Identifying relevant asymmetry features of EEG for emotion processing.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• The brain's left and right hemispheres process emotions differently, with models suggesting right hemisphere dominance or specialized processing of positive (left) and negative (right) emotions.
• Previous research leveraged hemisphere asymmetry models for emotion classification in machine learning, but understanding the link between asymmetry and emotion processing via machine learning remains underexplored.

Purpose of the Study:

• To investigate the role of brain hemisphere asymmetry in processing four distinct emotions: happiness, sadness, fear, and neutral states.
• To explore how machine learning models can identify associations between brain asymmetries and emotion processing using electroencephalogram (EEG) data.

Main Methods:

• Analyzed EEG signals from 15 subjects experiencing four emotions via video clips.
• Derived asymmetry features by calculating the log ratio of relative energy between symmetrical brain nodes.
• Trained four binary logistic regression models, one per emotion, to identify relevant predictive features.

Main Results:

• Achieved average AUC-ROC values ranging from 51.6% (fear) to 58.4% (happy), with comparable results on an independent dataset.
• Observed significant brain lateralization primarily in alpha frequency bands; other bands involved both hemispheres.
• Logistic regression highlighted gamma and alpha bands, especially in lateral frontal, parietal, and temporal EEG pairs (e.g., FT7-FT8), as most relevant for emotion prediction.

Conclusions:

• Brain hemisphere asymmetry plays a role in emotion processing, with specific frequency bands and brain regions being more critical than others.
• Findings indicate that alpha and gamma frequency bands, particularly in lateral frontal, parietal, and temporal areas, are crucial for developing accurate emotion recognition models.