Linear and non-linear methods for automatic seizure detection in scalp electro-encephalogram recordings

Area of Science:

• Neuroscience
• Signal Processing
• Computational Biology

Background:

• Electro-encephalogram (EEG) signals measure dynamic brain electrical activity.
• Assessing non-linear dynamics in EEG is crucial for understanding brain states.
• Existing linear methods may miss subtle changes in neural activity.

Purpose of the Study:

• Introduce and evaluate a novel non-linear technique, multi-dimensional probability evolution (MDPE).
• Compare MDPE's ability to detect dynamic changes in EEG signals against linear statistics.
• Determine if MDPE offers advantages over variance-based methods for seizure detection.

Main Methods:

• Developed MDPE based on the time evolution of probability density functions in a multi-dimensional state space.
• Utilized synthetic EEG data to demonstrate MDPE's sensitivity to underlying dynamic changes.
• Applied both MDPE and variance statistics to ten clinical scalp EEG recordings for seizure detection.

Main Results:

• Both MDPE and variance successfully identified seizures in all ten investigated EEG recordings.
• MDPE demonstrated a lower rate of false positive detections compared to variance.
• No conclusive evidence emerged to suggest MDPE or other non-linear methods significantly outperform variance for seizure identification.

Conclusions:

• MDPE is a viable non-linear method for analyzing EEG signals and detecting dynamic changes.
• While promising in reducing false positives, MDPE's superiority over linear variance methods for seizure detection requires further investigation.
• The study highlights the need for rigorous comparison of non-linear and linear techniques in neurophysiological signal analysis.