Classifying depression patients and normal subjects using machine learning techniques and nonlinear features from EEG...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Psychiatry

Background:

• Early diagnosis of depression is crucial for effective treatment and patient outcomes.
• Electroencephalogram (EEG) signals offer a non-invasive window into brain activity.
• Distinguishing depression from normal states using EEG requires sophisticated analytical methods.

Purpose of the Study:

• To investigate the efficacy of nonlinear EEG analysis for discriminating between unmedicated depressed patients and healthy controls.
• To evaluate the performance of various nonlinear features and machine learning classifiers in depression detection.
• To determine if combining nonlinear features enhances diagnostic accuracy.

Main Methods:

• Extracted four nonlinear features (DFA, Higuchi fractal, correlation dimension, Lyapunov exponent) and power of four EEG bands from 45 depressed patients and 45 controls.
• Employed k-nearest neighbor, linear discriminant analysis, and logistic regression classifiers.
• Utilized a genetic algorithm for feature selection to identify the most discriminative features.

Main Results:

• Correlation dimension with logistic regression yielded the highest accuracy (83.3%) among individual nonlinear features.
• Combining all nonlinear features with logistic regression achieved a classification accuracy of 90%.
• The proposed method, particularly with combined nonlinear features, demonstrated superior performance compared to other reported techniques.

Conclusions:

• Nonlinear analysis of EEG signals is a promising and effective method for discriminating depressed individuals from normal subjects.
• Combining multiple nonlinear EEG features significantly enhances diagnostic accuracy.
• This approach can serve as a valuable complementary tool for psychiatrists in diagnosing depression.