A Multiscale Reconstruction Framework Based on Edge-Enhanced Guidance for In-Line X-Ray Phase-Contrast CT With Limited-Angle Projections

The in-line X-ray phase-contrast computed tomography (IL-XPCCT) serves as an effective tool for studying organ function and pathologies. However, IL-XPCCT approaches often result in high radiation doses due to long scan times. To address this issue, a limited angel sampling strategy is frequently employed. However, limited-angle projection data can lead to severe divergence phenomena at sample edges or boundaries, significantly reducing image quality and affecting subsequent image analysis. In this work, we propose a novel multiscale reconstruction framework based on edge-enhanced guidance (MSRF-EEG) for IL-XPCCT limited-angle CT reconstruction. The MSRF-EEG framework consists of two primary subnetworks: an edge-enhanced subnetwork (EESN) and a multiscale reconstruction subnetwork (MSRSN). By fully considering the imaging characteristics of the IL-XPCCT, the EESN is designed to explore the edge enhancement characteristics from the IL-XPCCT images prior to phase retrieval, which is then utilized to compensate for edge distortions during the reconstruction process. The MSRSN performs image reconstruction with edge-enhanced guidance across multiple scales simultaneously. Furthermore, to better leverage the edge enhancement characteristics before phase retrieval for reconstructing the IL-XPCCT images from limited-angle projections, an edge aggregation (EAG) module and an edge attention (EAT) module are incorporated into the MSRSN. Simulations and real data experiments are conducted to evaluate the performance of our proposed MSRF-EEG framework. Compared with the existing competitive algorithms, both quantitative and qualitative results demonstrate that the proposed method can significantly enhance reconstruction quality under different sampling angles and improve detail restoration and edge preservation.