Lagged Correlogram Patterns-based seizure detection algorithm using optimized HMM feature fusion

Epileptic seizure detection by EEG signal processing as an offline procedure is dependent to appropriate features. In this paper, a novel feature extraction and a feature fusion method for EEG signal classification are introduced. After preprocessing, the windowed EEG signals are decomposed to five rhythms by filter bank of Discrete Wavelet Transform (DWT) structure. Three correlogram spectrums for 3 lags are computed for these rhythms. The extracted features of these spectrums have been combined by Hidden Markov Model (HMM) as a fusion method. The lag values are optimized using a hybrid approach based on Multi-Layer Perceptron (MLP) neural network and Genetic Algorithm (GA) with implementing the Hill-Climbing (HC) search technique. The final feature vector is obtained with optimal lags and optimized HMM feature fusion. This scenario for feature extraction and fusion is called Lagged Correlogram Patterns (LCP) algorithm. Meanwhile, the spectral entropy as a frequency model of signal is estimated. The maximum value of averaged spectrum on all windows has been considered as a feature. Finally, the feature vectors are classified by Support Vector Machine (SVM) classifier with Radial Basis Function (RBF) kernel. The average of accuracy rate of 81.40% is obtained for the performance of seizure recognition.