Improved Filtered Back-projection Reconstruction for Inverse Geometry CT without Gridding

• Published in: 2019 International Conference on Medical Imaging Physics and Engineering (ICMIPE) pp. 1 - 6
• Main Authors: Huang, Bangyan, Jiang, Xiao, Xu, Gongming, Cui, Hehe, Zhu, Lei
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.11.2019
• Subjects: filtered backprojection; interpolation; inverse geometry CT
• DOI: 10.1109/ICMIPE47306.2019.9098192

Inverse geometry CT (IGCT) with multiple sources and a small detector poses new challenges on accurate reconstruction. Current rebinning methods convert IGCT projections into a sinogram of single source via gridding, leading to inevitable loss of image resolution. Although direct filtered-backprojection (FBP) reconstruction with no gridding or iterations generates images with high spatial resolution, it will cause severe artifacts when the overlapping areas are insufficient, because sharp transitions on the edges will cause peaks in the filtered projections. In this paper, we propose a new FBP reconstruction algorithm to avoid filtering on those sharp edges. We introduce a simple interpolation prior to the filtration step to expand the projections to full size. After filtering the expanded projections, weighting functions are employed to eliminate redundancy. We design an IGCT system with nearly zero overlap, on which the proposed method is evaluated via both simulation studies on the Shepp-Logan phantom and phantom studies on the Catphan® 600. Our algorithm demonstrates a performance on numerical accuracy and spatial resolution almost identical to that of a direct FBP method implemented in situations with sufficient redundancy. While maintaining spatial resolution identical to that of the direct FBP method, no artifacts will occur in situations with insufficient overlap. And time required for interpolation is negligible. Moreover, here is potential on reducing the computation burden, because data required for backprojection is N (the number of sources) times less than that in the direct FBP method.