TIME-FREQUENCY ANALYSIS IN ICTAL AND INTERICTAL SEIZURE EPILEPSY PATIENTS USING ELECTROENCEPHALOGRAM

摘要

Conventional method to distinguish normal and seizure EEG by an epileptologist?s visual screening is tedious and operator dependent. Normal DWT-based seizure detection technique established before suffers from deteriorating of performance due to increasing number of non-relevant features by wavelet decomposition. PCA approach has been utilized in this paper to overcome this problem. Energy, amplitude dispersion and approximate entropy (ApEn) of each sub-band were used as feature of interest and fed to Support Vector Machine (SVM) classifier. Differences between ictal, interictal and normal EEG based on these features were explored. There are significant differences in delta, theta and alpha band in sub-band energy, whereas ApEn changes are found in beta and alpha for ictal EEG. Amplitude dispersion illustrates changes in all sub-bands. PCA approach has been proven to have better accuracy (98%) compared to non-PCA approach (97%) in detecting ictal seizure. The proposed method produced the highest accuracy (98%) compared to other existing methods. The algorithm shows potential to be used clinically.