Purpose: In penalized-likelihood Known Component Reconstruction (KCR), prior models of metal hardware are used to mitigate metal artifacts. We investigate the impact of scatter correction accuracy on the efficacy of KCR in orthopedic Cone-Beam CT (CBCT).
Funding Support, Disclosures, and Conflict of Interest: NIH grant R01 EB025470 Cone-beam CT, Scatter, Monte Carlo
Methods: An extremity CBCT system (1.3 magnification, 360 projections) operated at 120 kVp was simulated. The imaging phantom comprised of a water ellipsoid (12 by 10 mm) containing fat and bone inserts and an 4 mm diameter, 50 mm length Titanium screw. Adjacent to the screw, there were two trabecular regions-of-interest (ROIs) with BMD of 175 mg/ml and 100 mg/ml. Polyenergetic primary projections were obtained using ray-tracing, scatter was simulated with an accurate, low-noise Monte Carlo (MC) simulation with 10¹¹ photons. Prior to KCR, scatter estimates of varying fidelity were subtracted from the data: (i) projection-wise mean of the original MC simulation (mean correction), and (ii) fast MC-based corrections, involving noisy MC with low number of photons (10⁶ – 10¹⁰, corresponding to 25s-7h runtime per scan) followed by Gaussian denoising (standard deviation σ=5-40 mm). The KCR forward model incorporated a polyenergetic projection of the Ti screw pre-hardened to account for tissue attenuation.
Results: For KCR without scatter correction, the absolute BMD error with respect to a scatter-free reference was ~145 mg/ml for the 175 mg/ml ROI and 55 mg/ml for the 100 mg/ml ROI. With mean correction, the BMD accuracy was improved to 80 mg/ml for the 175 mg/ml ROI and 25 mg/ml for the 100 mg/ml ROI. Fast MC-based scatter correction reduced the BMD errors to <5 mg/mL across all evaluated photon track numbers and smoothing kernel sizes.
Conclusion: KCR is highly susceptible to scatter-induced artifacts. Fast MC (
Funding Support, Disclosures, and Conflict of Interest: NIH grant R01 EB025470
Cone-beam CT, Scatter, Monte Carlo