Abstract
BACKGROUND: Aerobic training is a promising rehabilitation strategy in multiple sclerosis (MS). However, individual responses to exercise vary. OBJECTIVES: To assess whether baseline demographic, clinical, and MRI features in MS patients predict response to aerobic or non-aerobic training. METHODS: Eighty-eight MS patients were randomized to 24-session aerobic (n = 43) or non-aerobic (n = 45) training over 2-3 months. Responders were defined by a ⩾ 21.6 meters improvement in the 6-minute walking test. Baseline assessments included global disability, fatigue, walking speed, peak oxygen consumption, and structural/functional MRI (lesion load, brain volumetrics, cortical thickness, white matter (WM) microstructural integrity, and resting-state functional connectivity). Random forest models identified baseline demographic, clinical, and MRI predictors of training response. RESULTS: Thirty-four MS patients (aerobic = 20, non-aerobic = 14) were responders. Predictors of treatment response were corticospinal tract (CST) and middle cerebellar peduncle (MCP) fractional anisotropy in the aerobic group and superior cerebellar peduncle (SCP) mean diffusivity in the non-aerobic group. SCP, MCP, and CST diffusivity metrics predicted training response in whole MS cohort. Repeated cross-validation confirmed identified predictors relevance (median area under the curve from 0.648 to 0.672). CONCLUSION: Microstructural integrity of motor-related WM tracts predicts training response in MS, highlighting the role of neuroimaging in identifying patients likely to benefit from rehabilitation.