Seizure detection in EEG using dynamic system analysis

An epileptic seizure is the irregularity of brain activity that interferes with normal functions. Electroencephalog-raphy (EEG) is widely used for detecting and localizing seizures. Almost all automatic detection methods pursue machine learning algorithms taking various features of EEG recordings as input. Previous studies focused on features in the time domain (e.g. variance), frequency domain (e.g. power spectrum), and time-frequency domains (e.g. wavelets, empirical modes). As an alternative, we present a dynamical system-based detection. From the perspective of a dynamic system, EEG is the output of a nonlinear system consisting of a network of neurons. The state of the system can be represented by a point in the phase space. Time evolution of the state creates a unique portrait in the phase space that captures the underlying dynamics of the system. In this paper, we have shown that entropy, as an important characterization of the dynamic system, of the EEG signals can be used to discriminate between the normal and seizure states. Our method differs from other methods in that it provides a biologically interpretable perspective to study the brain in healthy and diseased conditions.