Diagnosis of Epileptic Seizures in TLE Patients by Improved Cepstrum Analysis

Epilepsy is a common chronic brain disease characterized by recurrent seizures. The most effective way to detection the arrival of an epileptic seizure is EEG analysis. Various methods have been proposed in the detection of epilepsy. The article proposes the combination of the AutoRegressive and Cepstrum methods, taking into account the periodic characteristics of the amplitude and frequency of epileptic seizures compared to seizure-free periods in brain signals. Once the signal is received and pre-processed, the AR model of the signal is calculated and the Cepstrum is obtained on the modeled signal, then the Cepstrum coefficients are averaged within the given interval, and a detection of epileptic seizure is made by thresholding the averaged coefficients. TLE patients were selected in this study due to their higher prevalence compared to other types of epilepsy. The proposed detection algorithm was examined on 11 EEG patients with temporal lobe epilepsy. The results indicate that the proposed method has an accuracy of 98.3% and a false positive rate of 0.5%, which demonstrates its superiority over existing methods, especially considering the presence of various artifacts in the available data. Moreover, owing to its low computational complexity, the proposed method can be integrated into real-time implant detection systems.