An Optimal Channel Selection for EEG-Based Depression Detection via Kernel-Target Alignment.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Machine Learning

Background:

• Depression is a major global health issue characterized by persistent low mood and cognitive dysfunction.
• Low recognition and treatment rates highlight the urgent need for effective depression detection tools.
• Multichannel electroencephalogram (EEG) signals offer a promising objective measure of brain activity for depression diagnosis.

Purpose of the Study:

• To address information redundancy and computational complexity in multichannel EEG for depression detection.
• To propose an optimal channel selection method for enhancing the feasibility and practicality of EEG-based depression diagnosis.
• To improve the accuracy and efficiency of detecting depression using electroencephalogram data.

Main Methods:

• Developed an optimal channel selection strategy using a modified kernel-target alignment (KTA) approach.
• KTA measures similarity between channel selection kernel matrices and target matrices as an objective function.
• Optimized the objective function to identify the most informative EEG channels for depression detection.

Main Results:

• Channel selection significantly improved classification performance in EEG-based depression detection.
• Acceptable diagnostic results were achieved using only a small subset of selected EEG channels.
• The selected channels corresponded with expected patterns of cortical activity relevant to depression.

Conclusions:

• The proposed KTA-based channel selection method is effective for EEG-based depression detection.
• This approach enhances machine learning feasibility and clinical practicality by reducing data complexity.
• The method outperforms existing state-of-the-art channel selection techniques for depression diagnosis.