A Novel Deep Learning Approach for Detection of Sleep Apnea from ECG Signal

Sleep apnea is a sleeping disorder caused by interruption or inadequate breathing while sleeping. Despite the fact that sleep apnea is very common in middle-aged and elderly people, the majority of cases remain un-diagnosed due to high medical costs. Irregularity in sleep can be observed and subjectively classified into different stages using electroencephalography (EEG), electrooculography (EOG), electromyography (EMG), and polysomnography (PSG). Sleep apnea can be effectively identified by analyzing the asymmetrical RR intervals from the ECG signal. This paper proposes a novel method to identify sleep apnea from ECG signal using deep neural network (DNN) based on LSTM and BiLSTM. BiLSTM outperforms over LSTM due to its potential for determining optimal weights based on both forward and backward learning. Effectiveness of the model is evaluated on the ECG signal available in Physionet database. For various hyper-parameters, classification accuracy is computed. This technique attained an overall accuracy of 100%, with rmsprop optimizer, 20 epoch, and learning rate of 0.01.