Optimizing dynamical similarity index extraction window for seizure detection.

This paper addresses an optimization problem in choosing optimum window length for feature extraction in automatic seizure detection. The processing window length plays an important role in reducing the false positive and false negative rates and decreasing required processing time for seizure detection. This study presents an approach for selecting the optimum window length toward the extraction of dynamical similarity index (DSI) feature. Then, the optimal window value in DSI extraction was used to detect seizure onset automatically. The algorithm was applied to electroencephalogram (EEG) signals from European Epilepsy Database. Although the main purpose of this study was not the seizure detection and mainly focuses on proposing an approach for finding an optimum window length for feature extraction towards the early seizure detection, the results showed that the proposed method achieves 83.99% of sensitivity in seizure detection. The low false positive rate per hour (FPR/h) was also significant due to continuous EEG analysis. The method showed fast computation speed which promises a potential for the real time applications. The proposed method for the window optimization in feature extraction of DSI can be implemented for other features to further improve the performance of seizure detection.