Deep Image Prior for Sparse-View Reconstruction in Static, Rectangular Multi-Source X-Ray CT Systems for Cargo Scanning

A static, multi-source X-ray Computed Tomography (CT) system facilitates rapid multi-view X-ray radiography, significantly improving the efficiency of cargo scanning. However, reconstructing images from sparse-view X-ray data in cargo scanning is challenging, particularly when conventional deep learning reconstruction techniques are hampered by a scarcity of training data. This work proposes the application of Deep Image Prior (DIP), which does not require training data, to reduce undersampling reconstruction artifacts arising from sparse-view and restricted opening angle acquisition in X-ray CT systems tailored for large-scale cargo scanning in harbors.(1) The work particularly targets a rectangular multi-source X-ray CT system, featuring up to 40 equidistantly distributed static X-ray sources with a 30-degree opening angle.(2) Our study demonstrates that DIP improves the quality of of sparse-view cargo CT in terms of PSNR and SSIM compared to traditional reconstruction methods.