Video-Based Hand Pose Estimation for Remote Assessment of Bradykinesia in Parkinson’s Disease

Characterizing Disease Progression in Parkinson’s Disease from Videos of the Finger Tapping Test

“…Some patients also demonstrate increased movement variability and a progressive reduction in movement amplitude during the repetitive tapping movement (sequence effect) [ 32 ], [ 33 ]. Several studies have shown that is possible to quantify kinematic features related bradykinesia such as amplitude, speed, and rate from videos of the Finger Tapping videos using ML algorithms [ 8 ], [ 10 ], [ 11 ], [ 15 ], [ 16 ], [ 17 ], [ 18 ], [ 19 ], [ 34 ], [ 35 ]. In this study, we showed that other, non-traditional kinematic features such as opening and closing movement speed, amplitude decay, and measures related to movement and timing variability could also be quantified from videos.…”

“…MDS-UPDRS III videos were manually processed to identify the start and end time of the Finger Tapping test for the right and left hands. Finger Tapping videos were processed using our custom pipeline [ 10 ]. As Fig.…”

“…The landmarks consist of the hand’s base and four points on each finger, including the fingertips and joints. We have previously validated the accuracy of this algorithm in persons with PD [ 10 ]. Using the landmarks, we calculated the angular separation between the thumb and index finger by computing the angle between two vectors formed by connecting the finger’s tips with the hand’s base.…”

“…Recently, there has been a growing interest in establishing objective digital movement markers of PD based on video recordings of the MDS-UPDRS III using machine learning (ML) algorithms for pose estimation. In particular, the Finger Tapping test, a component of the MDS-UPDRS III used to asses upper limbs bradykinesia, has gained significant attention [ 8 ], [ 9 ], [ 10 ], [ 11 ]. Several research groups, including ours, have developed custom processing pipelines to extract objective movement features from videos of the Finger Tapping test using ML algorithms.…”

“…Several research groups, including ours, have developed custom processing pipelines to extract objective movement features from videos of the Finger Tapping test using ML algorithms. Such pipelines have been used to detect PD from videos with accuracy higher than 80% [ 11 ], predict disease severity [ 8 ], [ 9 ], and quantify the effect of therapies such as deep brain stimulation [ 10 ]. Assessing PD motor symptoms severity through objective kinematic features derived from videos of the MDS-UPDRS III benefits from the scale’s established clinical validity and widespread acceptance in the medical community while avoiding the limitations of the 5-point Likert scale.…”