A Comparative Study of Mismatch Projectors in Cone Beam CT

Fast projectors are widely used for reducing the computational costs of modeling X-ray optics in iterative CT reconstruction. This study examines the strengths of state-of-the-art GPU-based projectors in mismatch settings, where forward and back-projection combine the computational advantages of different methods in traditional iterative methods as well as momentum based methods. Specifically, we consider the Siddon, voxel-driven, and distance-driven projectors together with the recently introduced convolutional nonseparable footprint (CNSF) projector under realistic conditions that incorporate effects of the Beer-Lambert law as well as discretization errors. Our experiments suggest that specific mismatched combinations can achieve faster convergence and better image quality than some commonly-used matched projectors. We assess the performance of different projector combinations in terms of both reconstruction runtime and accuracy. These findings provide insights into selecting optimal GPU-based projector pairs, aiming to achieve an ideal balance between speed and quality in CT reconstruction.