Computed tomography-positron emission and tomography image reconstruction under compressed sensing constraints

A novel generative framework, RED-WGAN, is presented for the reconstruction of high-resolution CT-PET images under conditions of sparse sampling, motion corruption, and low-dose acquisition.The method integrates deformable motion compensation, residual deblurring, and Transformer-based denoising within a unified architecture.Joint priors from structural (CT) and functional (PET) data are combined with Ridgelet-domain sparsity and compressed sensing principles to enable the recovery of anatomically precise and perceptually coherent images from highly degraded inputs.The framework was evaluated using both digital phantom models (Shepp-Logan and Zubal) and clinical in vivo CT-PET datasets encompassing a range of anatomical sites and motion scenarios.RED-WGAN was benchmarked against twelve state-of-the-art super-resolution algorithms under varying levels of data sparsity (20-100%) and motion distortion.Superior performance was consistently observed, with PSNR values reaching 39.03 dB, SSIM up to 0.921, and LPIPS reduced to 0.118.Robust image quality was maintained at compression ratios as low as 40%.Furthermore, the framework achieved the lowest total registration error (1.44 voxels) across all evaluated motion compensation methods.These findings demonstrate that RED-WGAN offers a technically robust and clinically scalable solution for CT-PET image enhancement.Its ability to preserve structural integrity and improve cross-modality alignment under constrained acquisition conditions positions it as a promising tool for applications in low-dose imaging, pediatric diagnostics, rapid scanning protocols, and motion-prone clinical environments.