Using computer analysis to help assess Parkinson’s symptoms

Idiopathic Parkinson's disease (PD) is a neurodegenerative disease that progressively causes both motor and non-motor symptoms. As the second most common neurodegenerative disease and most common movement disorder, it affects over 8.5 million people worldwide and 13,000 people in Hong Kong. The most classical symptoms of PD are resting tremors, rigidity of the muscles, bradykinesia (slowing of movement), and gait difficulty. Other symptoms include sleep disorders, psychiatric symptoms, cognitive impairment, and autonomic dysfunction. Its pathophysiology is marked by the loss of dopaminergic neurons and the accumulation of aggregates called Lewy bodies. The severity of PD-related motor symptoms is usually semi-quantitatively ("normal", "slight", "mild", "moderate", and "severe") evaluated by expert physicians and physiotherapists according to the Movement Disorder Society-sponsored revision of the Unified Parkinson's Disease Rating Scale Part III (MDS-UPDRS III). However, the MDS-UPDRS III is semiquantitative and subjective, which might mask mild treatment effects or even provide false-positive results. Moreover, it takes significant time and effort for assessment with expected inter-observer variations. To address these issues, various artificial intelligence (AI) technologies and telemedicine approaches have been investigated for patient evaluation. However, previous studies did not incorporate items assessing rigidity and postural stability, which require physical contact as per the MDS-UPDRS III instructions. Zhu et al. explored a motor symptom machine-rating system for the complete MDS-UPDRS III. Nevertheless, they employed a depth camera and conducted the tests within a strictly controlled ideal laboratory environment. For the widespread implementation of AI-assisted rating, the RGB camera is a more accessible alternative.

This is a single-center, prospective, observational study designed to develop and validate an AI-based MDS-UPDRS III assessment system using RGB camera data. Participants will be recruited from Queen Elizabeth Hospital's neurology outpatient clinic. Each subject will undergo standard MDS-UPDRS III evaluation by a certified clinician or physiotherapist, alongside synchronized RGB-D video recording. The videos will be processed through a deep learning pipeline trained to estimate the MDS-UPDRS III scores. Blinded evaluations will be performed to compare AI-generated scores with ground truth clinician ratings. Statistical analysis will include inter-rater agreement metrics (e.g., ICC, Cohen's kappa), sensitivity to change, and subgroup analyses.