CastNet: A three-channel EEG-based deep learning model for cross-subject depression detection

Area of Science:

• Neuroscience
• Artificial Intelligence
• Psychiatry

Background:

• Depression is a prevalent mental health disorder impacting millions globally.
• Deep learning models have shown promise in electroencephalogram (EEG)-based depression diagnosis.

Purpose of the Study:

• To propose CastNet, a novel deep learning model for depression detection using three-channel EEG data.
• To enhance feature learning and address computational complexity in EEG analysis for depression diagnosis.

Main Methods:

• CastNet integrates Convolutional Neural Networks (CNN), Transformer, and Long Short-Term Memory (LSTM) for multi-level EEG feature learning.
• A feature enhancement module, depthwise separable convolution, and a novel coupled bidirectional LSTM are employed.
• Leave-One-Subject-Out (LOSO) cross-validation was used on MPHC and PRED+CT datasets.

Main Results:

• CastNet achieved classification accuracies of 97.2% on the MPHC dataset and 87.5% on the PRED+CT dataset.
• The model outperformed existing methods in EEG-based depression diagnosis.
• The study highlights the effectiveness of multi-level feature learning and advanced LSTM fusion.

Conclusions:

• CastNet demonstrates significant potential for accurate depression diagnosis using limited EEG channels.
• The findings offer new insights into EEG signal patterns associated with depression.
• This research supports the application of advanced deep learning techniques in clinical mental health assessment.