Seizure Prediction Using a Dynamic Model with Hidden Variable

Seizure is a kind of brain disorder and sometimes has vital effect on patients. Correct seizure prediction will provide huge help for patients and doctors. We have developed a novel approach to predict seizures. Wavelet transform is used to calculate the energy of a specific frequency band of 4-12 Hz to remove noise in the signal and to pick up useful information. A dynamic model is developed to describe this process with a hidden variable. We assume that the initial state of hidden variable has Gaussian distribution and it follows the second order autoregressive (AR) process. The hidden variable in the model can be solved by modified particle filters. Four patients´ intracranial EEG data are used to test our algorithm including 30 hours of ictal EEG with 15 seizures and 45 hours of normal EEG recordings. The first seizure from each patient is supposed to be known for parameter determination purpose. The results show that our algorithm can successfully predict 10 out of the 11 seizures and average prediction time is 33.5 minutes before seizure onset. The sensitivity is 90.9% and the false prediction rate is approximately 0.11FP/h.