Background: Multiple Sclerosis (MS) is a disease characterized by inflammation, demyelination, and/or axonal loss that disrupts white matter pathways that constitute the brain's structural connectivity network. Individual disease burden and disability in patients with MS (pwMS) varies widely across the population, possibly due to heterogeneity of lesion location, size and subsequent disruption of the structural connectome. Chronic active MS lesions, which have a hyperintense rim (rim+) on Quantitative Susceptibility Mapping (QSM) and a rim of iron-laden inflammatory cells, have been shown to be particularly detrimental to tissue concentration causing greater myelin damage compared to chronic silent MS lesions. How these rim+ lesions differentially impact structural connectivity and subsequently influence disability has not yet been explored.
Objective: We characterize differences in the spatial location and structural disconnectivity patterns of rim+ lesions compared to rim- lesions. We test the hypothesis that rim+ lesions' disruption to the structural connectome are more predictive of disability compared to rim- lesions' disruption to the structural connectome. Finally, we quantify the most important regional structural connectome disruptions for disability prediction in pwMS.
Methods: Ninety-six pwMS were included in our study (age: 40.25 ± 10.14, 67% female). Disability was assessed using Extended Disability Status Score (EDSS); thirty-seven pwMS had disability (EDSS ≥ 2). Regional structural disconnectivity patterns due to rim- and rim+ lesions were estimated using the Network Modification (NeMo) Tool. For each gray matter region, the NeMo Tool calculates a Change in Connectivity (ChaCo) score, i.e. the percent of connecting streamlines also passing through a lesion. Adaptive Boosting (ADA) classifiers were constructed based on demographics and the two sets of ChaCo scores (from rim+ and rim- lesions); performance was compared across the two models using the area under ROC curve (AUC). Finally, the importance of structural disconnectivity in each brain region in the disability prediction models was determined.
Results: Rim+ lesions were much larger and tended to be more periventricular than rim- lesions. The model based on rim+ lesion structural disconnectivity measures had better disability classification performance (AUC = 0.67) than the model based on rim- lesion structural disconnectivity (AUC = 0.63). Structural disconnectivity, from both rim+ and rim- lesions, in the left thalamus and left cerebellum were most important for classifying pwMS into disability categories.
Conclusion: Our findings suggest that, independent of the evidence that they have more damaging pathology, rim+ lesions also may be more influential on disability through their disruptions to the structural connectome. Furthermore, lesions of any type in the left cerebellum and left thalamus were especially important in classifying disability in pwMS. This analysis provides a deeper understanding of how lesion location/size and resulting disruption to the structural connectome can contribute to MS-related disability.
MoDL-QSM: Model-based Deep Learning for Quantitative Susceptibility Mapping
