Differentiating depression subtypes through three-electrode EEG and emotional stimuli: a machine learning approach

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Depression diagnosis often relies on subjective criteria, necessitating objective biomarkers.
• Understanding neural dynamics related to emotional processing is crucial for depression research.
• Subtyping depression (e.g., first episode vs. recurrent) may inform personalized treatment.

Purpose of the Study:

• To develop and validate an electroencephalography (EEG)-based framework for discriminating between depression subtypes.
• To investigate emotion-modulated neural dynamics elicited by immersive virtual reality (VR) experiences.
• To assess the efficacy of a novel Bi-Emotional Siamese Network (BESN) for depression classification.

Main Methods:

• Recorded EEG from 70 participants (healthy controls, first depressive episode, recurrent depressive episode) during positive and negative VR conditions.
• Utilized a linear mixed-effects model for group-by-condition interaction analysis.
• Developed and applied a Bi-Emotional Siamese Network (BESN) with path-signature temporal encoding for classification.

Main Results:

• Feature-level analyses indicated differential emotion-related neural modulation across participant groups.
• The BESN achieved 83.1% accuracy in distinguishing healthy controls from first depressive episode participants.
• Three-class classification (healthy controls, FDE, RDE) yielded 70.6% accuracy, outperforming baseline models.
• The framework demonstrated robustness on an external dataset, achieving 78.2% accuracy.

Conclusions:

• Baseline-anchored, emotion-modulated EEG dynamics are a viable objective measure for depression group discrimination.
• The BESN model, integrating temporal dynamics, offers a generalizable computational approach for psychiatric assessment.
• This framework holds potential for improving the objective diagnosis and subtyping of depression.