High-Precision CBCT Scatter Correction via Monte Carlo Simulation and Physics-Constrained Deep Learning

Cone beam computed tomography (CBCT) plays a critical role in image-guided radiotherapy by enabling accurate patient positioning and adaptive treatment planning. However, its image quality is significantly degraded by scatter artifacts arising from photon-tissue interactions such as Compton and Rayleigh scattering, which compromise both soft tissue contrast and dose calculation accuracy. To address this challenge, we propose a hybrid framework that integrates high-fidelity Monte Carlo (MC) simulation with data-driven deep learning for efficient and accurate scatter correction. The MC simulation component generates ground-truth scatter distributions under realistic clinical CBCT acquisition parameters, while a convolutional neural network (CNN) is trained to extract structural and texture priors from raw projections to estimate the scatter component rapidly and effectively. By combining the physical rigor of MC modeling with the generalization capacity of deep learning, the proposed method achieves robust and high-quality scatter estimation. Quantitative evaluations demonstrate that the corrected reconstruction yields a root mean square error (RMSE) of approximately 9 HU, with a structural similarity index (SSIM) of 0.93 and a universal quality index (UQI) of 0. 95. These results markedly outperform both conventional anti-scatter grid techniques (SSIM:0. 82, UQI:0. 85) and pure deep learning baselines (SSIM:0. 87, UQI:0. 89). Furthermore, the framework is computationally efficient and suitable for time-sensitive clinical workflows. This hybrid approach not only improves the fidelity of CBCT images- particularly in soft tissue regions-but also supports more accurate target delineation and dose calculation. Its modular design enables future integration of advanced neural architectures and adaptation to diverse anatomical scenarios, paving the way toward artifact-free, quantitative CBCT imaging in precision radiotherapy.