Understanding cognitive impairment in multiple sclerosis using structural and functional network measures and classification algorithms

Le résumé fourni par la source

Abstract Cognitive impairment (CI) is common in multiple sclerosis (MS), yet the network-level mechanisms underlying CI remain poorly understood. This study aimed to clarify how the joint organization of structural and functional brain networks contributes to CI in MS, and how network-derived features change when clinical information and MRI-derived measures are added. We analyzed neuropsychological and multimodal MRI data from the Amsterdam MS cohort (N = 330) and assessed generalizability externally (N = 27). Graph-theoretical measures of brain connectivity were extracted from diffusion-weighted and resting-state functional MRI. Machine learning models classified cognitively impaired versus preserved patients. Classification performance was quantified using the area under the receiver operating characteristic curve (AUROC), sensitivity, and specificity. Feature importance was evaluated using Shapley values. Network-derived features improved discrimination over demographic information alone (AUROC = 0.77; p < 0.001). Adding MRI-derived measures significantly improved performance (AUROC = 0.81; p < 0.001), whereas clinical variables added little additional information (AUROC = 0.79; p = 0.65). External validation demonstrated good discriminative ability (AUROC = 0.76) but low sensitivity. Structural connectivity within the dorsal attention network emerged as an informative feature. These findings suggest that network-derived features help classify cognitive status in MS and remain relevant alongside conventional features. Structural connectivity, especially within the dorsal attention network, may provide insight into systems-level correlates of CI in MS.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.