Speaker
Description
Introduction
The image quality reconstruction in Computed Tomography (CT) is dependent on different factors as the number of projections and angular sampling and the reconstruction method. In this work Fan‑Beam (FB) and Cone‑Beam (CB) CT geometries are studied, by different reconstruction algorithms applying reduced angular sampling.
Materials and Methods
It is implemented a methodology using a unified GPU‑accelerated based on Matlab ASTRA-CUDA framework. About analytical reconstruction algorithms, Back Projection (BP), and Filtered Back Projection (FBP) have been analyzed in FB geometry. And BP3D and Feldkamp-Davis-Kress algorithm (FDK) in CB geometry. Related to iterative algorithms, in FB geometry have been evaluated Conjugate Gradient Least Squares (CGLS), Simultaneous Iterative Reconstruction Technique (SIRT) and Simultaneous Algebraic Reconstruction Technique (SART). And in CB geometry CGLS3D and SIRT3D.
Two sampling regimes has been tested, many projections (Dense‑View, DV reconstruction) and reduced projections (Sparse‑View, SV reconstruction) to compare the impact of reduced angular sampling in FB and CB geometries.
Results
Efficiency in reconstruction technique is assessed using a high‑resolution, low‑noise image reference, applying standard metrics in CT reconstruction as L2, RMSE, PSNR, SSIM and NMAD.
Shifting from DV to SV caused image reconstruction degradation across both FB and CB geometries. Metrics L2, RMSE, and NMAD increased substantially. PSNR and SSIM decreased. Streaking artifacts and fine‑detail loss became more prominent.
Analytical reconstruction methods (BP, FBP/ FDK) results, show the strongest degradation under SV conditions, consistent with evidence of their angular‑sampling sensitivity. Iterative methods (SIRT, CGLS) results are more robust, preserving structure and mitigating SV artifacts, according its advantages under limited‑projection conditions.
Conclusions
Angular sparsity is the primary degradation factor in both CBCT and FBCT. Iterative algorithms (SIRT, CGLS) consistently outperform approaches under SV, yielding more stable and structurally accurate reconstructions. The unified ASTRA–CUDA framework enables the comparison and reconstruction strategies as a base for sparse‑view CT.