Online Epileptic Seizure Prediction Using Phase Synchronization and Two Time Characteristics: SOP and SPH

The successful prediction of epileptic seizures will significantly improve the living conditions of patients with refractory epilepsy. A proper warning impending seizure system should be resulted not only in high accuracy and low false-positive alarms but also in suitable prediction time.

In this research, the mean phase coherence index used as a reliable indicator for identifying the preictal period of the 14-patient Freiburg EEG dataset. In order to predict the seizures on-line, an adaptive Neuro-fuzzy model named ENFM (evolving neuro-fuzzy model) was used to classify the extracted features. The ENFM trained by a new class labeling method based on the temporal properties of a prediction characterized by two time intervals, seizure prediction horizon (SPH), and seizure occurrence period (SOP), which subsequently applied in the evaluation method. It is evident that an increase in the duration of the SPH can be more useful for the subject in preventing the irreparable consequences of the seizure, and provides adequate time to deal with the seizure. Also, a reduction in duration of the SOP can reduce the patient’s stress in the SOP interval. In this study, the optimal SOP and SPH obtained for each patient using Mamdani fuzzy inference system considering sensitivity, false-positive rate (FPR), and the two mentioned points, which generally ignored in most studies.

The results showed that last seizure, as well as 14-hour interictal period of each patient, were predicted on-line without false negative alarms: the average yielding of sensitivity by 100%, the average FPR by 0.13 per hour and the average prediction time by 30 minutes.

Based on the obtained results, such a data-labeling method for ENFM showed promising seizure prediction for online machine learning using epileptic seizure data. Apart from that, the proposed fuzzy system can consider as an evaluation method for comparing the results of studies.