ECG Beat Classification using CNN

This paper proposes an approach to classify cardiac arrhythmias according to the morphology of the electrocardiogram signal (ECG), using deep machine learning methods. Two hierarchical levels for classification are proposed, the first level classifies normal and abnormal beats, and the second level deals with the problem of multi-classification between classes of abnormal beats. The classifier is a U-Net convolutional neural network (CNN) architecture applied for feature extraction and classification of ECG arrhythmias acquired from MIT-BIH database. Results are discussed with sensitivity, accuracy and specificity as parameters of evaluation.