Sleep Apnea Diagnosis Using Complexity Features of EEG Signals

Sleep apnea syndrome is one the most prevalent sleep disorders. The accurate diagnosis and treatment of apnea by physicians can help to avoid its destructive effects in the long term. Electroencephalogram (EEG) records activity of the brain from different areas of scalp and can be an appropriate method to diagnose sleep apnea. In this work, we proposed a Computer Aided Diagnosis System (CADS) for sleep apnea based on complexity features of EEG. At first, EEG time series of 20 participants were decomposed into six frequency bands (delta, theta, alpha, sigma, beta, and gamma) by using bandpass Finite Impulse Response (FIR) filters. Then, complexity features such as fractals, Lempel-Ziv Complexity (LZC), entropies, and generalized Hurst exponent that was used for the first time to detect sleep apnea from EEG signals, were extracted from each frequency band. The minimumredundancy maximum-relevance (mRMR) algorithm was applied to sort 120 features of three EEG channels. Finally, two popular classifiers, Support Vector Machine (SVM) and K-Nearest Neighbors (KNN), were used to detect sleep apnea. 99.33% accuracy was obtained using the SVM classifier and generalized hurst exponent had an effective contribution to detect apnea.