Rat Tongue Kinematics with Markerless XROMM Using DeepLabCut and Marker Masking

X-ray Reconstruction of Moving Morphology (XROMM) enables 3D tracking of implanted radiopaque markers in animals to study biomechanics. In recent years, DeepLabCut (DLC) has emerged as an important tool for markerless pose estimation in areas of the biological sciences such as neuroscience or behavioral research. While the XROMM markers can be detected and tracked using traditional image processing, DLC has been proven to improve XMALab-based workflows. However, those methods are prone to the exact marker positions and are not robust concerning missing markers.We propose to use XMALab results to reconstruct camera calibrations for the prediction of DLC-guided 3D XROMM markers in videofluoroscopic data. Additionally, we propose a marker masking augmentation to enable tracking with incomplete marker sets and demonstrate that XROMM pose estimation is possible with subsets of visible beads and even without any visible markers without sacrificing accuracy. we further show an implementation of this method to extract tongue kinematic patterns, resulting in a novel insight into rodent tongue behavior during natural feeding behavior.