Abstracts

Rationale: Epilepsy is characterized by recurrent spontaneous seizures that occur without any forewarning in most cases. Recent studies on prediction or early seizure detection would facilitate possible clinical application of timely therapeutic intervention for acute seizure control. The purpose of this study was to improve the accuracy and speed of seizure detection by optimizing the detection algorithm.Methods: Twenty two male Sprague-Dawley rats were injected with pilocarpine (350-380 mg/kg) to induce initial status epilepticus and later spontaneous recurrent seizures after latent periods. Continuous in-vivo EEG monitoring (Twin Grass-Telefactor, USA) was performed in all rats with four epidural and depth electrodes (C3, C4, left and right hippocampi) to detect ictal EEGs. All EEG data were first analyzed visually, and then analyzed using a seizure detection algorithm based on support vector machine after stepwise applications of fast Fourier transform and derivative filtering process for denoising. Results from automated seizure detection were analyzed for different threshold and duration constraints and compared with those from visual analysis to estimate false positive, false negative, and timing of seizure detection.Results: Total 217 spontaneous recurrent seizures were analyzed so far. By adding derivative filtering from transformed EEG data, the accuracy was improved dramatically decreased 46% of false positive via denoising baseline EEG, and reduced 20% of false negative via denoising seizure activities. The average timing of seizure detection was 2.6 sec after seizure onset time determined by visual analysis.Conclusions: Using adjusted support vector machine algorithm, the accuracy was improved especially with the help of additional filtering for denoising. Further studies would need to optimize the detection algorithm, especially to improve the detection speed.