Abstract
AIMS: To characterize white matter geometric pathology in cerebellar subtype of multiple system atrophy (MSA-C) using director field analysis (DFA) and identify stage-specific biomarkers. METHODS: We analyzed single-shell diffusion MRI (b = 1000) in 31 MSA-C patients (15 early-, 16 late-stage) and 33 controls. DFA quantified axonal geometry (splay/bend/twist), complemented by fixel-based analysis (FBA) and brainstem volumetry. Group comparisons used threshold free cluster enhancement (TFCE) (p < 0.05 FWE-corrected). DFA-altered regions were correlated with clinical scores. AutoGluon evaluated classification performance using different feature sets. RESULTS: MSA-C exhibited distinct geometric degeneration patterns: cerebellar pathways showed reduced splay, bend, and twist (reflecting Wallerian degeneration), whereas brainstem tracts demonstrated dissociated geometry (increased splay/bend but decreased twist). Brainstem twist reduction strongly differentiated early- and late-stage MSA-C (AUC = 0.95). Clinically, middle cerebellar peduncle bend correlated with motor progression (UMSARS-II: r = 0.48), while cerebellar splay reduction predicted ataxia severity (SARA: r = -0.43). CONCLUSION: DFA captures circuit-specific white matter pathology in MSA-C, with brainstem twist emerging as a novel biomarker associated with disease stage. The integration of geometric metrics with automated machine learning provides a robust framework for early diagnosis and disease staging, highlighting distinct neurodegenerative mechanisms in cerebellar versus brainstem pathways.