HYBRID IMAGE- AND PROJECTION-DOMAIN RECONSTRUCTION METHODS FOR LOW-DOSE X-RAY CT IMAGING

Low-dose X-ray computed tomography (CT) is critical for minimizing patient radiation exposure, but it often results in sparse-view data that degrade image quality. Hybrid reconstruction methods, combining image-domain and projection-domain approaches, have emerged as effective solutions to address this challenge. This study reviews and evaluates recent developments in hybrid reconstruction techniques, including generative adversarial networks (GANs), convolutional neural networks (CNNs), attention-based deep learning, and multi-scale Fourier interpolation. We highlight how these methods enhance sparse-view CT images by leveraging complementary strengths of projection-domain interpolation and image-domain refinement. Experimental findings demonstrate significant improvements in peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and artifact suppression compared to traditional methods. Our results suggest that hybrid methods offer a promising framework for accurate and high-quality low-dose CT reconstruction, enabling safer imaging without compromising diagnostic performance.