Automatic registration of generic bone models for X-ray-based biomechanics: Application to 3D rat hindlimb kinematics...

Biplanar X-ray fluoroscopy offers sub-millimeter insight into skeletal motion, yet prevailing pipelines rely on subject-specific computed tomography (CT), implanted markers, digitally reconstructed radiograph (DRR)-based optimization, or deep networks trained on thousands of segmented images, which slow preparation and analysis, reducing throughput in rodent studies. We present a fully automatic, markerless, scan free framework that reconstructs six-degree-of-freedom poses of seven rat hindlimb bones from 100frames s^-1 stereo radiographs. Seventeen landmarks per view are detected by DeepLabCut and triangulated in XMALab; an extended singular value decomposition solver registers a generic OpenSim bone set to these points, estimating anisotropic scale, rotation and translation. Robustness is improved by replicating knee and ankle axis endpoints, stabilizing when landmarks are sparse or noisy. A constrained optimization refines selected landmarks within 1-4mm, driving residual error below 1mm without DRR search. In trials, projected bone silhouettes overlap manual segmentation with mean Dice coefficients of 0.94 for the femur and 0.89 for the tibia, averaged across the two X-ray views, confirming fidelity. After one refinement, a 500 paired-frame, seven bone sequence processes in 2.5s. Applied to 302 cycles from 18 female and 280 from 14 male rats, the method yields complete three-dimensional hip, knee and ankle kinematics and reveals sex specific patterns: greater hip internal rotation and ankle dorsiflexion in male rats, and greater hip abduction and ankle eversion in female rats. This streamlined pipeline delivers accurate, scalable skeletal tracking without CT, implanted markers or intensive optimization for routine biomechanical research.