Cross-subject affective analysis based on dynamic brain functional networks.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Emotion recognition is vital for human-computer interaction.
• Electroencephalography (EEG) signals offer direct insights into brain activity for enhanced emotion recognition.
• Inter-subject variability and signal non-smoothness pose challenges for EEG-based emotion recognition models.

Purpose of the Study:

• To develop a novel approach for robust EEG-based emotion recognition.
• To address the challenge of inter-subject variability in EEG signals.
• To improve the generalization performance of emotion recognition models across different subjects.

Main Methods:

• A novel approach combining time-frequency analysis and dynamic brain functional networks was proposed.
• Sliding time windows were used to construct dynamic brain functional networks, integrating time, frequency, and spatial domains.
• Mutual information quantified channel correlations for network construction, followed by extraction of global efficiency, local efficiency, and local clustering coefficients for feature extraction.

Main Results:

• The dynamic brain functional network approach outperformed static networks in subject-dependent, subject-independent, and subject- and trial-independent experiments.
• High classification accuracies of 90.89% (valence) and 91.17% (arousal) were achieved in subject-independent experiments.
• Analysis revealed distinct roles for temporal lobes (private emotional information) and other regions (basic emotional information).

Conclusions:

• The proposed dynamic brain functional network method significantly advances EEG-based emotion recognition by improving generalization.
• This approach effectively captures dynamic brain features, reducing inter-individual differences.
• The findings highlight the potential of dynamic brain network analysis for more accurate and reliable emotion recognition systems.