Purpose: Weight-bearing extremity cone-beam CT enables monitoring of fractures under physiological load. To support this capability, we utilize mechanical finite-element models of fracture fixation implants to obtain CBCT-based measurements of load-induced implant deformation. Furthermore, the registered implant models are incorporated in polyenergetic model-based reconstruction, mitigating artifacts and achieving quantitative measurements of bone mineral density (BMD) in the vicinity of surgical hardware.
Methods: A phantom was constructed using Ca inserts 200-300 mg/cc and Ti rods (8 cm x 6.3 mm diameter) with different degrees of bending. The phantom was imaged on a test bench emulating extremity CBCT (56 cm source-detector, 43 cm source-axis) using a nominal protocol (90 kVp, 210Â° rotation, ~15 mGy). Deformable registration was performed to obtain implant positions and deformations from a model of an intact rod. The registration utilized a finite-element model of load-induced deformation. The loading parameters yielding optimal gradient correlation with measured projections were found via CMAES optimization. The deformable component of the registration error was evaluated for each rod. Polyenergetic model-based reconstruction (PolyPL) incorporating the deformed implant models and Monte Carlo scatter correction was performed. Artifact reduction and BMD accuracy were measured in the reconstructed images and compared to PolyPL without implant models.
Results: Finite element models combined with 3D-2D registration resulted in accurate recovery of implant deformation, with <0.16 mm error in deflection for all Ti rods compared to air scans. By incorporating the models into PolyPL reconstruction and scatter estimation, streak artifacts decreased by 86% and BMD errors were decreased by 59% in the reconstructed image compared to the PolyPL framework without implant modeling.
Conclusion: Phantom studies indicate quantitative assessment of implant deformation and measurements of BMD in the vicinity of the implant can be achieved using a combination of deformable 3D-2D registration, mechanical modeling of the implant, and polyenergetic model-based reconstruction.