An Efficient Method to Detect Sleep Hypopnea- Apnea Events Based on EEG Signals

Hypopnea refers to the state in which insufficient alveolar ventilation during night sleep decreases the respiratory airflow by more than 50% of the airflow. However, sleep apnea is a more serious respiratory event, such as complete cessation of respiratory airflow for 10 seconds. The occurrence of hypopnea is a precursor to the occurrence of apnea events and the two are closely connected. In this paper, we propose a method based on the combination of discrete wavelet transform and approximate entropy of EEG signals to detect sleep apnea and hypopnea events. For this purpose, first, data preprocessing is performed on the EEG record data set obtained from Tianjin Chest Hospital, and then infinite impulse response (IIR) Butterworth bandpass filter is used to decompose the data into delta, theta, alpha, beta and gamma. Second, descriptive features are extracted based on sub-bands discrete wavelet transform such as the approximate entropy of high-frequency coefficients. Third, the features are filtered based on Support Vector Machine (SVM) recursive elimination. Finally, several machine learning algorithms including SVM, K-Nearest Neighbor (KNN) and Random forest (RF) are employed to identify the occurrence of sleep hypopnea-apnea events. The highest accuracy rate reached 94.33%, the sensitivity reached 93.10%, and the specificity reached 95.07%. The obtained results validate that the proposed method is an effective and practical diagnostic method to detect the occurrence of hypopnea-apnea events.