Using Machine Learning to Model EEG-Derived Brain Activity During Emotion Regulation.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Emotion Regulation (ER) is crucial for mental health and social functioning, particularly under stress.
• Understanding the neural mechanisms of ER is vital for developing effective interventions.
• Existing methods for assessing ER can be subjective; objective measures are needed.

Purpose of the Study:

• To investigate brain activity patterns during emotion regulation using electroencephalography (EEG).
• To develop and validate machine learning (ML) models for predicting successful versus unsuccessful ER.
• To identify specific EEG signal characteristics (time and frequency domains) associated with ER success.

Main Methods:

• Participants viewed emotional/neutral images and were instructed to either view normally or regulate emotions.
• EEG data were collected and analyzed in both time (Global Field Power) and frequency (Power Spectral Density) domains.
• ML models, including a neural network with Maximum Mean Discrepancy (MMD) loss, were trained on EEG features to predict ER success.

Main Results:

• Significant differences in Global Field Power (GFP) were found in frontal and central brain regions.
• Theta, beta, and gamma frequency bands were identified as important for ER.
• A subject-independent neural network achieved a 75.57% F1-score macro for predicting ER success, emphasizing frontal regions and beta frequencies.

Conclusions:

• Combining EEG data with advanced ML provides an accurate, objective framework for assessing ER.
• Frontal brain regions and beta frequency signals are key predictors of emotion regulation levels.
• This EEG-based approach offers a novel method for ER assessment and personalized mental health treatments.