TFFBN-HDLF: a hybrid deep learning framework based on time-frequency functional brain networks for epileptic seizure...

Area of Science:

• Neuroscience and Artificial Intelligence
• Medical signal processing and machine learning

Background:

• Epilepsy seizure detection in the elderly using electroencephalogram (EEG) is crucial for clinical decision support systems.
• Challenges exist due to slow background activity and complex non-stationary brain signal dynamics in elderly patients.
• Existing methods struggle to extract robust, discriminative features across diverse elderly individuals.

Purpose of the Study:

• To propose a hybrid deep learning framework, TFFBN-HDLF, to improve the reliability and diagnostic accuracy of AI-assisted epilepsy seizure monitoring in the elderly.
• To address the limitations of current methods in capturing complex EEG characteristics in older adults.

Main Methods:

• Developed a time-frequency functional brain network construction method (TFFBNC) using Pearson correlation coefficient (PCC) and phase lag index (PLV) to create a time-frequency fused functional brain network (TFPPNet).
• Constructed a hybrid deep learning architecture, SeizureTransNet, combining convolutional neural networks (CNNs) with enhanced Transformer modules.
• The architecture dynamically integrates multi-scale spatiotemporal features for adaptive seizure detection in aging brains.

Main Results:

• The TFFBN-HDLF framework achieved high accuracy on public datasets: 98.09% (AUC 99.45%) on CHB-MIT and 92.49% (AUC 95.64%) on Siena.
• Demonstrated effectiveness in capturing synchronous neural interactions in both time and frequency domains for elderly brain signals.
• Showcased adaptability to age-related EEG pattern differences.

Conclusions:

• The collaborative integration of attention-based time-frequency network fusion and deep feature learning significantly enhances diagnostic performance for elderly epilepsy detection.
• The TFFBN-HDLF framework shows strong potential for application in clinical care for epilepsy in the elderly.
• This approach provides a robust foundation for intelligent clinical decision support systems in geriatric neurology.