Accelerated Monte Carlo-driven statistical reconstruction for CBCT scatter correction

The traditional cone-beam computed tomography (CBCT) model has not sufficiently accounted for the effects of X-ray scattering, resulting in cupping and streak artifacts in the reconstructed images. These artifacts significantly degrade the quality of soft tissue imaging, severely impacting the accuracy of medical diagnosis. Accurately modeling the X-ray scattering process and efficiently solving it are crucial for scattering correction. Compared to traditional convolution-based methods, the Monte Carlo (MC) particle transport process offers higher precision in characterizing scattering effects. This paper presents a CBCT statistical imaging model based on the MC process and proposes an alternating iterative reconstruction algorithm. The algorithm employs a multi-step sparse scattering signal recovery strategy to accelerate the MC process, thereby improving the convergence speed. Additionally, it adopts a scattering removal scheme based on general scattering representation, effectively suppressing noise. Numerical and practical experimental results demonstrate that the proposed method effectively restores the structural details lost due to scattering effects and significantly shortens computation time to a few minutes, making it highly applicable in practice.