Cerebellar Normative Modeling Identifies Neuroanatomical Biotypes Predicting dTMS Response in Spinocerebellar Ataxia Type 3

BackgroundSpinocerebellar ataxia type 3 (SCA3) presents with significant clinical heterogeneity. Traditional case-control neuroimaging, based on group means, obscures inter-individual anatomical variability, hindering the identification of stratification biomarkers for interventions like Transcranial Magnetic Stimulation (TMS). MethodsTo quantify individual neuroanatomical deviations, we constructed a cerebellar normative model using a multi-center dataset of 2,071 healthy controls with 2,549 MRI scans. Gray matter volume deviations (Z-scores) were mapped across 27 cerebellar lobules in 114 genetically confirmed SCA3 patients, and unsupervised clustering was applied to identify neuroanatomical biotypes. Clinical relevance was assessed by associating biotypes with ataxia severity and deep TMS (dTMS) outcomes in a longitudinal subset . ResultsWe identified two distinct biotypes: Biotype 1 exhibited relative structural preservation (positive deviations) predominantly in the posterior cerebellum (lobules VIIB, VIIIA), whereas Biotype 2 was characterized by extensive atrophy (negative deviations) centered on the anterior motor cerebellum (lobules I-VI). Clinically, Biotype 2 patients presented with significantly more severe baseline ataxia. However, regarding treatment response, an inverse relationship was observed: Biotype 2 patients demonstrated significantly greater symptomatic improvement following dTMS compared to Biotype 1. To further identify the optimal neuromodulatory strategy for each biotype, we compared the therapeutic efficacy of repetitive TMS (rTMS) and dTMS. While both biotypes showed clinical improvement following rTMS, Biotype 1 exhibited a superior therapeutic response to rTMS relative to dTMS. Furthermore, feature weight analysis identified atrophy of the right lobule VIIB as a critical predictor of clinical severity in Biotype 2. ConclusionThis study demonstrates that normative modeling can decode SCA3 heterogeneity. The identification of these biotypes reveals a dissociation between baseline structural integrity and neuromodulatory responsiveness, suggesting that patients with severe anterior cerebellar atrophy may, counterintuitively, derive greater therapeutic benefit from dTMS. Furthermore, by comparing the therapeutic efficacy of rTMS and dTMS, we further clarified biotype-specific treatment responses. These findings support the use of individualized neuroanatomical mapping for patient stratification in precision medicine.