Short-Term Epileptic Seizures Prediction based on Cepstrum Analysis and Signal Morphology

This article aims to provide and implement apatient-specific seizure (for Intervention Time (IT) detection)prediction algorithm using non-invasive data to develop warningdevices to prevent further patient injury and reduce stress.Employing algorithms with high initial data volume and computationstime to increase the accuracy is an important problem inprediction issues. Consequently, reduction of calculations is metby applying only two effective EEG signal channels without manualremoval of artifacts by visual inspection as the algorithm’sinput. AR modeling and Cepstrum detect changes due to ITperiod. We carry out the goal of higher accuracy by increasingsensitivity to interictal epileptiform discharges or artifacts andreduce errors caused by them, taking advantage of the discretewavelet transform and the comparison of two channels epochsby applying the median filter. Averaging and positive envelopemethods are introduced to patient-specific thresholds becomemore differentiated as soon as possible and can be lead to soonerprediction. We examined this method on a mathematical modelof adult epilepsy as well as on 10 patients with EEG data.The results of our experiments confirm that performance of theproposed approach in accuracy and average false prediction rateis superior to other algorithms. Simulation results have beenshown the robustness of our proposed method to artifacts anderrors, which is a step towards the development of real-timealarm devices by non-invasive techniques.