An Improved Discrete Algebraic Reconstruction Technique for Limited-View Based on Gray Mean Value Guidance

Discrete Algebraic Reconstruction Technique (DART) as a heuristic algorithm is capable of computing better results even using fewer numbers of projections than the Algebraic Reconstruction Technique (ART), especially for a single material object. Although it has reconstructed the objects that consisting of only a few different materials robustly, the efficiency for parameter estimation such as prior gray values and segmentation threshold is time-consuming. In this work, we consider a new Gray Mean Value Guided DART(MVG-DART) method through optimizing the prior parameters to address the inverse problem of reconstructing the structure of an object from limited X-ray projections. Firstly, the preliminary image reconstructed from the algebraic reconstruction algorithm, such as SIRT, is segmented. And the local gray value in the region is corrected by residual projection data. Then, the process that minimizing the projection distance is used to obtain a better threshold to segmentation by advance and retreat method. Simulation and actual reconstruction experiments show that the proposed method improves the reconstruction quality effectively and speed up the convergence rate of the DART algorithm significantly. And the acceleration ratio is well improved about 7.7 times compared with the DART algorithm, which has demonstrated the potential of our method for incomplete projection reconstruction, such as limited projections.