Purpose: C-arm x-ray imaging systems have found widespread clinical use, for example taking 2D projections to guide interventional surgeries. Accurate 3D reconstruction from 2D images requires precise calibration of geometric parameters. Geometric calibration is typically offline, which restricts 3D imaging to a narrow library of acquisition arcs on systems with highly reproducible geometry. We present a real time online geometric calibration framework utilising a treatment couch attachment applicable to patient adaptive scans and portable systems, and show results obtained with a commercial c-arm.
Methods: We derived a Bayesian estimate of geometric parameters from observed markers. We designed a couch attachment that presents required markers during online imaging. A CIRS Dynamic Cardiac Phantom was imaged on a Siemens ARTIS pheno c-arm system with the couch attachment placed. First an automatic 3D acquisition with “gold standard” offline calibration was performed, then a manually controlled irreproducible acquisition. The first acquisition was reconstructed with uncalibrated, offline and online calibration geometry, the second with uncalibrated and online calibration as offline not applicable. The difference in parameter sets was quantified with Root-Mean-Square Deviation (RMSD) and the 5 3D images were inspected for calibration artefacts.
Results: The RMSD between offline calibrated and uncalibrated geometry was 0.37⁰ to 0.44⁰ and 1.0mm to 3.7mm while RMSD between offline calibrated and online Bayesian calibrated was 0.0013⁰ to 0.0017⁰ and 0.4mm to 0.9mm. The manual acquisition uncalibrated geometry had the most artefacts, then inbuilt acquisition uncalibrated, then automatic and manual acquisition online calibration with minimal artefacts completely absent in the automatic acquisition offline calibration reconstruction.
Conclusion: The online Bayesian calibration approach is able to recover geometric parameters and produce reconstructions to nearly the quality of “gold standard” offline calibration, with the benefit of being applicable to adaptive acquisitions and portable systems.
Funding Support, Disclosures, and Conflict of Interest: This research was supported by G201166 IPA2 with Siemens Healthineers; ACRF grant G175269; Ricky O'Brien Fellowship CI NSW Fellowship G195559, NHMRC Project Grant G193048; Tess Reynolds USyd Postdoctoral Fellowship Scheme G200793.
Cone-beam CT, Calibration, Bayesian Statistics
IM- Cone Beam CT: Development (New Technology and Techniques)