Brain-computer interfaces inspired spiking neural network model for depression stage identification.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Computational Psychiatry

Background:

• Depression diagnosis traditionally relies on subjective scales and clinical evaluations, risking misdiagnosis.
• Existing deep learning methods for diagnosis require significant computational power and lack physiological interpretability.
• Brain-Computer Interfaces (BCIs) offer a promising avenue for objective, physiologically-based depression diagnosis.

Purpose of the Study:

• To develop and evaluate a novel brain-like learning model for the assisted diagnosis of depression.
• To improve the accuracy and interpretability of AI-driven diagnostic tools for mental health.
• To reduce the energy consumption associated with deep learning models in clinical neuro-technologies.

Main Methods:

• Collected 128-channel electroencephalogram (EEG) data from individuals.
• Constructed a 128x128 brain adjacency matrix, reduced to a 90x90 matrix for input.
• Developed a spiking neural network (SNN) for functional classification and complex network analysis for structural topology.

Main Results:

• The spiking neural network achieved a diagnostic accuracy exceeding 97.5% for depression.
• The SNN model demonstrated significantly lower energy consumption compared to deep convolutional methods.
• Analysis of complex networks identified potential abnormal brain functional connections in individuals with depression.

Conclusions:

• The proposed brain-like learning model offers a highly accurate and energy-efficient method for depression diagnosis.
• The approach enhances physiological interpretability, addressing limitations of traditional deep learning in clinical applications.
• This study highlights the potential of SNNs and complex network analysis in advancing neuro-technologies for mental health.