Abstract:

Epileptic seizure with sudden and repeatability poses a great threat to patient safety. To effectively predict the epileptic seizure， an epileptic seizure prediction method based on multivariate Hilbert frequency domain model was proposed. Hilbert marginal spectrum， Hilbert weighted frequency and Hilbert marginal spectrum change direction were composed to a three dimensional feature vector as multivariate Hilbert frequency domain model， and then put it into support vector machine (SVM) to prediction epileptic seizure. The epileptic seizure prediction method was used to assess the prediction results. Experimental results showed that when the multivariate Hilbert frequency domain model was used to predict epileptic seizure for δ rhythm and θ rhythm， the seizure prediction horizon was 30~45minutes， so that patients could have enough time to take measures to deal with seizures. The seizure occurrence period was 5~10minutes， thus， the waiting time was shortened and the anxiety of patient was reduced. Compared with a variety of relevant methods， this method has lower false prediction rate and higher prediction sensitivity.

Key words: electroencephalogram, Hilbert-Huang transform, empirical mode decomposition, Hilbert marginal spectrum, Hilbert weighted frequency