Epilepsi hastalığı kişilerin normal ve sosyal hayatını olumsuz etkileyen bazı durumlarda ölümle sonuçlanabilen ciddi bir hastalıktır. EEG (elektroensefalogram) uzmanı tarafından EEG’nin incelenerek hastanın durumu hakkında bilgi vermesi hala klinikteki altın standart olarak kabul edilmektedir. EEG verilerini bu yöntemle analiz etmek çok uzun veri kaydının incelenmesini gerektirmektedir. Ayrıca, aynı EEG verisini, farklı EEG uzmanları farklı olarak yorumlayabilmektedir. Epilepsi hastalığının EEG verilerinden tespitinde otomatik bir yöntemin gerekliliği açıktır. Bu çalışmada yüksek dereceli istatistikler kullanan ikiz-izge ve ikiz-tutarlılık yöntemleri epileptik atak içeren subdural elektroensefalogram (SEEG) verilerine uygulanarak, atak zamanları tespit edilmiştir. Epilepsi atağının doğrusal olmayan yapısı göz önünde bulundurularak bir algoritma geliştirilmiştir. SEEG verileri durağan kabul edilen zaman parçalarına bölünerek ikiz-izge yöntemi uygulanmıştır. İkizizgenin normalize edilmiş hali ikiz-tutarlılık hesaplanmıştır. İkiz-tutarlılık matrislerinden doğrusal olmama ölçüsünü gösterecek değişkenler belirlenmiştir. Bu değişkenler kullanılarak epileptik atak tespit edilmiştir. Epileptik atak tespitinde değişkenlerin performansı ROC (Receiver Operating Characteristic curve) kullanılarak değerlendirilmiştir. Faz kuplajı olan ikiz-frekansların sayısını gösteren değişken ile 0.1853 bağıl toplam hata oranında %53.49 duyarlılık ve %90.25 özgüllük elde edilmiştir. Diğer bir doğrusal olmayan yöntem Lempel Ziv karmaşıklık ölçüsü ve doğrusal bir yöntem olan güç izge analizi aynı işaretlere uygulanmıştır. İkiz-izge temeline dayalı sunulan bu algoritma ile ataklar tespit edilmesine rağmen, Lempel Ziv karmaşıklık ölçüsü ve güç izge analizi atakları tespit etmede başarılı olamamıştır.

Epilepsy is a disease that has negative effects on human’s normal and social lives. On some cases epilepsy can even be lethal. Expert interpretation on EEG (electroencephalogram) data is still the ‘golden standard’ in clinical applications where experts analyze the EEG data and generate the reports for the patient’s situations. To generate the reports experts are obliged to examine and analyze very long EEG data in which the interpretation of the data may change from one expert to another. It is obvious that an automatic method is required to detect the epilepsy from EEG data. The starting time or early detection of the seizure is very important in terms of diagnosis, treating and controlling of the epilepsy. It is known that the noise, which is generated by recording devices and environment, has negative effects on analysis and interpretation of EEG while observing and analyzing the scalp EEG. For that reason, the works in literature are concentrated on subdural EEG (SEEG) in recent years. Subdural EEG is collected from patients, which do not answer to the medical treatment and candidate for brain surgery. The epileptic area in brain is decided by using MRI results, patient medical history, and clinical findings before medical operation. Patients who are decided that have epileptic area in their brain are taken to the medical operation. SEEG is recorded during the operation by using electrodes, which are placed on the celebral cortex. The SEEG data is loaded to the computers or CDs as digital data after required amplification and filtering. The aim of this work is to apply bispectrum and bicoherence methods of higher order statistics to epileptic seizure SEEG data and to detect the changes that occur during, before and just after the seizure. To achieve these objectives an algorithm is developed. In this algorithm the epileptic seizure is assumed as stationary and nonlinear. The first step is to apply bispectrum method to time framed EEG data segments, which is assumed to be stationary. Bicoherence, which is also called normalized bispectrum, is calculated from bispectrum. From the bicoherence matrix the parameters that are the measures for non-linearity, is determined. The values of bicoherence matrix are between 0 and 1 in theory. Whereas the values, which are showing uncoupled frequencies, might be close to zero but not every time in practice. A threshold level is used to determine true-coupled bifrequencies. The parameters that are the measures for non-linearity, is determined from the bicoherence matrix. Employing the non-linearity measures from each time frame, epileptic seizure is detected. The performance of the parameters on epileptic seizure detection is determined by ROC (Receiver Operating Characteristic curve). According to this method the parameter that shows the number of bifrequency, which is above a threshold level in bicoherence matrix, is determined as the most suitable parameter on epileptic seizure detection. Using this parameter epileptic seizure is detected. For this data package on a total error rate of 0.1853 sensitivity of %53.49 and specificity of %90.25 is achieved. Power spectrum is applied to the same signals to compare the performances of methods. Welch Method is used for estimation power spectrum. Normalized power spectrum and power in different frequency bands are used to detect the epileptic seizures. Power spectrum is unable to detect the epileptic seizures with acceptable total error rate for this data package. Lempel and Ziv Complexity is another nonlinear method. Lempel and Ziv Complexity and its variants are popular metrics for characterizing biological signals. Lempel and Ziv Complexity is also applied to the same signals to compare the performances of methods. Although Lempel and Ziv Complexity is applied to SEEG data to detect the seizures in literature it is unable to detect the epileptic seizures with acceptable total error rate for this data package. Therefore, parameters calculated from bicoherence provide a good characterization of epileptic data and detection of epileptic seizures. There is not any nonlinear method yet which is used for the clinical applications to detect epileptic seizure with high sensitivity and specificity. After the discovery of the possibility to control and treat the epileptic seizure by electrical stimulating the epileptic focus region of the brain, research to find methods for detecting the epileptic seizures are accelerated in a few years. Because the results of this study enlighten the literature, it will give an important contribution to a scientific hot topic.