A Novel Approach to K-complex Detection in Human Sleep EEG Using Entropies

K-complexes, which are transient, non-stationary and nonlinear waveforms with comparatively higher peaks, play a crucial role in sleep stage detection and specifically in sleep stage 2. K-Complex detection in electroencephalogram (EEG) signals is a challenging task due to its nonlinear and dynamic characterization. Extracting K-complexes visually is difficult, time-consuming and requires highly qualified experts. In this study, an effective method based on EEG entropies is proposed to detect K-complexes. EEG signals are first pre-processed and different entropy measures including sample entropy, approximation entropy, Shannon’s entropy and Tsallis entropy are extracted and then forwarded to feature selection stage. Most significant features are selected using t-test analysis and are finally fed to the classification model. In the classification phase, based on Adaboost algorithm, a fusion of five classifiers including linear discriminant analysis (LDA), Naïve Bayesian classifier (NB), multi-layer perceptron (MLP), K-nearest neighbor (KNN) and support vector machine (SVM) is employed to enhance classification performance. We have evaluated our proposed method by employing two different datasets and compared our results with previous studies. Results show that the proposed method is considerably effective in comparison with existing methods. The average classification accuracy is 95.6% and the method overall computation time is acceptable. Significant EEG channels and brain lobes are also reported. Our proposed method can be an efficient tool for physicians and neurologists to automatically score sleep stages and also in diagnosis and treatment of sleep disorders.

Keywords:

Sleep Stage Detection , K-complex Detection , Electroencephalogram , Nonlinear EEG Analysis , EEG Entropy , Adaboost.