Amplitude-integrated EEG processing and its performance for automatic seizure detection

Amplitude-integrated EEG (aEEG) is an easy-to-use screening tool for monitoring electrical seizures which is widely used in hospitals nowadays. The purpose of this paper is to develop the process for creating aEEG in details and to present an algorithm for a single-channel automatic seizure detection. To create an aEEG signal, a raw EEG signal was passed through a band-pass filter, peak-to-peak rectification, smoothing, semi-logarithmic compression and time compression. The processed signal was qualitatively evaluated by widely used commercial software. The algorithm was developed based on statistic methods and tested with CHB-MIT Scalp EEG Database in order to detect seizures which last longer than 50 seconds (total 85 seizures with annotations in 683 EDF files) from a single selected channel of each data record which has maximum amplitude in ictal period. The performance measurement showed average sensitivity of 88.50% (33.33-100.00%), false detection rate of 0.18 (0.00-0.58) false positives per hour and the percentage of false detection duration of 0.34% (0.00-0.93%). The results showed that aEEG with automatic seizure detection developed by our process was an effective screening tool to monitor seizures.