A Fine-grained Hemispheric Asymmetry Network for accurate and interpretable EEG-based emotion classification.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Emotion classification from electroencephalography (EEG) data is crucial for understanding brain activity.
• Leveraging hemispheric asymmetry in EEG signals offers a promising avenue for improved emotion recognition.
• Existing methods often lack the fine-grained spectral analysis needed to capture subtle hemispheric differences.

Purpose of the Study:

• To propose a novel deep learning model, the Fine-grained Hemispheric Asymmetry Network (FG-HANet), for accurate and interpretable emotion classification.
• To utilize fine-grained hemispheric asymmetry features within narrow frequency bands from raw EEG data.
• To investigate hemispheric dominance and asymmetry patterns during different emotional states.

Main Methods:

• Developed an end-to-end deep learning model (FG-HANet) incorporating hemispheric asymmetry.
• Extracted features from original and mirrored EEG inputs.
• Applied Finite Impulse Response (FIR) filters with 2-Hz granularity for fine-grained spectral analysis.
• Implemented a three-stage training pipeline to enhance attention to asymmetry features.

Main Results:

• Achieved state-of-the-art accuracy on two public datasets (SEED: 97.11%, SEED-IV: 85.70%).
• Demonstrated the model's superior performance in emotion classification.
• Identified hemispheric dominance and asymmetry within 2-Hz frequency bands across individuals and emotional states.

Conclusions:

• The FG-HANet model effectively leverages fine-grained hemispheric asymmetry for robust emotion classification.
• Results align with and extend previous neuroscience findings on brain lateralization and emotion.
• The study provides novel insights into the neural mechanisms of emotion generation through EEG analysis.