Identification of Biological Subtypes of Friedreich Ataxia with Structural MRI-based Machine Learning

Background Friedreich ataxia (FRDA) is an inherited, progressive neurodegenerative disease. Interindividual heterogeneity in the rate and phenotypic profile of disease progression indicates a biologic variability in the pattern and spatial evolution of underlying changes, but the occurrence of possible FRDA subgroups, which could aid in clinical trial design and treatment, are still unknown. Purpose To obtain a structural MRI-based stratification of participants with FRDA using the Subtype and Stage Inference (SuStaIn) algorithm and determine whether these subgroups are biologically meaningful and clinically relevant. Materials and Methods This multicenter secondary analysis of prospectively acquired data included structural MRI and clinical-demographic data from participants from the ENIGMA-Ataxia working group. MRI biomarkers were analyzed using the SuStaIn algorithm to identify subgroups with distinct patterns and disease stages. The clinical and genetic relevance of these subgroups were assessed within a linear model framework. Results This study included 565 participants (mean age, 32 years ± 13.1 [SD]; 286 women; 275 participants with FRDA and 290 healthy controls). SuStaIn identified three subtypes: <i>(a)</i> a classical subtype (66.5% [183 of 275 participants]), characterized by an ascending gradient of damage from brainstem to cerebellar cortex to cerebrum; <i>(b)</i> an early cerebral subtype (25.8% [71 of 275 participants]) with cerebral atrophy preceding the involvement of cerebellar cortex; and <i>(c)</i> and an early cerebellar subtype (7.64% [21 of 275 participants]) showing cerebellar lobule atrophy before upper brainstem or cerebral involvement. More advanced disease stages (MRI-based SuStaIn stages) correlated with greater symptom duration (unstandardized coefficient <i>B</i> = 0.422, standard error = 0.065, <i>P</i> < .001) and severity (<i>B =</i> 1.404, standard error = 0.201, <i>P</i> < .001), and these relationships were moderated by subtype, with biologic stage progression in the early cerebral subtype mapping less strongly to clinical variables relative to the others (interaction term early cerebral subtype × stage: <i>B</i> = -0.925, standard error = 0.410, <i>P</i> = .02). Conclusion Using the SuStaIn algorithm, three distinct structural MRI-based subtypes of FRDA were identified, with different patterns of brain degeneration and associations with clinical severity. © RSNA, 2026 <i>Supplemental material is available for this article.</i>