Fig. 1: A flowchart of characterization of clinical and imaging heterogeneity in Parkinson’s disease.

a A total of 567 individuals (347 in discovery dataset and 220 in validation dataset) were enrolled in this study; five clinical variables and 6 imaging variables were used for model construction; these imaging variables were extracted from 6 brain structures including substantia nigra, locus coeruleus, amygdala, hippocampus, entorhinal cortex, and basal forebrain by using neuromelanin-sensitive imaging and diffusion tensor imaging. b These variables were normalized relative to healthy control subjects using z-score; the SuStaIn model was conducted to identify distinct patterns of progression (progression trajectories of clinical and neurodegenerative events) using these variables and to cluster patients into distinct patterns (subtypes) and stages. c The differences of clinical/imaging variables, levodopa response, and longitudinal progression were assessed between subtypes; the consistency between the inferred SuStaIn stage and the disease severity measured by the UPDRS, PDQ-39, and other measures was evaluated. The box plot and line chart were used for visualization purposes only. Box plots with center line indicating median, bounds of boxes showing upper and lower quartile, whiskers illustrating 1.5 * interquartile range, and dots representing the distribution of raw data (minima and maxima are included). MRI magnetic resonance imaging, SuStaIn Subtype and Stage Inference, REM rapid eye movement. UPDRS Unified Parkinson’s Disease Rating Scale, PDQ-39 Parkinson’s Disease Questionnaire-39 items.