Epileptic Seizure Prediction Using Spectral Entropy-Based Features of EEG

About 1 % of the world population suffer from Epilepsy. Epileptic seizures are generated by excessive and abnormal activation of neurons in the cortex. Unpredictable nature of these seizures motivated us to develop an algorithm to predict them. In our proposed algorithm, we used the data from MIT Physionet database which contains EEG signals of 24 epileptic patients. We employed spectral entropy to predict epileptic seizures. With the calculation of the power spectral density and adopting its frequency components as probability density functions which are used in the calculation of Shannon entropy, we managed to extract our desired features. In the next step, 2 classifiers which are support vector machine (SVM) and K-nearest neighbor (KNN) classifier were used as predictors of epileptic seizures. Our proposed algorithm can predict occurrence of a seizure using the first 9 minutes of a 10-minute interval before the seizure. The proposed method using SVM achieved sensitivity of 83.8% and specificity of 71%. KNN classifier achieved sensitivity of 83.8% and specificity of 67.8%. The proposed algorithm not only had an acceptable accuracy but also was one of the best algorithms compared to the other researches in the literature in terms of computational complexity, required energy for the calculations, and time delay. The achieved delay of 0.9 seconds is, as far as we know, the shortest time delay among all algorithms.