Automatic Detection of Sleep Apnea Using Sub-Band Features from EEG Signals

Sleep Apnea is a major sleep disorder which leads to a partial or complete stopping in breathing for a short duration of time during sleep. The out-turn of this is extreme daytime drowsiness. Among different polysomnographic signals, Electroencephalogram (EEG) signal reflects the electrical activity of the brain by sensing and recording the brain's activities. Thus, EEG serves as a valuable information source for detecting the sleep apnea events. In this research, an efficient automatic method is proposed for differentiating apnea and non-apnea events of an apnea patient which is a very burdensome task if performed manually. The features energy, entropy, mean absolute deviation and kurtosis, extracted from five sub-bands, provide better accuracy, sensitivity and specificity as compared to other recently published studies. The performance evaluation of the proposed technique is carried out using the publicly available Physionet dataset. The highest classification accuracy of 95.10%, sensitivity of 93.20% and specificity of 96.80% is achieved by using Ensemble decision tree methods using the bagging technique. The novel method proposed in this research offers superior results in terms of accuracy, sensitivity and specificity as compared to the recently published work on the same database.