A machine learning framework involving EEG-based functional connectivity to diagnose major depressive disorder (MDD)

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Major Depressive Disorder (MDD) presents diagnostic challenges and significant societal impact.
• Early and accurate diagnosis of MDD is crucial for effective treatment and management.
• Existing diagnostic methods may lack the precision required for timely intervention.

Purpose of the Study:

• To propose and evaluate a machine learning framework for the automatic diagnosis of MDD.
• To investigate the efficacy of electroencephalography (EEG)-derived synchronization likelihood (SL) features in discriminating MDD patients from healthy controls.
• To compare the performance of SL features against other EEG-based measures like interhemispheric coherence and mutual information.

Main Methods:

• A machine learning framework was developed using EEG-derived SL features as input.
• Classification models including Support Vector Machine (SVM), Logistic Regression (LR), and Naïve Bayesian (NB) were employed.
• These models were trained to differentiate between individuals with MDD and healthy controls based on their EEG SL features.

Main Results:

• The machine learning models demonstrated classification rates significantly better than chance.
• SVM achieved the highest accuracy (98%) with high sensitivity (99.9%) and specificity (95%).
• LR and NB models also showed strong performance, with NB achieving 100% sensitivity.

Conclusions:

• Synchronization Likelihood (SL) derived from EEG is a promising feature for the accurate diagnosis of MDD.
• The developed machine learning framework shows potential for creating robust computer-aided diagnosis (CAD) tools for clinical use.
• These findings could facilitate earlier and more reliable detection of depression, improving patient outcomes.