Discrimination of Interictal and Preictal States Using Wavelet Analysis on EEG Signals of Patients with Intractable Convulsions

Abstract

The discrimination of the interictal and preictal states contributes to the construction of an efficient system of seizure prediction. Here, we performed the classification of the interictal and preictal states for EEG signals of the scalp. The energies of the levels obtained by the signal decomposition of the Wavelet Discrete Transform were used as features for classification. The kNN and SVM classifiers were used in the analysis of the individual EEG channels, which gave indications that the occipital lobe region channels are the most relevant to differentiate between the interictal and preictal states. Using these channels, the classification into two states achieved accuracy of 97.29%, sensitivity of 96.25% and specificity of 98.33%. In addition, the different frequency ranges obtained by Wavelet for the classification were analyzed, and it was observed that the range of 32 Hz to 128 Hz presented greater relevance in the task.