Abstracts

Anticipation of epileptic seizures is, among others, the most challenging aspect in epileptology. Recent findings indicate that particularly measures quantifying relations between recording sites to characterize interaction between different brain regions, show a promising performance that exceeds chance level if tested by statistical validation. Despite the conceptual simplicity of a number of these bivariate measures, real-time applications are currently limited by calculations for large number of combinations of electrodes. Promising systems for measuring synchronization while minimizing space and energy are cellular neural networks (CNN) as they offer a massive computing power, are capable of universal computation, and are already available as analog integrated circuits., We studied multi-channel intracranial EEG recordings from two epilepsy patients (duration 5 and 3.5 days; 10 and 5 epileptic seizures of focal origin) along with the symmetric and asymmetric (for detection of driver-responder relationship) profiles of the nonlinear interdependency measure [italic][/italic]. In order to find optimum network settings we first performed an in-sample supervised training using randomly selected EEG segments amounting to about 5 minutes from patient #1. For an out-of-sample validation and in order to study the long-term behavior of our CNN-based approximation for generalized synchronization, network settings were then tested on EEG recordings from both patients., Our CNN reproduced the temporal variability of [italic][/italic] with a sufficient quality, the average deviation amounting to 6% over 5 days only for patient #1. For patient #2 we achieved a comparable good approximation (9% deviation over 3.5 days) without re-adjusting the network. Since the nonlinear interdependency measures allow to detect driver-responder relationships, we exchanged the EEG data (patient #1) from the two recording sites. When comparing the obtained [italic][/italic] profile from the CNN with the corresponding calculated profile we observed a clear-cut deviation along with an increase of the difference measure which might indicate that our CNN is also able to approximate the asymmetrical part of the generalized synchronization measure., CNN allow to characterize aspects of generalized synchronization in continuous long-lasting EEG recordings from epilepsy patients. Without using a priori knowledge as to the various states of the patients for training our CNN we were able to reproduce the temporal variability of generalized synchronization in brain dynamics with a sufficient quality, which might be helpful for improving predictability of epileptic seizures., (Supported by Deutsche Forschungsgemeinschaft.)