Remote AI Screening for Parkinson's Disease: A Multimodal, Cross-Setting Validation Study

Area of Science:

• Neurology
• Artificial Intelligence
• Digital Health

Background:

• Parkinson's disease (PD) diagnosis often relies on specialized clinical assessments, which can be inaccessible.
• Remote screening tools are needed to improve early detection and accessibility.

Purpose of the Study:

• To evaluate the performance and usability of PARK, a web-based AI tool for remote Parkinson's disease screening.
• To assess PARK's accuracy and generalizability across diverse populations and settings.

Main Methods:

• Utilized video and audio recordings of speech, facial expression, and motor tasks from 1,865 participants.
• Trained and validated the AI model on independent test sets (n=389) and compared performance with movement disorder specialists.
• Conducted usability studies to assess participant satisfaction and preference.

Main Results:

• PARK achieved high accuracy (80.2%-80.6%) and AUROC (0.85-0.87) in classifying individuals with and without PD.
• Demonstrated strong agreement with clinical specialists and generalized well across demographic subgroups.
• Usability studies reported high participant satisfaction and preference for remote screening.

Conclusions:

• PARK shows potential as an accessible, scalable, and clinically aligned tool for remote Parkinson's disease identification.
• The tool can support early detection, especially where traditional healthcare access is limited.
• AI-powered remote screening offers a promising approach to overcome healthcare access barriers for neurological conditions.