A Functional Connectivity-Based Model With a Lightweight Attention Mechanism for Depression Recognition Using EEG...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Attention mechanisms are commonly used for feature extraction in depression recognition from EEG data.
• Standard multi-head self-attention in the spatial domain of EEG is computationally intensive and parameter-heavy.
• Model performance can be unstable due to random parameter initialization.

Purpose of the Study:

• To design a lightweight attention mechanism for efficient EEG-based depression recognition.
• To develop a deep learning model, the Functional Connectivity Attention Network (FCAN), utilizing this mechanism.
• To improve computational efficiency and model stability compared to standard methods.

Main Methods:

• Developed a lightweight attention mechanism inspired by multi-head self-attention.
• Constructed the Functional Connectivity Attention Network (FCAN) incorporating spatial attention and feature integration modules.
• Evaluated FCAN performance on a public EEG dataset against baseline models.

Main Results:

• The proposed lightweight attention mechanism significantly reduces model parameters and computational costs.
• FCAN achieved a classification accuracy of 95.20% (±3.99%) on the EEG dataset.
• FCAN demonstrated superior classification performance compared to existing baseline models.

Conclusions:

• The developed lightweight attention mechanism and FCAN model offer an efficient and effective approach for depression recognition using EEG.
• FCAN provides a stable and computationally feasible solution for clinical applications.
• This work contributes to advancing automated mental health diagnostics through improved deep learning techniques.