A Study of Seizure Prediction Based on EEG Phase Synchronization

Around 0.6-0.8% of the world’s population is affected by a re-current neurological disorder of epilepsy. Second only to stroke, epileptic seizures interrupt lives of 50 million people. The sudden and seemingly unpredictable nature of seizures is one of the most defeating facts of epilepsy. The ability to predict a rising seizure reliably can mitigate the severity of the disorder and open up new theraupetic possibilities: closed-loop seizure prevention systems and EEG triggered ondemand therapy. No study has yet reported a method that can reliably predict a seizure. This thesis investigates EEG phase synchronization as a promising algorithm for seizure prediction. The algorithm is evaluated on a database of three patients with a total of 30 seizures and 230 hours of labeled EEG data. The results of the study show 37% sensitivity for an average seizure occurrence period (SOP ) of 30 minutes and maximum false prediction rate (FPRmax) of 0.15 seizures/hr. A review of seizure prediction research and methodology is presented. Motivated by potential implementation on an implantable chip, a realizable FPGA architecture of the algorithm is proposed. Acknowledgements I would like to thank my advisor Professor Roman Genov for pointing me to the prominent researchers in the field of seizure prediction, for his encouragement and enthusiasm communicated in meetings and stimulating discussions. I appreciate the kindness of Professor Paul Chow in providing a Xilinx board for the FPGA component of this thesis. I would like to thank a group of researchers in University of Freiburg, Germany, for providing the EEG dataset used in this work and establishing an international workshop to advance the study of seizure prediction. Finally, I thank my parents for their ever-present guidance and support.