MVCA-Net: Multi-View Convolution Attention Network for measuring EEG rhythms representing Anxiety.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Psychiatry

Background:

• Anxiety significantly impacts daily life, with traditional assessments relying on self-report questionnaires.
• Advances in neuroimaging and computer-aided technologies offer potential for enhanced anxiety diagnosis.
• Neural patterns associated with anxiety are complex and require sophisticated analysis.

Purpose of the Study:

• To develop and validate a novel deep learning model for anxiety assessment using electroencephalogram (EEG) signals.
• To extract frequency-based features from EEG data to identify neural patterns indicative of anxiety.
• To evaluate the model's accuracy in classifying different levels of anxiety severity.

Main Methods:

• A deep learning model combining a convolutional neural network (CNN), multi-head attention transformer, and attention module was designed.
• The model processed frequency-based features extracted from EEG signals.
• Validation was performed on the publicly available DASPS EEG dataset, categorizing participants into normal and anxious states (further subdivided by severity).

Main Results:

• The model achieved 82.94% accuracy for binary classification (normal vs. anxious).
• Average accuracy for multi-class classification (normal, mild, moderate, severe anxiety) was 74.05%.
• The findings demonstrate the model's capability in distinguishing anxiety levels based on EEG features.

Conclusions:

• The proposed deep learning model effectively utilizes EEG frequency features for anxiety assessment.
• This approach shows promise for improving the accuracy and objectivity of anxiety diagnosis across various severity levels.
• Leveraging advanced AI techniques with neuroimaging data can offer new avenues for understanding and managing anxiety disorders.