Nonlinear similarity analysis for epileptic seizures prediction

The prediction of epileptic seizures can promise a new diagnostic application and a novel approach for seizure control. This paper proposes an improved dynamical similarity measure to predict epileptic seizures in electroencephalographic (EEG). First, mutual information and Cao's method are employed to reconstruct a phase space of preprocessed EEG recordings by using the positive zero crossing method. Second, a Gaussian function replaces the Heavyside function within correlation integral at calculating a similarity index. The crisp boundary of the Heavyside function is eliminated because of the Gaussian function's smooth boundary. Third, an adaptive detection method based on the similarity index is proposed to predict the epileptic seizures. In light of test results of EEG recordings of rats, it is found that the new dynamical similarity index is insensitive to the selection of the radius value of Gaussian function and the size of segmented EEG recordings. Comparing with the dynamical similarity index proposed by Le Van Quyen et al. [Anticipating epileptic seizures in real time by a non-linear analysis of similarity between EEG recordings, NeuroReport 10 (1999) 2149–2155], the tests of twelve rats show the new dynamical similarity index is better to predict the epileptic seizures.