Data-driven, cross-sectional image-based subtyping and staging of brain volumetric changes in Parkinson's disease

Abstract

Background: Several subtyping methods have been proposed to characterize Parkinson's disease (PD) progression, yet the trajectory of subcortical and cortical neurodegeneration and its clinical implications remain unclear.

Objectives: We aimed to conduct a strictly image-based, data-driven classification of PD progression through Subtype and Stage Inference (SuStaIn) algorithm.

Methods: Brain volumetric data from 565 patients with PD and 150 propensity-matched healthy controls were analyzed. 16 regions of interest, including 9 cortical and 7 deep grey matter structures, were segmented from T1-weighted magnetic resonance images. Clinical data, including REM sleep behavior disorder (RBD), levodopa equivalent daily dose (LEDD), and motor complications were collected. SuStaIn was trained and tested using a 10-folds cross-validation and identified two distinct PD progression subtypes, which were compared for differences in clinical and radiological characteristics.

Results: We found two distinct neurodegenerative trajectories: deep grey matter (DG)-first and cortex (CO)-first. The CO-first subtype had a higher prevalence of RBD (p = 0.009) and levodopa-induced dyskinesia (p = 0.024) than the DG-first subtype. Disease progression was faster in the CO-first subtype (0.203 year/stage, LEDD increase 59.3 mg/year), than in the DG-first subtype (0.081 year/stage, LEDD increase 45.1 mg/year, respectively). Regardless of the subtypes, the sensorimotor and auditory cortices were the earliest affected cortical regions, while the amygdala was the first affected subcortically. A subset of participants (n = 186) showed no significant atrophy progression.

Conclusions: Our findings support the existence of two distinct subtypes of PD progression based on neuroimaging data. Longitudinal studies are warranted to track their evolution.