Multimodal Remote Digital Phenotyping for Detecting and Tracking Early Parkinsonian Change in LRRK2 Carriers

Area of Science:

• Neuroscience
• Genetics
• Artificial Intelligence

Background:

• Identifying prodromal Parkinson's disease (PD) in individuals carrying LRRK2 mutations is crucial for early intervention but remains challenging.
• LRRK2 gene mutations are a significant genetic factor in Parkinson's disease, necessitating focused research on carriers.

Purpose of the Study:

• To develop and validate a remote, multimodal video analysis framework for early PD detection in LRRK2 carriers.
• To distinguish between non-manifest LRRK2 carriers and healthy controls, and to monitor early clinical changes indicative of PD onset.

Main Methods:

• Analysis of video data from 829 participants, including 158 LRRK2 carriers (36 with manifest PD, 122 non-manifest) and controls.
• Development of an AI model for classification and a continuous score ('PD Weigh-In') for monitoring disease progression.
• Validation of the AI model's accuracy and correlation of the 'PD Weigh-In' score with expert clinical ratings.

Main Results:

• The AI model achieved 92.9% accuracy in distinguishing non-manifest LRRK2 carriers from controls (AUROC 0.92, AUPRC 0.82).
• The 'PD Weigh-In' score effectively captured clinical decline in carriers who later developed PD.
• The score demonstrated strong correlation with expert assessments (Pearson r = 0.77, Spearman ρ = 0.79) in the LRRK2 test cohort.

Conclusions:

• A remote, AI-driven video analysis framework can effectively identify individuals at high risk for Parkinson's disease among LRRK2 carriers.
• This technology holds promise for early detection and continuous monitoring of prodromal Parkinson's disease, facilitating timely therapeutic strategies.