Feature Selection and Classification of Electroencephalographic Signals: An Artificial Neural Network and Genetic...

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Major Depressive Disorder (MDD) diagnosis often relies on subjective assessments.
• Electroencephalography (EEG) offers objective biomarkers, but high dimensionality poses challenges.
• Feature selection is crucial for improving classification accuracy in complex datasets.

Purpose of the Study:

• To develop and evaluate an optimized classification method for MDD using EEG data.
• To enhance diagnostic accuracy by reducing feature dimensionality.
• To assess the efficacy of a combined genetic algorithm (GA) and back-propagation neural network (BPNN) approach.

Main Methods:

• Utilized pre-treatment 6-channel EEG data from 147 MDD patients.
• Employed a genetic algorithm (GA) for feature selection to identify discriminant EEG patterns in theta and delta bands.
• Applied a back-propagation neural network (BPNN) for classification using the reduced feature set.
• Validated performance using 6-fold cross-validation.

Main Results:

• The GA effectively reduced redundant and less discriminant EEG features.
• The optimized classification method achieved an overall accuracy of 89.12%.
• The area under the receiver operating characteristic curve (AUC) reached 0.904, indicating strong classification performance.

Conclusions:

• The GA-BPNN approach significantly enhances classification accuracy for MDD diagnosis compared to traditional methods.
• Reduced feature sets derived from frontal EEG slow bands can yield high diagnostic performance.
• This method offers a promising tool for objective MDD assessment in clinical settings.