Determination of epileptic eeg signals by model-based spectral analysis techniques

Electroencephalography (EEG) is a significant signal that represents the electrical activities of the brain. It provides useful information for the diagnosis of many diseases and the major one of these diseases is epilepsy. In this thesis, one of the parametric methods which has a wide variety of application in spectral analysis of signals and is called Autoregressive (AR) method is used. Since AR method characterizes especially narrow band spectrum successfully and it has great performance for estimation of spectrums with sharp peaks and also it has computationally efficient estimation methods, it is one of the most preferred methods for spectral estimation. EEG data of people with epilepsy and without epilepsy are analyzed with AR method. The power spectral densities of epileptic and non-epileptic EEG records are estimated with Yule-Walker and Burg methods. Feature extraction is necessary to detect the difference between epileptic and non-epileptic EEG. For this end, features of EEG data are extracted from the power spectral density estimations and in that respect the performances of Yule-walker and Burg methods are analyzed. In parametric methods, the accurate estimation of power spectral density is possible with the selection of suitable model order. For this reason, AIC (Akaike?s Information Criterion), MDL (Minimum Description Length), FPE (Akaike?s Final Prediction Error), CAT (Parzen?s Criterion of Autoregressive Transfer function), BIC (Bayesian Information Criterion) ve KIC (Kullback Information Criterion) methods are used to determine the AR model order. The performances of these methods are analyzed for Yule-Walker and Burg methods. According to the analysis results, it is understood that the performance of the model order determination and the feature extraction differ with EEG data. The performance of every method is evaluated for every EEG data set.