TAU-DI Net: A Multi-Scale Convolutional Network Combining Prob-Sparse Attention for EEG-based Depression...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Informatics

Background:

• Major Depression Disorder (MDD) diagnosis relies heavily on clinical assessments, with electroencephalogram (EEG) signals offering potential for objective biomarkers.
• Existing deep learning models (CNN, LSTM, attention) for EEG-based MDD detection often overlook signal-level pathological features or redundant information in resting-state EEG.

Purpose of the Study:

• To develop a novel deep learning architecture for improved EEG-based detection of Major Depression Disorder (MDD).
• To address limitations in current models by extracting multi-frequency information and handling redundant data in resting-state EEG signals.

Main Methods:

• Proposed an adaptive time-frequency distribution network combining frequency-periodic transformation and multi-scale CNNs.
• Employed adaptive weighted fusion of spatiotemporal representations across frequencies.
• Utilized down-sampled Prob-Sparse Attention to distill reliable patterns from resting-state EEG data.

Main Results:

• The proposed adaptive network achieved a classification accuracy of 94.91% for MDD detection from EEG signals.
• Demonstrated superior performance compared to existing self-attention and convolutional neural network models.
• Highlighted the efficacy of adaptive, frequency-specific processing for EEG-based depression analysis.

Conclusions:

• The novel adaptive time-frequency distribution network offers a promising approach for accurate and objective MDD diagnosis using EEG.
• Adaptive methods leveraging different frequency bands enhance the processing of depression-related EEG signals.
• This approach has the potential to significantly aid in the early detection and treatment of MDD.