Electrocardiogram based arrhythmia classification using wavelet transform with deep learning model

High-risk patients of cardiovascular disease can be provided with computerized electrocardiogram (ECG) devices to detect Arrhythmia. These require long segments of quality ECG which however can lead to missing the episode. To overcome this, we have proposed a deep-learning approach, where the scalogram obtained by continuous wavelet transform (CWT) is classified by the network based on the signature corresponding to arrhythmia. The CWT of the recordings is obtained and used to train the 2D convolutional neural network (CNN) for automatic arrhythmia detection. The proposed model is trained and tested to identify five types of heartbeats such as normal, left bundle branch block, right bundle branch block, atrial premature, and premature ventricular contraction. The model shows an average sensitivity, specificity, and accuracy to be 98.87%, 99.85%, and 99.65%, respectively. The result shows that the proposed model can detect arrhythmia effectively from short segments of ECG and has the potential for being used for personalised and digital healthcare.