Human epileptic seizure prediction with fuzzy clustering of wavelet and polyspectra-based features of the EEG

Summary form only given. Epileptic seizures affect about 1% of the human population. Many patients (20%) are resistant to preventive drug treatment. The ability to anticipate the onset of seizures in such cases would permit clinical intervention. In previous work, the unsupervised optimal fuzzy clustering (UOFC) algorithm was applied to wavelet coefficient features in predicting hyperbaric-oxygen-induced generalized epileptic seizures in rat EEG. The two phases have the ability of processing and identifying transients (spikes), as well as long-term activity. In 16 instances, a pre-seizure state, lasting between 0.7 and 4 minutes, was defined. Projecting these results to human epilepsy is uncertain, given the different origin of the seizure as well as species differences in cerebral organization. In the present work, we attempt to predict spontaneous focal seizures in the EEG of epileptic patients who are monitored for surgical intervention. The UOFC is applied to two main classes of features: Fourier-transform based features (polyspectra) and multi-resolution based features (wavelets and multi-wavelets). Cluster formation shows the ability to detect 88%-100% of the 25 focal seizures and a prediction rate of in about 30% of the seizure, all using the wavelet coefficients as features. Less promising results were obtained using Fourier based features.