A study of interdependence in electroencephalographic recordings of epileptic seizures: comparison of state-space and phase-based measures

Abstract

Treball fi de màster de: Master in Computational Biomedical Engineering Tutor: Prof. Dr. Ralph G. Andrzejak

Epilepsy is a neurological disorder characterized by synchronous neuronal activity in the brain, which causes seizures. Studying epileptic brain activity allows researchers to identify the seizure onset zone, which is crucial for epilepsy treatment. Various methods, such as nonlinear interdependence measures, assess the interaction between brain areas and can be used to detect seizure onset zones in electroencephalographic (EEG) recordings. A state-space-based measure L is able to detect the direction and strength of interaction between two dynamics. A phase-based measure R quantifies the degree of coherence between two dynamics using mean phase coherence. Although both measures are used to assess interdependence, their agreement in realworld EEG data remains to be tested. In this work, we compare the two measures applied to pairs of signals obtained from the EEG of 100 epileptic seizures to evaluate their reliability in clinical EEG data. We first assess the measures under controlled conditions using Rössler dynamics. We study the measures applied to noise-free signals obtained from coupled systems. Then, we add noise and study the behavior of the measures with the varying level of noise. Subsequently, the measures are applied to signals obtained from EEG data of patients with focal-onset epilepsy. The results of both approaches are compared against each other, identifying the overall consistency and potential deviations. The study demonstrates while the two measures generally produce consistent results, their disagreement is observed during seizures. This indicates they recognize different features derived from the complex dynamics of the brain. Although the exact cause of this divergence remains unclear, it highlights the need for careful interpretation when analyzing seizure-related data. These findings underscore the importance of using complementary approaches in EEG analysis and contribute to the understanding of different bivariate approaches in neuroscience.