Predicting Major Depressive Disorder Using Neural Networks from Spectral Measures of EEG Data.

Area of Science:

• Computational Neuroscience
• Psychiatry
• Medical Informatics

Background:

• Electroencephalogram (EEG) processing with neural networks is crucial in modern medicine.
• Major Depressive Disorder (MDD) diagnosis often relies on subjective assessments.
• Objective biomarkers for MDD are needed to improve clinical decision-making.

Purpose of the Study:

• To develop and evaluate a neural network-based method for MDD diagnosis.
• To combine spectral characteristics of resting-state EEG with psychological questionnaire data.
• To assess the performance of different algorithms for classification and regression tasks in MDD.

Main Methods:

• Collected resting-state EEG and psychological data from 71 participants (42 healthy, 29 with MDD).
• Evaluated traditional machine learning, deep learning (LSTM), ablation analysis, and feature importance analysis.
• Performed binary classification (healthy vs. MDD) and regression for predicting Beck Depression Inventory (BDI) scores.

Main Results:

• An LSTM network on delta-rhythm EEG data achieved R² = 0.742 for predicting depression severity, an 86% improvement over Ridge regression.
• Ablation studies identified delta and alpha EEG rhythms as key neurophysiological biomarkers.
• Feature importance analysis highlighted ruminative thinking, age, and hostility as dominant psychometric predictors.

Conclusions:

• LSTM applied to spectral EEG data offers superior quantitative assessment of depression severity.
• Logistic Regression on psychometric or EEG data provides a reliable tool for MDD screening and diagnosis.
• This objective methodology enhances clinical decision-making and personalized treatment monitoring for MDD.