Less is more: downsampling x-ray images improves pose estimation accuracy

Biplanar videoradiography (BVR) is a gold-standard technique for quantifying<i>in vivo</i>bone motion, yet the influence of x-ray image resolution on pose estimation accuracy remains unexplored. This study investigates how downsampling x-ray images impacts model-based pose estimation, using high-speed BVR data from a participant with implanted tantalum beads. Images were downsampled from 2048 × 2048 to 512 × 512 using bicubic and nearest-neighbour interpolation. Across multiple bones and varying perturbation levels, downsampling significantly reduced rotational and translational errors when compared to full-resolution images for both interpolation results. Bicubic interpolation led to slightly improved pose accuracy for certain bones, demonstrating enhanced edge clarity that benefits the optimisation algorithm. Pose estimates for full-resolution images exhibited more outliers and greater variability for all the bones investigated. These findings highlight that downsampling images improves pose estimation accuracy even for challenging anatomical areas such as the ankle. We recommend bicubic downsampling to 512 × 512 pixels as a best practice for BVR tracking of the ankle complex, when using both automated optimisation and manual workflows.