SparNet: A Convolutional Neural Network for EEG Space-Frequency Feature Learning and Depression Discrimination.

Area of Science:

• Neuroscience and Computational Psychiatry
• Application of Artificial Intelligence in Clinical Diagnosis

Background:

• Major depressive disorder (MDD) is a significant global health issue.
• Electroencephalogram (EEG) signals offer insights into brain mechanisms underlying MDD.
• Existing deep learning models for depression recognition often lack specialized spatial and frequency domain feature learning for different brain regions.

Purpose of the Study:

• To propose SparNet, a novel convolutional neural network (CNN) designed for EEG-based depression detection.
• To enhance the classification accuracy of depression recognition by learning space-frequency domain characteristics in different brain regions.
• To evaluate the effectiveness of SparNet in distinguishing between individuals with MDD and healthy controls.

Main Methods:

• Development of SparNet, a CNN incorporating five parallel convolutional filters and the Squeeze-and-Excitation network (SENet).
• Utilizing EEG signals to learn combined spatial and frequency domain features.
• Employing a subject-wise cross-validation method for model training and testing.

Main Results:

• SparNet achieved a high classification accuracy of 94.37%.
• The model demonstrated strong performance with a sensitivity of 95.07% and a specificity of 93.66%.
• The findings indicate successful differentiation between depressive and normal control groups.

Conclusions:

• The proposed SparNet model is effective for detecting depression using EEG signals.
• Integrating spatial and frequency domain information significantly improves the identification of individuals with depression.
• This approach holds promise for advancing objective diagnostic tools for MDD.