Using effective connectivity-based predictive modeling to predict MDD scale scores from multisite rs-fMRI data.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Medical Imaging

Background:

• Major depressive disorder (MDD) is a severe mental illness impacting millions globally.
• Accurate quantification of MDD symptom severity is crucial for effective treatment.
• Current methods often lack the granularity to capture the complexity of brain network dysfunction in MDD.

Purpose of the Study:

• To develop a predictive model for MDD symptom severity using machine learning.
• To leverage effective connectivity (EC) derived from resting-state functional magnetic resonance imaging (rs-fMRI) data.
• To compare the predictive power of EC against traditional functional connectivity (FC) methods.

Main Methods:

• Utilized large-scale rs-fMRI data and Hamilton Depression Rating Scale (HAMD) scores from the REST-meta-MDD dataset.
• Computed brain EC features using Granger causality analysis and symbolic path coefficients.
• Constructed and evaluated machine learning models, including support vector regression, to predict HAMD scores.

Main Results:

• The Dosenbach brain atlas yielded the best predictive performance among tested atlases.
• EC-based models significantly outperformed FC models in predicting HAMD scores (r=0.81, p<0.001).
• Support vector regression demonstrated superior performance in the machine learning models.

Conclusions:

• Brain network EC features effectively predict HAMD scores in MDD patients.
• The identified EC network serves as a potential biomarker for MDD symptom severity.
• This approach offers clinically significant insights for early MDD diagnosis and personalized interventions.