Abstract
Parkinson's disease (PD) exhibits clinical and neuropathological heterogeneity, potentially driven by distinct spatiotemporal neurodegenerative patterns. This study utilizes a connectivity-based MRI epicenter model combined with unsupervised clustering to identify unique degenerative epicenters in PD. Analyzing cross-sectional multi-modal MRI data from 278 PD patients and 177 healthy controls, this work identifies two distinct neurodegenerative epicenter patterns. Subtype 1 exhibits epicenters predominantly in cerebellar and midbrain regions associated with severe motor symptoms and rapid progression. Subtype 2 shows epicenters primarily in cortical and striatal regions with milder progression. These patterns are validated in an independent cohort of 66 PD patients and shows consistency in longitudinal follow-up. Additionally, a predictive model incorporating epicenter traits and structural connectivity properties is developed, accurately forecasting individualized neurodegenerative progression. Spatial correlation analyses further reveal overlapping epicenter distributions between PD subtype 1 and other movement disorders, including essential tremor and multiple system atrophy, suggesting potential shared pathological mechanisms. These results delineate PD heterogeneity through distinct epicenter-driven neurodegenerative trajectories, bridging the gap between neuroanatomical spread patterns and clinical variability. This novel framework not only enhances the understanding of PD's neuropathological complexity but also advances personalized prognosis and highlights connectivity-based epicenters as promising biomarkers for PD subtyping and therapeutic targeting.