A computer vision framework for finger-tapping evaluation in Parkinson's disease.

Area of Science:

• Biomedical Engineering
• Movement Disorder Analysis
• Computer Vision Applications

Background:

• The rapid finger-tapping test (RFT) is crucial for evaluating movement disorders like Parkinson's disease (PD).
• Current clinical assessment of RFT relies on subjective, coarse visual judgment, limiting precision.
• Objective quantification methods are needed for accurate PD symptom tracking.

Purpose of the Study:

• To introduce a novel computer-vision (CV) method for objective quantification of RFT symptoms.
• To utilize facial features for normalizing tapping amplitude and camera-subject distance variations.
• To develop a CV algorithm for detailed motion analysis of index-finger tapping.

Main Methods:

• Collected 387 RFT videos from 13 advanced PD patients and 84 from 6 healthy controls.
• Clinicians rated PD patient tapping severity using the UPDRS-FT scale (0-3).
• A CV algorithm tracked index-finger motion, extracting features for speed, amplitude, rhythm, and fatigue, then used in a support vector machine classifier.

Main Results:

• A novel rhythm feature, 'cross-correlation between normalized peaks,' strongly correlated with clinical ratings (Guttman correlation μ2=-0.80).
• Support vector machine classification achieved 88% accuracy in categorizing PD patient symptom severity.
• The system discriminated between PD patients and healthy controls with 95% accuracy.

Conclusions:

• The CV-based RFT analysis is feasible for objective PD symptom assessment.
• This approach is suitable for remote, home-based PD monitoring.
• It offers advantages over existing sensor-based technologies for tapping symptom evaluation.