ST-SCGNN: A Spatio-Temporal Self-Constructing Graph Neural Network for Cross-Subject EEG-Based Emotion Recognition...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Disorders of Consciousness (DOC) present significant challenges in clinical assessment due to limited training data for EEG-based emotion recognition.
• Accurate emotion recognition is crucial for understanding brain function and detecting covert consciousness.

Purpose of the Study:

• To propose a novel spatio-temporal self-constructing graph neural network (ST-SCGNN) for cross-subject emotion recognition and consciousness detection.
• To evaluate the ST-SCGNN's effectiveness in recognizing emotions and detecting consciousness in patients with DOC.

Main Methods:

• Extraction and combination of activation and connection pattern features for spatio-temporal feature generation.
• Implementation of a self-constructing graph neural network that dynamically updates its structure based on input signals.
• Cross-subject emotion recognition experiments using SEED and SEED-IV datasets, followed by clinical application in patients with DOC.

Main Results:

• The ST-SCGNN achieved average accuracies of 85.90% (SEED) and 76.37% (SEED-IV), surpassing state-of-the-art metrics.
• In a clinical setting, two out of eight patients with DOC showed accuracies significantly higher than chance level.
• These patients exhibited neural patterns similar to healthy subjects, indicating covert consciousness and emotion-related abilities.

Conclusions:

• The ST-SCGNN is a powerful tool for cross-subject emotion recognition, outperforming existing methods.
• The model demonstrates potential as a non-invasive tool for consciousness detection in patients with disorders of consciousness.
• The findings suggest the presence of covert consciousness and emotion processing capabilities in some DOC patients.