A trend-detection algorithm for intraoperative EEG monitoring

Intraoperative EEG-monitoring needs to discriminate random fluctuations from real systematic variations (trends). This task is made more difficult by several types of artifacts. With the goal of supporting visual EEG evaluation, a new trend-detection algorithm is presented which is based on spectral analysis and a post-processing dynamic linear model, the latter introduced by Harrison and Stevens. A gradient value provided by this model is exploited to determine the onset and relative extent of an existing trend. Artifacts are detected by several threshold measures for the original signal and its first derivative. The system was validated using a set of intraoperative EEGs recorded during carotid endarterectomy.