Discrimination of generalized anxiety and major depressive disorders by machine learning of resting-state functional...

To address the clinical difficulty of differentiating Generalized Anxiety Disorder (GAD) from Major Depressive Disorder (MDD), this study aimed to create a machine learning framework with strong interpretability, leveraging a comprehensive suite of resting-state functional MRI features to enhance potential relevance.

A comprehensive set of functional measures, encompassing both local neural activity and global functional connectivity, was extracted as neuroimaging features from resting-state fMRI data. These data were obtained from 91 patients with Generalized Anxiety Disorder (GAD), 94 with Major Depressive Disorder (MDD), and 71 healthy controls (HCs). Five machine learning algorithms were used for classification, with performance estimated within a rigorous nested cross-validation framework. Then, partial correlations were conducted to assess the associations between top-ranked contributive neuroimaging features and symptom severity.

For the discrimination tasks, the machine learning models achieved area under the curve (AUC) values of 0.783 (GAD vs. MDD), 0.824 (GAD vs. HCs), and 0.867 (MDD vs. HCs), demonstrating moderate classification performance. Feature importance analysis revealed that precuneus function served as a prominently differential neurobiological signature and it showed opposing relationships with the severity of anxiety (positive correlation) and depression (negative correlation) across different diagnostic categories.

These findings could guide the creation of a computational framework based on neuroimaging to effectively differentiate GAD from MDD. This is underscored by the pivotal role that the precuneus appears to play in the neurobiological processes associated with symptoms of both disorders.

This is a non-interventional study with a control group. Clinical trial registration is not applicable.