Automated EEG signal classification using chaotic local binary pattern

Abstract

Background: Electroencephalography (EEG) signals are the electrical signals which depicts the brain's neuronal activities. The EEG signals inherently have nondeterministic patterns. Hence, we have used a chaotic feature generation function to classify normal and abnormal EEG signals in this work. Method: This research presents a new generation abnormal EEG detection model using a chaotic one-dimensional local binary pattern (CLBP) and wavelet packet decomposition (WPD) techniques. The Temple University Hospital (TUH) EEG dataset is used to develop and evaluate our chaotic feature generation model. In this work, the WPD is performed on EEG signals, and then CLBP is applied on the decomposed signals to extract the features. The iterative minimum redundancy maximum relevancy (ImRMR) is applied to select the clinically significant features. Finally, these features are classified into normal and abnormal EEG classes using a support vector machine (SVM) classifier. Results: Our developed model yielded the detection accuracies of 93.84% to 98.19% for 24 channels using SVM classifier with ten-fold cross-validation strategy. Conclusion: We have obtained the highest classification performance of 98.19% for the PZ channel that is the highest performance so far using this database. Our developed model is ready to be tested with more EEG data.