Sleep Apnea Identification Based on EEG Signals Using Hybrid Spatio-Temporal Deep Learning

Sleep Apnea is a serious sleep disorder in which breathing stops for more than 10 seconds during sleep. It increases the risk of heart disease, high blood pressure, and stroke. This paper proposes methods to detect Sleep Apnea based on electroencephalogram (EEG) signals. EEG signals are recorded from multiple channels. Before detection, a filtered signal (0,5-30 Hz) needs to be extracted spacially first. It uses Convolutional Neural Networks (CNN). Then, the sequence of feature extraction signals was classified using Recurrent Neural Networks (RNN). So, it is called spatio-temporal. The experiment used the PhysioNet 2018 Challenge dataset every 30 seconds. Testing of the proposed method showed an accuracy of 98.96-99.59% or an average of 99,304%. The CNN extraction effectively captures spatial features from EEG signals across different channels, while the RNN is suitable for sequence data. Experiments also show that using GRU in a memory setting is better than LSTM in terms of accuracy and time performance, with only 40 minutes to achieve convergence. Extracting the appropriate frequency spectrum using the Wavelet transform effectively improves learning performance for Sleep Apnea detection. EEG signals capture the electrical characteristics of the brain and have a certain frequency that indicates Sleep Apnea.