Tracking Hand Position During the Purdue Pegboard Test using Markerless Pose-Estimation

Paul Marinis

Background: The Purdue Pegboard Test is a popular and reliable assessment of manual dexterity but does not assess hand kinematics, such as the trajectory or speed of the hand during difference phases of prehensile tasks. Assessment of hand kinematics may help clinical researchers better characterise manual dexterity impairments. Markerless motion capture is an emerging technology that could be used to track hand kinematics, however, it’s unclear whether it can do so accurately enough for clinical or research purposes. Objective: Establish the technical validity of using single camera markerless pose estimation to track hand movement during the Purdue Pegboard Test. Methods: Using a single-camera approach, we recorded participant’s hands during the Purdue Pegboard Test and trained pose estimation models with DeepLabCut software to track the metacarpophalangeal joint of the second phalange. Technical validity was assessed by comparing spatial agreement between manually labelled frames and those estimated by the trained model. RESULTS: Eleven adults who were free from visual, auditory, cognitive and motor impairments affecting hand function participated. Consistent with our hypothesis of a test error of below 5 mm, median test errors were 1.32 mm for models trained to 60,000 iterations and 0.86 mm for models trained to 90,000 iterations with corrected outliers. Secondary analysis showed excellent retest reliability of manual labelling of the metacarpophalangeal joint (median = 1.56 mm; mean = 1.70 mm, SD = 0.89 mm; 95% CI = 1.53 mm to 1.87 mm). Darker skin tone was moderately correlated, albeit not significantly with, poorer reliability of manual labelling but not model test error. CONCLUSIONS: Single-camera markerless pose estimation is a valid method for tracking the position of the hand during the Purdue Pegboard Test. Markerless pose estimation using a single camera has potential to assess manual dexterity impairments observed in developmental, ageing, and neurological populations. However, further experimental validation is required, such as testing whether hand kinematics derived from markerless pose estimation are sensitive to increasing task and cognitive demands.

Tracking Hand Position During the Purdue Pegboard Test using Markerless Pose-Estimation

Files and links (1)

Abstract

Details

Metrics