Modified Deep Learning Models for ECG Heart Beat Classification
This paper introduces a deep learning approach to classify Electrocardiograms (ECG) signals. It is necessary to detect early and accurate types of arrhythmias in the management of myocardial dysfunction and providing proper medication at the right time. The proposed solution uses the MIT-BIH Arrhythmia Dataset. The emphasis is on the scanning and classification of the patient’s heartbeats into Non-Ectopic beats (Normal beats), Supraventricular Ectopic beats, Ventricular Ectopic beats, Fusion Beats and Atrial Premature Contraction beats. To overcome the problem of imbalanced data, resampling and augmentation have been performed on the dataset. Convolutional neural network (CNN) is used to develop three different models by altering the architecture based on VGG Net, AlexNet, LeNet, and Long Short-Term Memory with Fully Convolutional Network (LSTM-FCN). Batch normalization is implemented in each architecture which has significantly reduced the chance of overfitting on data and also reducing the overall time for training. The model’s efficiency is measured on the basis of accuracy, specificity and sensitivity. All the proposed models show robust performance with short training time, out of which AlexNet outperforms with an average accuracy of 99%, specificity of 99.74% and sensitivity of 99.72%. The proposed methodology is capable of real-time prediction of ECG signals, which will enhance the effectiveness of clinical treatment and prevention of its serious complications.