A Novel and Efficient CNN Architecture for Detection and Classification of ECG Arrhythmia

Cardiac arrhythmia is a result of irregular beating of the heart and occurs when the electrical impulses in our heart fail to send signals regularly. Depending on the type, arrhythmia can cause fainting or dizziness and in some cases, even heart failure which can be fatal. Early and quick detection of arrhythmia can be quite beneficial in many situations, and thus, there has been a surge in research on using artificial intelligence to counter this problem. Until recently, researchers have been using traditional machine learning algorithms to process ECG signals and classify them into different types of arrhythmias. In this manuscript, we propose a novel convolutional neural network (CNN) architecture to classify ECG signals after converting them into two-dimensional images. Our process mirrors the procedure used by medical practitioners in the real world to detect arrhythmia. We also implemented three popular CNNs—DenseNet 169, ResNet-50, and MobileNet V1 to evaluate and compare our proposed model against them. Our model demonstrated better performance than these networks on our evaluation metrics and will be useful for future tasks of ECG arrhythmia classification.