GNN-Based Depression Recognition Using Spatio-Temporal Information: A fNIRS Study.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Medical Imaging Analysis

Background:

• Depression is a growing global health concern, necessitating advanced diagnostic tools.
• Functional near-infrared spectroscopy (fNIRS) shows promise for auxiliary depression diagnosis.
• Both temporal (direct data) and spatial (functional connectivity) fNIRS features are effective for depression recognition.

Purpose of the Study:

• To propose and evaluate a novel Graph Neural Network (GNN) method for automatic depression recognition.
• To integrate both temporal and spatial features from fNIRS data within a unified framework.
• To enhance the accuracy and reliability of depression detection using neuroimaging data.

Main Methods:

• Collected and pre-processed fNIRS data from 96 subjects.
• Extracted temporal features (statistical metrics per channel) and spatial features (channel connectivity: coherence, correlation).
• Developed a GNN model treating subject data as graphs, with temporal features as node features and spatial features as edge weights.

Main Results:

• The GNN-based method demonstrated superior performance in depression recognition compared to classical machine learning approaches.
• Achieved significant improvements in accuracy, F1 score, and precision.
• Specifically, the F1 score showed an improvement of over 10%.

Conclusions:

• The proposed GNN method effectively combines temporal and spatial fNIRS data for robust depression recognition.
• This approach offers a promising advancement for computer-aided diagnosis of depression.
• The findings highlight the potential of GNNs in analyzing complex neuroimaging data for mental health applications.