Detection of Obstructive Sleep Apnoea by ECG signals using Deep Learning Architectures

Abstract

Obstructive Sleep Apnoea (OSA) is a breathing disorder that happens during sleep and general anaesthesia. This disorder can affect human life considerably. Early detection of OSA can protect human health from different diseases including cardiovascular diseases which may lead to sudden death. OSA is examined by physicians using Electrocardiography (ECG) signals, Electromyogram (EMG), Electroencephalogram (EEG), Electrooculography (EOG) and oxygen saturation. Previous studies of detecting OSA are focused on using feature engineering where a specific number of features from ECG signals are selected as an input to the machine learning model. In this study, we focus on detecting OSA from ECG signals where our proposed machine learning methods automatically extract the input as features from ECG signals. We proposed three architectures of deep learning approaches in this study: CNN, CNN with LSTM and CNN with GRU. These architectures utilized consecutive R interval and QRS complex amplitudes as inputs. Thirty-five recordings from PhysioNet Apnea-ECG database have been used to evaluate our models. Experimental results show that our architecture of CNN with LSTM performed best for OSA detection. The average classification accuracy, sensitivity and specificity achieved in this study are 89.11%, 89.91% and 87.78% respectively.