Classification of Arrhythmia using Electrocardiogram based Image Features with Deep Learning

Digital health technology leverages artificial intelligence for real-time monitoring of patients' arrhythmia conditions. Heartbeat data are captured via an electrocardiogram (ECG) and processed to extract pivotal features of each heartbeat. These extracted characteristics serve as the foundation for the training of data-driven models, facilitating the detection and measurement of arrhythmias. The success and reliability of these algorithmic decision-making systems heavily depend on the quality of features extracted, which typically necessitate significant empirical and domain-specific knowledge. To address these challenges, the current work introduces a novel method for arrhythmia classification that hinges on image features, derived using deep learning techniques, with the aim of augmenting the accuracy of the decision-making algorithms. Denoised one-dimensional (1D) ECG signals corresponding to various arrhythmia conditions are transformed into scalograms via the continuous wavelet transform (CWT) process. These scalograms are then classified with the help of a convolutional neural network (CNN), a powerful deep learning model. The proposed methodology exhibits an impressive classification accuracy of 91.52%, underscoring its potential effectiveness. This approach opens avenues for the establishment of trustworthy healthcare systems, enabling precise monitoring of patients' cardiac health, thereby facilitating timely and appropriate intervention.