DC-Swin: A Dual-domain Collaborative Swin Transformer for Truncation-artifact Correction in Photon-counting Computed Tomography

To address the issue of projection-data truncation in photon-counting computed tomography (CT), arising from a sample exceeding the effective field of view of the detector, a dual-domain collaborative truncation-artifact correction network based on the Swin transformer is proposed. First, an edge-extrapolation module is designed in the projection domain, which employs a multiscale feature-fusion approach to extrapolate the edge information of truncated sinograms. Second, a reconstruction algorithm is utilized to simultaneously process both the truncated projection data and the extrapolated projection data, thereby achieving dual-channel information fusion. Finally, an artifact-correction module is constructed in the image domain to capture the transition from detailed features to global structural features within the dual-channel information, thus realizing the correction of truncation artifacts. Experimental results demonstrate that the proposed network can effectively extrapolate truncated data, suppress the interference of truncation artifacts on reconstructed images, and render the corrected images more consistent with real data. Compared to the state-of-the-art dual-Swin method, the results obtained by the proposed method exhibit a 6.4% improvement in peak signal-to-noise ratio (PSNR) and a 6.3% improvement in structural similarity index measure (SSIM).