Investigating functional brain networks in multiple sclerosis patients: a graph theory approach for evaluating...

This study introduces a graph theory-based machine learning framework to analyze functional brain networks in Multiple Sclerosis (MS) patients during working memory tasks. Using fMRI data from 22 participants (8 MS patients/14 controls) performing a Persian n-back task, we constructed functional connectivity networks with an automated anatomical labeling atlas (116 regions). Novel methodological contributions include: a proportional thresholding approach (5-80% connectivity strength) to optimize network analysis, and extraction of six graph-theoretic features (degree, clustering coefficient, betweenness/eigenvector/page-rank/k-coreness centrality) for classification. Machine learning models (SVM, decision trees, kNN) were trained on task-specific networks (1-back, 2-back, 3-back), with leave-one-subject-out validation. The 3-back task (high cognitive load) yielded superior classification (95.5% accuracy) using SVM with degree, k-coreness, and eigenvector centrality features, outperforming traditional resting-state approaches. Key innovations include identification of the cerebellum (1-back) and orbitofrontal cortex (3-back) as novel discriminative hubs, and demonstration that task difficulty modulates network separability. Results indicate MS patients exhibit impaired high-load connectivity patterns, with the proposed framework showing potential as an auxiliary diagnostic tool. The study provides both a technical blueprint for task-based network analysis and clinically relevant insights into MS-related cognitive impairment, bridging engineering methodologies with neurological applications.