An Arrhythmia Classification on ECG Signals Using Convolutional Neural Network and Long-Short Term Memory

Authors

  • Mr.Sreehari Kundella, Dr.R. Gobinath

Abstract

The rate of heartbeat disorder is knows as an Arrhythmia. At the time of an arrhythmia the heart beat rate would be high, suddenly reduces and shows an irregularity in the rate. The two major types of arrhythmia are tachycardia and bradycardia. If the heart beat frequency is high, then the condition is known as tachycardia. If the heart beat frequency is low, then the condition is known as bradycardia. In this paper, arrhythmias ECG signal preprocessing and classification has done. The new approach comprised noise reduction, feature extraction and classification with finally tuned parameters.  The study is comparing the classification algorithms for the large database consisting of 12 lead ECG from the thousands of patients with classes of arrhythmia. In this research, we have done the arrhythmia ECG signal preprocessing that contains, noise reduction, feature selection, feature extraction and classification. In this research, ECG data used that is MIT-BIH Arrhythmia Dataset is publicly available. MIT – BIH Arrhythmia dataset is the large dataset that contains the ECG recordings. The heart beat classifier has assessed by using of the MIT-BIH arrhythmia dataset using machine learning approach. In this research, the classification of ECG signals has done through CNN and LSTM (Convolutional Neural Network and Long – Short Term Memory). The main objective of the paper is classification of ECG signals using CNN and LSTM. Arrhythmia includes tachycardia, bradycardia, supraventicular arrhythmia, and ventricular arrhythmia etc. These are encouraged researchers to do research on arrhythmia dataset that includes ECG signals using some deep learning techniques like Convolutional Neural Network (Convolutional NN), Classic Neural Network (Classic NN), and Recurrent Neural Network (RNN) etc. The method classified 5 classes in arrhythmia with high accuracy of 97.55% and 98% using CNN and LSTM respectively. In this paper, Feature extraction and feature selection are done by gradient booting detection and random forest tree. CNN and LSTM are used for classification of arrhythmia dataset.

Keywords- Big Data, Convolution Neural Network, Long Short Term Memory, Recurrent Neural Network, ECG, Arrhythmia, Random Forest Tree, Gradient Boosting Tree.

Published

2020-12-07

Issue

Section

Articles