Computer Vision Analysis for Objective Motor Assessment in Parkinson's Disease: A Retrospective Study

Area of Science:

• Neurology
• Biomedical Engineering
• Medical Imaging

Background:

• The Movement Disorder Society-Unified Parkinson's Disease Rating Scale-Part III (MDS-UPDRS-III) is subjective and lacks sensitivity for early Parkinson's disease (PD) detection.
• Computer vision (CV) offers a method for objective kinematic analysis from routine videos, potentially improving PD motor assessment.

Purpose of the Study:

• To identify CV-derived finger-tapping features that differentiate PD patients from healthy controls (HC).
• To quantify the relationship between these CV features and clinical measures, including MDS-UPDRS-III and DAT-SPECT.

Main Methods:

• Retrospective analysis of finger-tapping videos from PD patients and HC.
• Utilized a Mediapipe-based pipeline to quantify tapping velocity, amplitude changes, and amplitude/rhythm variability.
• Assessed diagnostic performance using ROC AUC and correlations with clinical scores via Spearman analysis.

Main Results:

• Amplitude and rhythm variability demonstrated high discriminatory capacity between PD and HC (AUCs > 0.83).
• CV-derived amplitude variability and decrement correlated with MDS-UPDRS-III scores and finger-tapping severity (item 3.4).
• Tapping velocity negatively correlated with MDS-UPDRS-III, while amplitude variability correlated with disease duration.

Conclusions:

• CV-based analysis of finger-tapping provides objective kinematic measures.
• These objective measures effectively discriminate PD from HC and correlate with clinical motor severity and disease progression.