Classification of Obstructive Sleep Apnoea from single-lead ECG signals using convolutional neural and Long Short Term Memory networks

Abstract

Obstructive Sleep Apnoea (OSA) is a breathing disorder that happens during sleep. Polysomnography (PSG) is typically used as a reference standard for the diagnosis of OSA which uses different physiological signals such as Electrocardiography (ECG), Electroencephalogram (EEG) and Electromyogram (EMG) in a sleep laboratory. This procedure is time-consuming, expensive and inconvenient. However, detection of OSA by using a wearable sensor to collect Electrocardiography (ECG) signals is a practical and effective alternative. Previous studies of OSA classification from ECG signals focused on feature engineering methods which involves extracting specific features from ECG signals and using the extracted feature as inputs to the machine learning methods. In this study, we propose a novel method of OSA classification of ECG signal where deep learning methods automatically extract the features from the ECG signals and classify them. Our deep learning approach uses a hybrid model involving Convolution Neural Networks (CNN) and Long Short Term Memory (LSTM) networks. PhysioNet Apnea-ECG database is used for training and evaluation of our proposed deep learning model. For the released training dataset, our proposed model achieves the accuracy of 94.27%, sensitivity of 94.57%, specificity of 93.93% and F1 score of 95.41%. While for the testing dataset, the achieved accuracy, sensitivity, specificity and F1 score for the proposed model are 90.92%, 91.24%, 90.36% and 92.76% respectively. The performance of our model is compared with state of the art techniques and we found our model to achieve the best performance to classify OSA and health ECG signals.