An approach for automatic sleep apnea detection based on entropy of multi-band EEG signal

Sleep apnea is a very common sleep disorder affecting a large number of people all over the world. Electroencephalography (EEG) signal analysis is an important process that enables neurologists and sleep specialists to diagnose and monitor sleep apnea events. In view of exploiting the variation in random characteristics of multi-band EEG data between apnea and non-apnea events, in this paper, an entropy based feature extraction scheme is proposed. It is shown that the proposed feature set, extracted from five different band-limited EEG signals, offers satisfactory feature quality in terms of standard performance criteria, such as geometric separability index. For the purpose of classification, the K-nearest neighborhood (KNN) classifier is used. The proposed method is tested on several subjects taken from publicly available Physionet database. It is found that the proposed method offers superior classification performance with lower feature dimension in comparison to that obtained by existing methods, in terms of sensitivity, specificity and accuracy.