A CNN-based synchronization analysis for epileptic seizure prediction: Inter- and intraindividual generalization properties

We investigate the generalization capability of our proposed CNN-based approach to measure the strength of generalized synchronization in EEG recordings from epilepsy patients. With an in-sample optimization on short-lasting EEG data taken from two recording sites of a single patient we obtain a CNN with polynomial-type templates that allows us to approximate the strength of generalized synchronization in continuous long-lasting multichannel EEG recordings from this patient at a high accuracy. In an out-of-sample study we use the same CNN to analyze days of multichannel EEG data from other patients and observe that the strength of generalized synchronization between different brain regions in different patients can be approximated with a sufficient accuracy. These inter- and intraindividual generalization properties render CNN highly attractive for the development of miniaturized seizure prediction devices.