ECG signal detection and classification using machine learning and artificial intelligence

Abstract

The accurate recognition and classification of different types of arrhythmias are crucial for the correct treatment of cardiac patients. consequently, the automatic arrhythmia detection from an electrocardiogram (ECG) has been a very important subject. 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) have 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. In this work, in 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 ECG heartbeats. As a first step, the ECG signals from the MIT-BIH database are segmented into heartbeats of about 0.8 s of duration. In the second step, 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. We compared the effectiveness of LSTM network training in the variant with a raw ECG signal at the input, in the variant with two input spectral features (IF and SE) and in the variant with CNN learned features. Keywords: ECG signals; Deep learning; CNN; LSTM network; Feature extraction; Classification; CNN learned features.