Epileptic Seizure Detection in Separate Frequency Bands Using Feature Analysis and Graph Convolutional Neural Network (GCN) from Electroencephalogram (EEG) Signals

A team of researchers has published a new AI model that significantly improves the detection of epileptic seizures from EEG data. The framework, detailed in an arXiv preprint, moves beyond conventional methods by analyzing brainwave activity in five separate frequency bands—delta, theta, alpha, lower beta, and higher beta. For each band, the model extracts eleven distinct features before applying a Graph Convolutional Neural Network (GCN). The GCN is key, as it treats EEG electrodes as nodes in a graph, allowing the AI to understand the complex spatial dependencies and connections across the scalp, which is critical for identifying seizure patterns.

Tested on the widely-used CHB-MIT scalp EEG dataset, the model delivered exceptional results. It achieved an overall accuracy of 99.01% across all frequency bands. Performance varied by band, with the alpha and beta bands showing remarkable accuracy of 99.5% and 99.7%, respectively, highlighting their strong discriminative power for seizure activity. In contrast, the higher beta band had a much lower accuracy of 51.4%, revealing specific frequency ranges where seizure signatures are less distinct. This frequency-aware approach not only boosts accuracy but also provides neurologists with interpretable, neurophysiologically relevant insights, marking a step forward from less transparent 'black box' deep learning models.