Parkinson's disease (PD) exhibits marked clinical and neurobiological heterogeneity, complicating its prognosis and therapeutic development. We aimed to identify imaging-defined PD subtypes by integrating diffusion and structural MRI features, using a data-driven clustering approach. Baseline neuroimaging data from 103 de novo PD participants and 40 normal controls (NC) were obtained from the Parkinson's Progression Markers Initiative (PPMI). 209 imaging features were extracted, including fractional anisotropy (FA) and mean diffusivity (MD) from diffusion MRI and cortical and subcortical volumes from structural MRI. Imaging features were adjusted for demographic, anatomical, and scanner-related confounders using a normative, interaction-aware regression model. Hierarchical clustering with Ward's linkage was applied to the principal components of the adjusted features. The optimal cluster number was determined using Silhouette and Calinski–Harabasz indices, and cluster stability was evaluated by cross-algorithm validation and bootstrapping. Two imaging-defined PD subtypes were identified. Subtype A exhibited widespread elevations in FA and reductions in MD across multiple white matter tracts, with no significant volumetric atrophy, higher Symbol Digit Modalities Test performance, and an earlier age at onset. Subtype B showed widespread MD increases, regionally specific FA reductions, trend-level cortical volumetric reductions, ventricular enlargement, and lower cross-sectional cognitive performance. Over the 48-month follow-up, a linear mixed-effects model demonstrated divergent cognitive trajectories: Subtype A showed stable cognitive scores, whereas Subtype B showed significant cognitive decline. The two subtypes showed divergent cognitive trajectories in the longitudinal analysis. These findings identify two imaging-defined structural phenotypes in early de novo PD with divergent cognitive trajectories, providing a multimodal imaging framework for understanding PD heterogeneity at the earliest stages of the disease.
Arabi et al. (Mon,) studied this question.