Using convolutional neural networks to decode EEG-based functional brain network with different severity of...

Area of Science:

• Neuroscience
• Computational psychiatry

Background:

• Acrophobia prevalence is increasing, necessitating advanced diagnostic tools.
• Neuroscience approaches, particularly for virtual reality exposure therapy (VRET), are emerging.
• Electroencephalography (EEG) is underutilized for acrophobia assessment.

Purpose of the Study:

• To assess the efficacy of EEG in objectively quantifying acrophobia severity.
• To explore the potential of functional brain networks (FBNs) derived from EEG for acrophobia diagnosis.

Main Methods:

• EEG data collected during virtual reality (VR) exposure experiments.
• Functional brain networks (FBNs) constructed using synchronization likelihood.
• Machine learning (ML) and convolutional neural networks (CNNs) trained on FBN features.

Main Results:

• FBN features effectively distinguished between different acrophobia severity groups.
• CNNs outperformed traditional ML algorithms in classification accuracy.
• A ResNet-based CNN achieved a high accuracy of 98.46 ± 0.42%.

Conclusions:

• Community structures within specific cerebral cortex regions are indicative of acrophobia severity.
• The proposed CNN framework offers objective feedback for potential closed-loop VRET systems.