A deep Learning Scheme for Automatic Seizure Detection from Long-Term Scalp EEG

Epilepsy is a chronic brain disorder that is expressed by seizures. Monitoring brain activity via electroencephalogram (EEG) is an established method for epilepsy diagnosis and for monitoring epilepsy patients. Yet, it is not favorable to visually inspect EEG signals to diagnose epilepsy, especially in the case of long-term recordings. This process is time consuming and tedious error-prone exercise. In recent years, the sub-field of machine learning called deep learning has achieved remarkable success in various artificial intelligence research areas. In this paper, we present a method based on the deep convolutional neural networks (CNNs) to perform unsupervised feature learning framework for automated seizure onset detection. The proposed system was evaluated on 526 hours duration of scalp EEG data, including 181 seizures of 23 pediatric patients. The different parameters of CNNs were optimized through 4-fold nested cross-validation. The resulting generalized CNN seizure detection model achieved an average sensitivity of 86.29%, an average false detection rate of 0.74 h-1 and an average detection latency of 2.1 sec.