FINDING OF ARRYTHMIA BY 2D IMAGE

CLASSIFICATION
Selvanathan N#1 Kaavya Lakshmanan#2 Agallya A N#3 Diloshaa Sri R#4 Dharani Shree A S#5
#1
ASSISTANT PROFESSOR, DEPARTMENT OF INFORMATION TECHNOLOGY, SONA COLLEGE OF
TECHNOLOGY, SALEM, TAMIL NADU.
#2#3#4#5
STUDENTS, DEPARTMENT OF INFORMATION TECHNOLOGY, SONA COLLEGE OF
TECHNOLOGY, SALEM, TAMIL NADU.
#1
[email protected],#2 [email protected], #[email protected], #[email protected],
#5
[email protected]

ABSTRACT-- The purpose of this study is I. INTRODUCTION

to examine the viability of using 2-
Dimentional Electrocardiogram image Any unusual deviation from the regular
detection for arrhythmia classification. The beat of the heart is referred to as an
suggested method involves employing arrhythmia, which is a type of
time-frequency analysis and wavelet cardiovascular disease. Atrial fibrillation,
transform to transform the ECG signal into premature contraction, ventricular
a 2-Dimentional picture representation. fibrillation, and tachycardia are only a few
Convolutional neural networks are used to examples of arrhythmia. Continuous
automatically classify arrhythmia using the arrhythmia beats can lead to deadly
output images. 10,000 2-Dimentional situations, even though a single
Electrocardiogram pictures from 1000 arrhythmia heartbeat may not have a
patients with 10 different forms of serious influence on life.
arrhythmia make up the dataset used in
this study. With sensitivity and specificity Here we are using Convolution Neural
ranging from 87% to 100%, the CNN Network (CNN) model. Whereas the
model classified arrhythmias with an CNN model is used for Heartbeat
overall accuracy of 95%. The findings Classification based on ECG Image to
point to the proposed methodology as a classify various forms of Arrhythmia.
potential one for 2-Dimentional
Electrocardiogram picture identification The purpose for building this
and automated arrhythmia classification. classification proposed model for making
an intelligent output that can categorise
Keywords- Electrocardiogram, 2- any heartbeat signal according to a certain
Dimentional Image, Convolutional Neural form of arrhythmia. The project's
Network, Deep Learning, Image accuracy efficiency and competitiveness
Classification, Arrhythmia. in terms of sensitivity and specificity are
both demonstrated by the final findings.
We are classifying six types of enhanced by the capacity to extract
arrhythmias among 10 forms of features specific to the patient.[3]
arrhythmia.

II. LITRATURE SURVEY III. PROBLEM STATEMENT

[1] An exact scientific classification of The objective of this research is to create

Electrocardiogram signals is incredibly a model that can correctly categorise
supportive in the counteraction and various arrhythmias, including
determination of CVDs. Profound model bradycardia, tachycardia, and atrial
demonstrated valuable in upgrading the fibrillation, using 2D ECG pictures. The
exactness of determination calculations in outcomes of this study could have a big
the combination of medication and impact on how arrhythmias are identified
present day AI advances. The proposed and treated early, which could eventually
CNN-based characterization calculation, lead to better patient outcomes and lower
utilizing 2-D pictures, can group eight healthcare expenditures.
sorts of arrhythmia, specifically, NOR,
VFW, PVC, VEB, RBB, LBB, PAB, and IV. METHODOLOGY
APC, and it accomplished 97.91% normal The suggested method uses a
awareness, 99.61% particularity, 99.11% Convolutional Neural Network (CNN)
normal exactness, and 98.59% positive which is trained on two dimensional
prescient worth (accuracy). These method whereas the electrocardiogram
outcomes show that the expectation and beat pictures is to identify different types
arrangement of arrhythmia with 2-D ECG of arrhythmias. ECG signals, which are
portrayal [1] split into heartbeats and comprise six
[2] It is noteworthy that the suggested different types of arrhythmias, are first
CNN network has considerably higher converted into 2D grayscale pictures. The
accuracy and robustness when compared photos are then sent into a new CNN
to the BP neural network, random forests, architecture that is being trained to
and other CNN networks. For the identify heartbeats.
characterization of the miniature classes Here we are done our project by using
of the Arrhythmia dataset, the proposed Jupyter notebook and spyder. Jupyter
CNN network shows extraordinary notebook is to train the dataset and save
execution with a general arrangement our model in a file. For user interface we
exactness of 97.41%, awareness of are created a web page by using html,
97.05%, particularity of 99.35%, and bootstrap, css where the trained file and
positive expectation pace of 97.21%.[2] html are connected to spyder.
[3] This study describes a rapid and To train our data we collected the datasets
precise electrocardiogram monitoring and form the Kaggle which has set of data
classification system that is patient- collections that contain images of 2D
specific. An individual and simple CNN grayscale images. We use that dataset to
will be created for each patient utilising a train our model which we have build in
small amount of common and patient- Jupyter notebook. There we first import
specific training data.. As a result, the our models like tensorflow and keras for
classification performance can be further
our in order to perform Machine Learning image but choosing the ecg image to
(ML) tasks, we combine different layer check which type of arrhythmia that they
types to create a model that may be used have.
to train using data to forecast future
values. TESTING

For image pre-processing we copy the path TYPES OF LAYERS:

of our file location to check the number of A convolution neural network has
classes that available in the dataset. Then numerous hidden layers that help in
we created our model (CNN) information extraction from a picture.
Convolutional Neural Network and took CNN's three key layers and they are
the summary of it. Then we compile our Convolution , ReLU and Pooling layers.
model to get the accuracy next we fit our
model and run to check the epoch
CONVOLUTION LAYER-- Take into
accuracy whether the given dataset fit or
account the following 5x5 image, where
overfitting the file then we saved our
model. each pixel's value is either 0 or 1. A filter
matrix with a 3x3 dimension is also
After saving the model we predict our present. To obtain the convolved feature
result to check by uploading the path of a
matrix, move the filter matrix over the
single sample image in the dataset. In
below diagram we have shown the flow of image and compute the dot product.
our work that done in Machine Learning. Figure 2 shows us the diagrammatic
This figure.1 helps us to learn about the representation of this layer.
process easily.

Figure.2: Convolution layer

RELU LAYER-- The rectified linear unit

is referred to as ReLU. The next step is to
transfer the feature maps to a ReLU layer
Figure.1: Flow diagram after they have been retrieved.
For user interface we have created three The negative pixels are all set to 0 by
pages using html, css and bootstrap which ReLU, which performs an operation
contain home page, information page and element-by-element. Figure 3 graph of a
the predict page. Home page shows the ReLU function is shown below.
basic information about the Arrhythmia
disease and the info page contain the
details of six types of arrhythmia and in
the predict page the user can upload the
 Figure 5: Home page

Figure.3: ReLU layer

POOLING LAYER-- The down sampling

process of pooling lowers the feature
Figure 6: Info page
map's dimensionality. The corrected
feature map is now run through a layer of
pooling to create a pooled feature map.
Figure 4 shows us how the pooling layer
works on our project.

Figure 7: Prediction page

VI. CONCLUSION

The irregular beats and slow heartbeat

cause major problems like Arrhythmia,
Stroke etc., Here we suggested a 2-D
Figure 4: Pooling layer CNN-based classification algorithm to
automatically classify cardiac arrhythmias
using ECG signals. When modern machine
V. RESULT learning technology and medicine are
combined, Deep CNN has been effective
The accuracy of the result is done by in increasing the precision of diagnosis
using machine learning and the algorithm algorithms. Six different types of
we used is CNN which is highly efficient arrhythmia can be classified using the
us to provide a better accuracy over 98% proposed CNN-based classification system
with 10 epoch we got such an output. using 2-D pictures. And finally an average
Figure 5,6,7 shows our project home, info accuracy of 98% was attained.
and predict pages.
 VII. REFERENCE [8] The future of remote ECG
monitoring systems. J Geriat.
[1] Classification of Arrhythmia by Cardiol. 13, 528. doi:
Using Deep Learning with 2-D 10.11909/j.issn.1671-
ECG Spectral Image 5411.2016.06.015. Guo, S.-L.,
Representation Amin Ullah Han, L.-N., Liu, H.-W., Si, Q.-
1,2 , Syed Muhammad Anwar J., Kong, D.-F., and Guo, F.-S.
1,2 , Muhammad Bilal 3 and (2016).
Raja Majid Mehmood 4
[9] ECG arrhythmia classification
using transfer learning from 2-
[2] A Study on Arrhythmia via dimensional deep CNN
ECG Signal Classification features. Salem, M.; Taheri, S.;
Using the Convolutional Neural Yuan, J.S. In Proceedings of
Network Mengze Wu1, the 2018 IEEE Biomedical
Yongdi Lu2, Wenli Yang3 and Circuits and Systems
Shen Yuong Wong2* 05 Conference (BioCAS),
January 2021 Cleveland, OH, USA, 17–19
October 2018; IEEE:
[3] [3]Real-Time Patient-Specific Piscataway, NJ, USA, 2018;
ECG Classification by 1-D pp. 1–4.
Convolutional Neural
Networks, Serkan Kiranyaz; [10] “Global public health
Turker Ince; Moncef Gabbouj problem of sudden cardiac
death,” R. Mehra, Journal of
[4] Heartbeat sound signal Electrocardiology, vol. 40, no.
classification using deep 6, pp. S118–S122, 2007.
learning. Raza A, Mehmood A,
Ullah S, et al. [11] “Detecting pulmonary
embolism using deep neural
networks” Akilandeswari, J.,
[5] Pattern Recognition of the
Jothi, G., Naveenkumar,
ECG, J.L.Talmon, Berlin,
A., ...Iyyanar, P., Paramasivam,
Germany: Akademish M.E. International Journal of
Proefscript, 1983. Performability Engineering,
2021, 17(3), pp. 322–332

[6] A comparison of 1-D and 2-D [12] Brain Tumor Detection

deep convolutional neural Using Deep Learning Neural
networks in ECG classification. Network for Medical Internet
Wu, Y.; Yang, F.; Liu, Y.; of Things Applications
Zha, X.; Yuan, S. Rajkumar, S., Karthick, K.,
Selvanathan, N., ...Murali, M.,
[7] Applications of artificial neural Dhiyanesh, B. Proceedings of
networks for ECG signal the 6th International
detection and classification. J. Conference on Communication
Electrocardiol. 1993, 26, 66– and Electronics Systems,
73. Hu, Y.H.; Tompkins, W.J.; ICCES 2021, 2021, 9489187
Urrusti, J.L.; Afonso, V.X.