Data Driven Subtyping Reveals Two Distinctive Patterns of Deep Gray Matter atrophy and Dopamine Availability in early Parkinson’s Disease

Motivation: Recent imaging evidence suggests that machine-learning technique may be useful to identify subtypes with distinct spatial patterns of Parkinson's disease. Goal(s): We investigated possible data-driven subtypes of PD. Approach: Using the Subtype and Stage Inference machine-learning technique, we categorized PD patients into subtypes by analyzing their progression patterns in deep gray matter atrophy and dopamine availability. Results: Subtype 1 displayed early dopamine availability reduction, severe cardiac denervation, mild cognitive dysfunction in the early stage, and rapid decline in motor and cognitive function in the later stage, whereas subtype 2 showed early brain atrophy, mild cardiac denervation, severe cognitive dysfunction in the early stage. Impact: Two subtypes displayed distinctive patterns in dopamine availability, deep gray matter volume, cognition, motor symptom, and cardiac denervation, supporting the body-first and brain-first concepts of PD through imaging-machine learning approach.