Prediction of Abnormal EEG Epileptic Activities Based on Sub-band Decomposition Techniques: System Architecture and Signal Processing

This paper presents an efficient approach that works on various bands of brain waves for channel selection and epileptic seizure prediction using digital Infinite Impulse Response (IIR) filters. The proposed approach begins with a filtering process to separate the bands of brain wave signals (gamma, beta, alpha, theta, and delta) using various kinds of IIR filters including Butterworth (BW), Chebyshev Type I (CH I), and Chebyshev Type II (CH II). Each filter is used for delimitating a frequency band of the EEG signals based on a band of brain waves and removing the fundamental noise component and its harmonics. After that, the bands are transformed into frequency domain using Fractional Fourier Transform (FrFT) with different orders. Moreover, both magnitude and phase of the Fast Fourier Transform (FFT) are used. The seizure prediction method depends on a statistical approach that has training and testing phases. Simulation results have been obtained on the CHB-MIT dataset, which is divided to 70% for training and 30% for testing. These results proved the feasibility of FFT and FrFT for seizure prediction. A prediction rate of 100% has been achieved for FrFT with all filters and all selected orders. The best result of Patient 1 (P1) is for CH I filter for absolute FFT over δ wave. The best result of Patient 8 (P8) is for the absolute FFT for the γ wave using BW filter. Finally, the best result of Patient 20 (P20) is for FrFT with σ = 2 for the δ wave with CH II filter.