An Effective and Interpretable EEG-Based Depression Recognition Method Using Hybrid Feature Selection.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Machine Learning

Background:

• Deep learning models for EEG-based depression detection are complex and lack interpretability.
• High computational demands limit the clinical practicality of current automated methods.

Purpose of the Study:

• To propose an efficient and interpretable method for automated depression recognition using EEG signals.
• To enhance classification performance while reducing computational complexity for practical clinical application.

Main Methods:

• Extracted multi-domain features from preprocessed EEG signals.
• Integrated RankSearch for efficient feature screening and Genetic Algorithm (GA) for interactive optimization.
• Evaluated performance using various machine learning classifiers on the HUSM dataset.

Main Results:

• Achieved high classification performance: accuracy = 95.08%, sensitivity = 95.99%, specificity = 94.30%, F1-score = 95%, AUC = 0.9514.
• Demonstrated feature-level interpretability through feature importance analysis.
• Showcased lower computational complexity and higher clinical practicality compared to existing models.

Conclusions:

• The proposed efficient feature selection method significantly improves EEG-based depression detection.
• This approach offers a more interpretable and computationally practical solution for objective depression diagnosis.