LG-TriCapsNet: A lightweight graph capsule framework with nearest neighbor graphs for multi-disease EEG...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Accurate classification of neurological disorders via EEG signals is critical but challenging due to signal complexity.
• Traditional methods struggle to capture intricate temporal and spatial features in EEG data.
• Existing approaches often lack efficiency and the ability to handle diverse neurological conditions.

Purpose of the Study:

• To introduce a novel framework, Lightweight Graph Triplet Capsule Networks (LG-TriCapsNet) with Nearest Neighbor Graphs (NNG), for enhanced EEG signal classification.
• To effectively leverage both temporal and spatial information inherent in EEG data for improved diagnostic accuracy.
• To provide a computationally efficient and real-time solution for automated neurological disease detection.

Main Methods:

• Utilizing graph-based representations to enhance information spread and feature extraction from EEG signals.
• Integrating Nearest Neighbor Graphs (NNG) to dynamically capture spatial-temporal dependencies between EEG channels.
• Employing a combination of graph structures and capsule networks (LG-TriCapsNet) to handle complex EEG data characteristics.

Main Results:

• LG-TriCapsNet achieved high performance metrics: 98.32% F1 score, 98.34% accuracy, 98.30% sensitivity, and 98.40% specificity.
• The proposed model demonstrated superior performance compared to current state-of-the-art methods in EEG classification.
• The graph-based approach with capsule networks effectively improved feature discrimination across various neurological conditions.

Conclusions:

• LG-TriCapsNet offers a computationally efficient and effective solution for the automated classification of neurological disorders using EEG signals.
• The novel combination of graph structures and capsule networks significantly enhances the analysis of complex EEG data.
• This framework holds substantial potential for advancing clinical decision-making and patient care in automated neurological disease diagnosis.