Radiomics-based classification and inference of subtypes and stages in social anxiety disorder using resting-state functional images.

BACKGROUND

This study aimed to leverage advanced radiomics analysis of resting-state functional magnetic resonance imaging (rs-fMRI) data to investigate the potential of radiomics in distinguishing patients with social anxiety disorder (SAD) from healthy controls and identifying distinct subtypes within patients.

METHODS

We analyzed the rs-fMRI data from 147 participants (78 controls, 69 patients) using three rs-fMRI metrics: regional homogeneity, fractional amplitude of low-frequency fluctuations, and degree centrality. From each of these metrics, we extracted 91 radiomics and mean signals from the amygdala, hippocampus, insula, and medial/ventromedial prefrontal cortex (mPFC/vmPFC). We employed machine learning algorithms for classification and utilized Subtype and Stage Inference (SuStaIn) model to identify disease subtypes and symptom progression.

RESULTS

Classification using radiomics from individual regions, particularly the left amygdala (accuracy: 84.3%), right hippocampus (74.2%), and mPFC (74.1%), significantly outperformed classification using mean signals from all regions (52.2%). Furthermore, the right hippocampal-based SuStaIn model revealed two distinct subtypes of SAD, social anxiety-led and general anxiety-led, with the former demonstrating more severe comorbid symptoms and poorer prognosis.

CONCLUSIONS

Radiomics features from rs-fMRI effectively classified patients with SAD and revealed clinically meaningful subtypes through SuStaIn modeling. These findings demonstrate the value of quantitative imaging approaches that capture subtle functional patterns and underscore the potential of disease-progression modeling for understanding heterogeneity in SAD.