Spatio-Temporal-Spectral Hierarchical Graph Convolutional Network With Semisupervised Active Learning for...

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Electroencephalogram (EEG) signal analysis is crucial for seizure prediction.
• Current deep learning methods often overlook spatial and temporal dependencies in epileptic brains, leading to suboptimal performance.
• Accurate seizure prediction requires capturing complex spatio-temporal-spectral characteristics of brain activity.

Purpose of the Study:

• To propose a novel patient-specific EEG seizure predictor.
• To address limitations in current deep learning approaches for seizure prediction.
• To enhance the robustness and accuracy of automatic seizure prediction.

Main Methods:

• A spatio-temporal-spectral hierarchical graph convolutional network with an active preictal interval learning scheme (STS-HGCN-AL) was developed.
• The framework infers a hierarchical graph to characterize epileptic cortex activity across different rhythms.
• Temporal dependencies and spatial couplings are extracted using spectral-temporal convolutional neural networks and a self-gating mechanism, integrated via a hierarchical graph convolutional network.

Main Results:

• The STS-HGCN-AL framework effectively characterizes epileptic brain activity under different rhythms.
• It successfully captures and integrates critical intrarhythm spatiotemporal properties.
• The active learning strategy optimizes the preictal interval for individual patients, enhancing prediction robustness.

Conclusions:

• The proposed STS-HGCN-AL method demonstrates efficacy in extracting critical preictal biomarkers from EEG signals.
• This approach offers promising capabilities for automatic and patient-specific seizure prediction.
• The study highlights the importance of integrating spectral, temporal, and spatial information for improved seizure prediction accuracy.