FSTU-Net: A sparse-view CT reconstruction framework for enhanced imaging efficiency and quality in neutron tomography

Computed tomography (CT) imaging has revolutionized non-destructive testing and scientific research by enabling three-dimensional visualization of internal structures. Among the various CT modalities, neutron CT stands out for its exceptional capabilities in material characterization and analysis. However, both neutron and X-ray CT face significant challenges when dealing with sparse-view data acquisition, which often leads to reconstruction artifacts and degraded image quality. To address these issues and enhance imaging efficiency, we propose a novel sparse-view CT reconstruction framework, Filtered Swin Transformer U-Net (FSTU-Net). This framework combines filtered back projection for initial reconstruction with Swin Transformer U-Net for refinement. The FSTU-Net leverages the global modeling power of the Swin Transformer and the multi-scale feature fusion of U-Net to recover fine structural details while mitigating streak artifacts. The shifted window attention mechanism of the Swin Transformer ensures efficient feature extraction, making the method computationally efficient and suitable for large-scale neutron CT imaging facilities like the China Spallation Neutron Source. Our experiments on multiple neutron CT datasets validate the effectiveness of FSTU-Net, demonstrating superior denoising and structural preservation, even when trained on mixed datasets. These results underscore the framework’s potential in improving imaging efficiency and quality for neutron sparse-view CT.