Seizure detection from multi-channel EEG using entropy-based dynamic graph embedding.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Epileptic seizures are characterized by abnormal, sudden neuronal discharge in the brain, leading to temporary dysfunction.
• Accurate and timely detection of epileptic seizures is crucial for patient management and treatment.
• Existing methods for seizure detection often face challenges in capturing the complex dynamics of brain activity.

Purpose of the Study:

• To propose a novel dynamic graph embedding model for the detection of epileptic seizures.
• To leverage graph entropy measurements to differentiate seizure activity from normal brain function.
• To enhance the accuracy of epileptic seizure detection using multi-channel EEG signals.

Main Methods:

• Constructing a dynamic graph by identifying correlations within multi-channel EEG signals.
• Utilizing graph entropy to measure similarity between graphs across time intervals.
• Developing an entropy-based dynamic graph embedding model for clustering and discriminating seizure graphs.
• Applying the model to the Children Hospital Boston-MIT Scalp EEG database.

Main Results:

• The proposed entropy-based dynamic graph embedding model effectively clusters EEG graphs.
• Seizure-related graphs were successfully discriminated from non-seizure graphs.
• The approach demonstrated a 1.4% improvement in accuracy compared to baseline methods.

Conclusions:

• The dynamic graph embedding model, incorporating graph entropy, is a promising approach for epileptic seizure detection.
• This method offers a novel way to analyze complex EEG signal dynamics for neurological disorder identification.
• The findings suggest potential for improved diagnostic tools in epilepsy management.