Purpose: Cone-beam computed tomography (CBCT) is used in image-guided radiotherapy to ensure correct patient positioning prior to treatment delivery. However, patient-generated x-ray scatter degrades CBCT image quality. Monte Carlo (MC) simulations offer the most physically accurate means to predict scatter signals, which can improve CBCT images via subtraction in projection space. Unfortunately, MC simulations can be too computationally costly for clinical application. We present a novel Bayesian Monte Carlo extrapolation (BMCE) algorithm that uses Bayesian statistics to mitigate the statistical noise present in MC-estimated scatter signals. The BMCE scatter correction method is then used to correct a 200-projection CBCT scan of a lung cancer patient.
Methods: The BMCE method uses MC simulations to estimate the scatter component of each CBCT projection by transporting a 120 kVp x-ray beam through the planning 4DCT volume. Statistical noise present in the MC-simulated scatter signals was then corrected using an iterative Bayesian extrapolation postprocessing algorithm. Finally, the Bayesian-extrapolated scatter signals were subtracted from CBCT projections and images were reconstructed in 3D and 4D using the Feldkamp-Davis-Kress (FDK) algorithm. The quality of the CBCT images corrected using our BMCE algorithm was quantified in terms of contrast-to-noise ratio (CNR) and Hounsfield unit (HU) integrity. Image quality was then compared to images reconstructed without any scatter correction.
Results: The BMCE scatter correction algorithm increased CNR for 3D and 4D reconstructions by factors of 3.4 and 3.8 respectively, and decreased the root mean square error (RMSE) between the HU-calibrated 3DCBCT reconstruction and the planning 3DCT by 30% relative to the 3DCBCT image reconstructed without any scatter correction.
Conclusion: We demonstrate that Bayesian extrapolation of scatter signals estimated using Monte Carlo simulations with low numbers of primary x-rays can render otherwise noisy scatter distributions suitable for implementation in a CBCT scatter correction algorithm.
Funding Support, Disclosures, and Conflict of Interest: This work was funded by an NHMRC project grant 1138899 and partially by the Cancer Australia PdCCRS project grant number 1123068. RO was funded by a Cancer Institute NSW career development fellowship.