Publication: Classification of arrhythmia signals by using convolutional neural network
Date
2021-07-01
Authors
Evelyn Siao, Yung Ern
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Arrhythmia is one of the fatal cardiovascular diseases. Arrhythmia is a problem with the rate of the heartbeat that either beats too fast, or too slowly, or with an irregular rhythm. Atrial fibrillation (AF) is the most common arrhythmia that can be diagnosed by using an electrocardiogram (ECG) pattern. Identification of arrhythmia through ECG can be very challenging if we fully depend on experts because it is very time-consuming. This paper presents the use of deep learning in helping the detection of arrhythmia using one-dimensional (1-D) input. Deep learning is preferable compared to standard neural networks because it has higher accuracy of disease detection. This is because while neural
networks use neurons to transmit data in the form of input values and output values through connections, deep learning is associated with the transformation and extraction of features. 1-D signals are used instead of 2-D images because 1-D is less complex. The experimental data is obtained from the Physionet CinC Challenge 2017 Database. The ECG signals undergone pre-processing including filtering, QRS detection, segmentation, and median wave selection. In this project, 1-D convolutional neural network (CNN) model, CNN-Long Short Term Memory (CNN-LSTM) hybrid model, and CNN-Bidirectional LSTM (CNN-biLSTM) hybrid model are developed to classify ECG signals into four classes, which are the normal rhythm, AF rhythm, other rhythms, and noisy signal. The proposed 1-D CNN model achieved an accuracy of 91.67%, the CNN-LSTM model achieved 82.33% of accuracy, while the 1-D CNN-biLSTM hybrid model achieved 94.67% of accuracy. From the experimental results, it shows that the proposed CNN models are able to aid AF diagnosis in clinics.