Classification of ECG heartbeats using deep neural networks
Selma Harrane,Mohammed Belkhiri
DOI: https://doi.org/10.1007/s42600-021-00176-x
2021-09-25
Research on Biomedical Engineering
Abstract:PurposeThe electrocardiogram (ECG) classification is an important process in assisting doctors in detecting various types of cardiac arrhythmias. The extraction of features from the ECG is a key step for a good classification. Deep learning architecture such as convolutional neural networks (CNN) and long-short-term memory (LSTM) has recently gain popularity in real-world applications. The main reason for this popularity is that it can automatically extract features and classify them so that there is no need for handcrafted feature extraction and selection. The aim of this work is to demonstrate the success of CNN in the extraction of features from ECG signals and LSTM neural networks in ECG classification.MethodsThis paper presents an ECG heartbeat classification approach based on CNN and LSTM networks for the extraction of heartbeat features and the classification of six types of heartbeats, namely, normal beat, N; left bundle branch block beat, L; right bundle branch block beat, R; premature ventricular contraction, V; atrial premature contraction, A; and paced beat, P, provided from the well-known MIT-BIH arrhythmia database. Firstly, the ECG signals from the MIT-BIH database are segmented into heartbeats of about 0.8 s of duration, and then these raw heartbeats are introduced to the CNN in order to extract relevant features, and finally the feature vectors resulting from the CNN are fed to the LSTM network for classification.ResultsThe classification performance of the proposed model is validated on the MIT-BIH arrhythmia database. The training and test accuracies of the proposed method for MIT-BIH arrhythmia database are 99.9% and 98.60% respectively for the modified lead II. The classification accuracy for the N, L, R, P, V, and A classes was 99.40%, 98.00%, 97.30%, 99.70%, 98.40%, and 93.80%, respectively.ConclusionIn order to gain from the power of CNN in extracting features and LSTM networks in classification, we proposed an accurate and robust deep neural network system that combines CNN and LSTM to automatically classify six types of ECG heartbeats from lead II ECG records from the MIT-BIH database. The obtained results prove that combining CNN and LSTM for ECG feature extraction and classification achieves a higher performance compared to other techniques proposed from the literature. Another important advantage of the proposed method is that it does not require any preprocessing of data, so no filtering or denoising is applied to the raw ECG signals before classification. Therefore, our implemented model is less complex than other models proposed in the other works and achieved better classification accuracy.