A machine learning algorithm for the automatic detection of ictal activity using energy and synchronization features

This paper proposes a novel method for the early detection of ictal activity in epileptic patients. The detection system relies on features extracted from multi-channel EEG data collected from pediatric patients undergoing video-EEG monitoring. Two types of EEG features are extracted from each patient's EEG data; frequency-specific energy levels in the EEG data and the amount of neural synchrony displayed between the EEG channels. Prior to energy extraction, the EEG data undergoes a signal enhancement step designed to improve the differentiation between seizure and non-seizure EEG and enhance performance of the detector. Synchronization levels are calculated by evaluating the condition number (CN) of EEG data. The features are then fused to form a single feature vector for classification by two separate machine learning algorithm, support vector machine (SVM) and a feed-forward neural network. The proposed method achieves a sensitivity of 100%, detection latency of 2.74 seconds, and a false positive rate of 2.69 per hours for SVM based detection. Neural network based detection decreases the false positive rate by 47.9% from SVM based detection. The novality of this work lies in exploiting the benefits of combining both features prior to classification and training the detector using two machine learning algorithms. It is shown that the proposed system outperforms state-of-the-art detectors in terms of sensitivity and detection latency.