Screening for Parkinson's disease using "computer vision"

Area of Science:

• Neurology
• Biomedical Engineering
• Computer Science

Background:

• Bradykinesia is a key indicator for Parkinson's disease (PD) diagnosis.
• Traditional finger-tapping tests rely on subjective physician assessments.
• Computer vision presents a non-contact, objective, and cost-effective diagnostic approach.

Purpose of the Study:

• To detect Parkinson's disease (PD) by identifying bradykinesia.
• Utilize computer vision analysis of the finger-tapping test.
• Apply machine learning models for automated diagnosis using both hands' data.

Main Methods:

• Recruited 100 participants (PD patients and healthy controls).
• Analyzed 10-second finger-tapping movements recorded via smartphone using Google MediaPipe Hands.
• Trained six machine learning models with a nested cross-validation framework.

Main Results:

• PD patients showed significantly greater differences in tapping scores between hands compared to controls (p=0.001).
• Tapping amplitude variations and decremental parameters significantly differed in PD patients.
• Machine learning models, particularly support vector machines, accurately classified PD based on tapping features.

Conclusions:

• Computer vision effectively detects bradykinesia from finger-tapping tests.
• This method provides an objective and accurate approach for Parkinson's disease diagnosis.
• Simultaneous tapping analysis of both hands enhances diagnostic capability.