An FPGA Accelerator for 3D Cone-beam Sparse-view Computed Tomography Reconstruction

Computed tomography (CT) technique is a widely used clinical medical imaging method that employs X-rays to obtain detailed cross-sectional images of the human body. In order to minimize the ionizing radiation from X-rays, sparse-view computed tomography (SVCT) proves to be an effective solution. However, traditional reconstruction algorithms face challenges in producing satisfactory results in sparse-view scenarios. In this paper, we propose a 3D self-supervised projection network named SCOPE-3D to address the challenging sparse-view computed tomography problem. We also introduce a resource and energy efficient FPGA-based hardware accelerator that implement the proposed SCOPE-3D. Evaluation results demonstrate that the proposed accelerator reconstructs CT images with a speed and quality comparable to that of a graphics processing unit (GPU) card when running the same SCOPE-3D algorithm, while it consumes significantly fewer resources and less power.