Disentangling neurodegeneration from ageing in multiple sclerosis: the brain-predicted disease duration gap

Abstract Disentangling brain ageing from disease-related neurodegeneration in patients with multiple sclerosis (PwMS) is increasingly topical. The brain-age paradigm offers a window into this problem but may miss disease-specific effects. Here, we statistically modelled disease duration (DD) in PwMS as a function of brain MRI scans and evaluated whether the brain-predicted DD gap (i.e., the difference between predicted and actual duration) could complement the brain-age gap as a DD-adjusted global measure of multiple sclerosis-specific brain damage. In this retrospective study, we used 3D T1-weighted brain MRI scans of PwMS (i) from a large multicentric dataset (n = 4,392) for age and DD modelling, and (ii) from a monocentric longitudinal cohort of patients with early multiple sclerosis (n = 252 patients, 749 sessions) for clinical validation. We trained and tested a deep learning model based on a 3D DenseNet architecture to predict DD from minimally pre-processed brain MRI scans, while age predictions were obtained with the previously validated DeepBrainNet algorithm. Model predictions were scrutinised to assess the influence of lesions and brain volumes, while the DD gap metric was biologically and clinically validated within a linear model framework assessing its relationship with brain-age gap values and with physical disability measured with the Expanded Disability Status Scale (EDSS). Our model predicted DD better than chance (mean absolute error = 5.63 years, R 2 = 0.34) and was nearly orthogonal to the brain-age model, as suggested by the very weak correlation between DD gap and brain-age gap values ( r = 0.06). DD predictions were influenced by spatially distributed variations in brain volume, and, unlike brain-predicted age, were sensitive to the presence of lesions (mean difference between unfilled and filled scans: 0.55 ± 0.57 years, p < 0.001). The DD gap metric significantly explained EDSS scores (β = 0.060, p < 0.001), adding to brain-age gap values (ΔR 2 = 0.012, p < 0.001). Longitudinally, increasing annualised DD gap was associated with greater annualised EDSS changes ( r = 0.50, p < 0.001), with a significant incremental contribution in explaining disability worsening compared to changes of the brain-age gap alone (ΔR 2 = 0.064, p < 0.001). The brain-predicted DD gap metric appears to be sensitive to multiple sclerosis-related lesions and brain atrophy, adding to the brain-age paradigm in explaining physical disability both cross-sectionally and longitudinally. Potentially, it may be used as a multiple sclerosis-specific biomarker of disease severity and progression.