TSUnet-CC: Temporal Spectrogram Unet embedding Cross Channel-wise attention mechanism for MDD identification.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Major Depressive Disorder (MDD) diagnosis relies on clinical assessment, lacking objective biomarkers.
• Electroencephalography (EEG) signals offer a potential avenue for objective MDD detection.
• Existing deep learning models face challenges with redundant information in EEG data.

Purpose of the Study:

• To develop an advanced deep learning model for accurate MDD detection using multi-channel EEG signals.
• To address the issue of redundant information in temporal and spectrogram EEG features.
• To enhance the identification of key EEG features indicative of MDD.

Main Methods:

• Proposed a novel Temporal Spectrogram Unet with cross channel-wise attention (TSUnet-CC).
• Utilized a symmetrical two-stream U-net architecture for processing temporal and spectrogram features.
• Incorporated a multi-scale saliency-encoded spectrogram using a Fourier-based approach.
• Employed a cross channel-wise attention mechanism to prioritize informative EEG channels.

Main Results:

• TSUnet-CC achieved high classification accuracy: 98.55% for eyes closed and 99.22% for eyes open datasets.
• The model outperformed several state-of-the-art methods in MDD detection.
• The cross channel-wise attention effectively highlighted crucial EEG features for MDD identification.

Conclusions:

• The TSUnet-CC model demonstrates significant potential for objective and accurate MDD detection.
• The proposed method offers a promising tool for clinical application in mental health.
• This approach advances the use of EEG signal processing in psychiatric diagnostics.