Neural network based seizure detection system using raw EEG data

In this paper we present a new seizure detection system based on neural networks. The system takes raw EEG data without any explicit feature extraction. The removal of explicit feature extraction steps can improve flexibility and also mitigate hardware overhead and high clock rate often associated with feature extraction steps. We also propose a scheme to construct training data from raw EEG data which naturally has a significant length imbalance between seizure and non-seizure state waveform. The proposed detection system is validated with the CHB-MIT database. The results show that the proposed system can achieve detection performance compared to existing systems having explicit feature extractors.