Automated recognition of epilepsy from EEG signals

In this study, it is aimed to design an automatic pattern recognition system for the detection of epilepsy which distinguishes healthy and seizure electroencephalography (EEG) signals. During the study, 100 EEG signals from patients were used during the opened eyes and healthy epileptic seizures. Each EEG signal consisting of 4096 samples was divided into 256 samples and a total of 3200 signals were obtained. The designed pattern recognition system has been developed in 3 basic parts. In the first part, the power spectral density (PSD) estimation is performed with the periodogram and Welch methods and the frequency domain information of the EEG signals is obtained. In the second part, the feature vectors are found from the frequency domain information obtained in the periodogram and Welch PSD estimation. In the third part, healthy EEG signals from the eigenvectors obtained by using K-Nearest Neighbor Algorithm (K-NN) and Support Vector Machine (SVM) classifiers are distinguished from pathological EEG signals. 5-fold cross-validation method was used in evaluating the accuracy performance of the designed system. The total classification accuracy of the system was found to be 99.66% with K-NN, 99.72% with SVM for periodogram PSD estimation and 99.72% with K-NN, 99.75% with SVM for Welch PSD estimation. The results of the pattern recognition system designed in the study are promising because they are close to the work done with different approaches in the literature. The pattern recognition system designed here is not a diagnostic tool. It is foreseen that physicians may be useful in evaluating preliminary diagnosis.