Improved epileptic seizure detection combining dynamic feature normalization with EEG novelty detection.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Critical Care Medicine

Background:

• Continuous electroencephalogram (EEG) monitoring is crucial for critically ill patients.
• Seizure detection algorithms, like support vector machines (SVM), are vital but require accurate EEG feature normalization.
• Existing normalization methods, including Median Decaying Memory (MDM), can be negatively impacted by including non-seizure or seizure-like activity in their baseline calculations.

Purpose of the Study:

• To introduce and evaluate a novel EEG normalization method, Novelty-Median Decaying Memory (Novelty-MDM), for seizure detection.
• To compare the performance of Novelty-MDM against fixed baseline (FB) and MDM normalization methods.
• To enhance the accuracy of SVM-based seizure detection in long-term intensive care unit (ICU) EEG recordings.

Main Methods:

• Developed Novelty-MDM, which uses a novelty detection algorithm to automatically select appropriate EEG epochs for the baseline buffer.
• Evaluated three normalization methods (FB, MDM, Novelty-MDM) within an SVM-based seizure detection framework.
• Assessed performance using the area under the sensitivity-specificity ROC curve across 17 long-term ICU EEG recordings.

Main Results:

• The MDM approach did not show significant improvement over the fixed baseline (FB) method (p < 0.27).
• Novelty-MDM demonstrated a significant performance improvement compared to both FB (p = 0.015) and MDM (p = 0.0065).
• The inclusion of seizure-like episodes in the baseline buffer negatively affected the performance of the MDM method.

Conclusions:

• Novelty-MDM is a superior EEG normalization technique for seizure detection compared to FB and MDM.
• Automated selection of baseline EEG data using novelty detection enhances the reliability of seizure detection algorithms.
• This improved normalization method holds promise for more accurate continuous EEG monitoring in critical care settings.