Abstract
We developed VisionMD, an AI computer vision platform, analyzing over 1200 clinical videos of Parkinson's patients' hand movements across 13 years. This large-scale, markerless analysis identified three kinematic domains (speed, consistency, timing/scale) reliably improved by levodopa. Our method offers objective, quantitative motor assessment, reducing subjectivity and enhancing reproducibility compared to traditional scales.