Crucial rhythms and subnetworks for emotion processing extracted by an interpretable deep learning framework from EEG...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Electroencephalogram (EEG) brain networks reveal relationships between brain regions and can identify emotional states.
• Current methods lack interpretable structural features for EEG brain networks.
• Identifying distinct emotional states from brain activity remains a challenge.

Purpose of the Study:

• To propose a novel deep learning model for extracting discriminant and interpretable features from EEG brain networks for emotion recognition.
• To enhance the contribution of crucial rhythms and subnetworks for improved emotion classification.
• To improve the generalization performance of models for cross-subject emotion recognition tasks.

Main Methods:

• A novel deep learning structure combining an attention mechanism and a domain adversarial strategy was developed.
• The attention mechanism highlights significant rhythms and subnetworks for emotion recognition.
• A domain-adversarial module was employed to enhance cross-subject generalization.

Main Results:

• The proposed method significantly improved classification accuracy for different emotions compared to existing methods.
• The model demonstrated effectiveness in subject-independent emotion recognition tasks.
• Crucial rhythms and subnetwork structures for emotion processing were identified and validated.

Conclusions:

• The novel deep learning approach effectively improves EEG brain network classification for emotion recognition.
• The method provides a valuable tool for understanding emotion processing mechanisms.
• Interpretable features extracted from brain networks offer insights into neural correlates of emotion.