Validating marker-less pose estimation with 3D x-ray radiography

Abstract

To reveal the neurophysiological underpinnings of natural movement, neural recordings must be paired with accurate tracking of limbs and postures. Here we validate the accuracy of DeepLabCut (DLC) by comparing it to a 3D x-ray video radiography system that tracks markers placed under the skin (XROMM). We record behavioral data simultaneously with XROMM and high-speed video for DLC as marmosets engage in naturalistic foraging and reconstruct three-dimensional kinematics in a shared coordinate system. We find that DLC tracks position and velocity of 12 markers on the forelimb and torso with low median error (0.272 cm and 1.76 cm/s, respectively) corresponding to 2.9% of the full range of marker positions and 5.9% of the range of speeds. For studies that can accept this relatively small degree of error, DLC and similar marker-less pose estimation tools enable the study of more naturalistic, unconstrained behaviors in many fields including non-human primate motor control. Summary StatementWe validate the accuracy of DeepLabCut, a widely used marker-less pose estimation tool, using a marker-based 3D x-ray radiography system (XROMM).