Seizure prediction using low frequency EEG wavesfrom WAG/Rij rats

摘要

Frontal lobe epilepsy is the second most common form of epilepsy which initiates during sleep and causes death. Early detection is the solitary measure to control seizure. Electroencephalography (EEG)is the only confirmatory test for seizure. However, epileptic research still depends on studies based on animal models. In this research, our main objective is to study the significance of low frequency brainwaves (which is dominant during sleep) in the prognosis of seizure from WAG/Rij rat data. Wavelet based decomposition coefficients and power spectral density (PSD) are selected as features to take care ofnon-stationary nature of brain waves. A comparative study of classifiers' performance is formulated between low frequency wave range (0.5-13 Hz) and the total frequency range (0.5-40 Hz) where RBF-SVM provides the maximum classification accuracy. The average classification accuracy of RBF-SVM for low frequency wave range is found to be 92.50%, which lies within a range of <;2%, compared to the total frequency wave range as 94.03%. QDA has the second highest classification accuracy. Both RBF-SVM and QDA perform better for total frequency wave range compared to low frequency wave range. However, LSVM and LDA produce a different pattern - higher classification accuracies for low frequency wave range EEG data. The novelty of this paper lies to the fact that selection of low frequency brain wave is more significant in the prognosis of frontal lobe epilepsy that is not only useful detecting spikes during sleep but also removes environmental and physiological artifacts with higher frequencies.