy

1. Mathunjwa ECG arrhythmia 2D recurrent plot PhysioNet’s 1 stage : 95.3
,et.al classification using with 2 stages they MIT-BIH AR, ± 1.27%
a recurrent plot compared Different AF, VT. 2 stage :
and convolution CNN architecture such 95.41 ±
neural network as AlexNet 0.11%
(2020) in ,VGG(Visual
Biomedical signal Geometry Group) 16
processing and &VGG 19
control journal of
Elsevier
 9

Sl.no Researchers Title Proposed model Data base Accuracy

2. Ruxin Wang Deep Multi-Scale Deep Multi-Scale CPSC - 2018 F1 scores
et.al Fusion Neural Fusion convolution (12-lead 82.8%,84.1%
network for multi- neural Network dataset), For respective
class arrhythmia (DMSFNet) Physionet/CinC datasets
detection(2020) in 2017 (1 -lead
IEEE journal dataset)
biomedical and
health informatics
 10

Sl.no Researchers Title Proposed model Data base Accuracy

3. Abdalla .F.Y.O Deep CNN - 11 layers MIT-BIH 93.84%
et al. convolutional • Convolutional ,
neural network Max pooling -4
application to layers each,
classify the ECG • Fully connected
arrhythmia(2019) layer -3
In signal, Image • ReLU(Rectified
and Video Layer Unit) &soft-
processing of max function used
Springer for classification
 11

Sl.no Researchers Title Proposed model Data base Accuracy

4. Hannun A Y, Cardiologist-level Network architectures iRhythm Receiver
Andrew Ng Y arrhythmia - 34 layers Technologies, Operating
et.al classification in • Optimization of inc Curve(ROC)-
ambulatory network tracking 97%
Electrocardiogram • Residual blocks-16 F1 score- 83.7
s using a Deep 2 convolution layer
neural in each block
network(2019) in
Nature medicine
journal of Nature
publishing group
 12

Sl.no Researchers Title Proposed model Data base Accuracy

5 Alfaras M. A fast Machine Echo State MIT-BIH AR, • Single lead
et al, Learning model Network(ESN) with AHA II
for ECG-Based on ring topology Sensitivity-
Heart beat 92.7%
classification and Positive
Arrhythmia predictive
detection(2019) in value - 86.1%
“Frontiers in • Lead V1’
Physics journal” Sensitivity -
by Frontiers 95.7%
Positive
predictive
value - 75.1%
 OBJECTIVE
13

The main objective of this project is to analyze ECG and classify different
types of arrhythmia with efficient algorithm using CNN and RNN
Backpropagation Algorithm
 In order to have some numbers to work with, here are the initial weights, the biases, and training
inputs/outputs:
 The goal of backpropagation is to optimize the weights so that the neural network can learn how to correctly
map arbitrary inputs to output
 The Forward Pass

We then use logistic function to calculate

output of h1.

Similarly for output of h2,we get.

 Calculating the Total Error

The target output o1 is 0.01 but the neural network output is

0.75136507
 The Backwards Pass

We want to know how much change in w5 effects the total error
 First, how much does the total error change with respect to the output

The partial derivative of the logistic function is the output multiplied by 1 minus the
output:

Finally how much does the total net output change with respect to w5.
 21

Finally how much does the total net output change with respect to
w5.

Putting it alltogether
 22

To decrease the error, we then subtract this value from the current
weight (optionally multiplied by some learning rate, eta, which we’ll set
to 0.5):
 Hidden Layer

Next, we’ll continue the backwards pass by calculating new values

for w1,w2,w3,w4
Finally, we’ve updated all of our weights!. . It might not seem like much, but
after repeating this process 10,000 times, for example, the error plummets to
0.0000351085. At this point, when we feed forward 0.05 and 0.1, the two
outputs neurons generate 0.015912196 (vs 0.01 target) and 0.984065734 (vs
0.99 target).
INTRODUCTION TO CNN
25
 26

A CNN is a feed forward neural network that is used to analyze visual

images by processing data with a grid like topology. Convolution
operation forms the basis of convolution neural network. Every image is
represented in the value of pixels.
 CONVOLUTION LAYER
27

Consider a Sample 5x5 matrix whose pixel values are 0 and 1.The
patterns are detected by sliding the filter matrix over the image and
computing the dot product.
1 1 1 0 0 4 3 4
1 0 1
0 1 1 1 0 2 4 3
0 1 0
0 0 1 1 1
1 0 1 2 3 4
0 0 1 1 0
0 1 1 0 0 Filter Convolved
Image pixels features
 28

 1 1 1 0 0
0 1 1 1 0 1 0 1 4
0 0 1 1 1 0 1 0
0 0 1 1 0 1 0 1
0 1 1 0 0

1 1 1 0 0 1 0 1 4 3
0 1 1 1 0 0 1 0
0 0 1 1 1
1 0 1
0 0 1 1 0
0 1 1 0 0
 29

 1 1 1 0 0
1 0 1 4 3 4
0 1 1 1 0
0 0 1 1 1 0 1 0
0 0 1 1 0 1 0 1
0 1 1 0 0

1 1 1 0 0 1 0 1 4 3 4
0 1 1 1 0 0 1 0 2
0 0 1 1 1
1 0 1
0 0 1 1 0
0 1 1 0 0
 ReLU LAYER
30

Once the features are extracted the next step is to move them to the ReLU
Layer.
This layer sets zero all the pixel values which are lesser than zero and
preserves all the pixel values which are greater than zero.
 Pooling layer
31

The rectified feature map goes through a pooling layer.

Pooling is a down sampling operation that reduces the dimensionality of the
feature map and helps to identify the features in a image.

Rectified feature map

pooled feature map
1 4 2 7
6 8
2 6 8 5
3 4 0 7 4 7
1 2 3 2

Max(3,4,1,2)
BIDIRECTIONAL RECURRENT NEURAL NETWORK
Bidirectional Recurrent Neural Networks (BRNN) connect two hidden layers
of opposite directions to the same output. With this form of generative deep
learning, the output layer can get information from past (backwards) and
future (forward) states simultaneously

For ex:-
He said “Teddy Roosevalt is a president”
He said “Teddy Bear is on sale”
METHODOLOGY
34
 35

Our model is constructed on a collective construction of CNN blocks,

supported by an attention layer, bidirectional Gated Recurrent Unit
(GRU) which is modern type of RNN , and finally a dense, i.e. fully
connected, layer.
Within each CNN block there were two convolution layers and they are
supported by a pooling layer to decrease the amount of parameters and
computation in the system and monitor over-fitting.
 We determined to utilize CNN and RNN because of their demonstrated
capability to manage time series data and noisy signals in studies which
included ECG classification .
 36

GRU is proposed lately that can need fewer training time and a smaller
amount of number of iterations.
We utilized batch normalization to adapt and scale the input from the
attention layer, which defines a vector of significance weights, to the
dense layer.
The subjects are at random split into 10 equal parts to put up an 8-1-1
train, validation and test arrangement of machine learning.
Some models are constructed based on the execution on the validation
set. The models are further evaluated by calculating its F1 score on the
test set. Further by use this model for evaluating on single lead data.
 37

To do that, for a given lead we just given zero to everyone the ECG values of the other 11

leads and developed the model using the same network architecture and similarly the models
are further evaluated by calculating its F1 score.
Evaluation Criteria of our model
The recall rate, precision and F1 score are implemented for determining the classification
performances.
Precision = T P/ (T P + F P)
Recall = T P /(T P + F N)
F1 = 2 × (Precision × Recall) /(Precision + Recall)
Where, TP - True Positive
FP - False Positive
FN - False Negative
 REFRENCE
38

1. Chen, Tsai-Min, et al. "Detection and Classification of Cardiac Arrhythmias by a Challenge-Best Deep Learning
Neural Network Model." Iscience 23.3 (2020): 100886
2. Mathunjwa, Bhekumuzi M., et al. "ECG arrhythmia classification by using a recurrence plot and convolutional
neural network." Biomedical Signal Processing and Control 64 (2020): 102262.
3. Wang, Ruxin, Jianping Fan, and Ye Li. "Deep Multi-Scale Fusion Neural Network for Multi-Class Arrhythmia
Detection." IEEE journal of biomedical and health informatics 24.9 (2020): 2461-2472.
4. Wang, Ruxin, Jianping Fan, and Ye Li. "Deep Multi-Scale Fusion Neural Network for Multi-Class Arrhythmia
Detection." IEEE journal of biomedical and health informatics 24.9 (2020): 2461-2472
5. Abdalla, Fakheraldin YO, et al. "ECG arrhythmia classification using artificial intelligence and nonlinear and
nonstationary decomposition." Signal, Image and Video Processing 13.7 (2019): 1283-1291
6. Hannun, Awni Y., et al. "Cardiologist-level arrhythmia detection and classification in ambulatory
electrocardiograms using a deep neural network." Nature medicine 25.1 (2019): 65
7. Rajkumar, A., M. Ganesan, and R. Lavanya. "Arrhythmia classification on ECG using Deep Learning." 2019 5th
International Conference on Advanced Computing & Communication Systems (ICACCS). IEEE, 2019
8. Mustaqeem, Anam, Syed Muhammad Anwar, and Muahammad Majid. "Multiclass classification of cardiac
arrhythmia using improved feature selection and SVM invariants." Computational and mathematical methods in
medicine 2018 (2018).
9. Kachuee, Mohammad, Shayan Fazeli, and Majid Sarrafzadeh. "Ecg heartbeat classification: A deep transferable
representation." 2018 IEEE International Conference on Healthcare Informatics (ICHI). IEEE, 2018
THANK YOU
for spending valuable time

39