Iterative reconstruction algorithms for limited-angle tomography of thin slab objects: a performance review

Abstract As computed tomography (CT) applications have expanded, scenarios with restricted angular coverage, known as limited-angle tomography (LAT), are increasingly encountered. LAT commonly arises in industrial and non-destructive testing applications, where CT cannot fully traverse an object due to its size or due to obstructions at certain angles. Conventional filtered backprojection (FBP) cannot restore the missing data in LAT, causing artifacts. In this study, iterative reconstruction (IR) algorithms—namely the simultaneous algebraic reconstruction technique, conjugate gradient least-squares, adaptive steepest-descent projection-onto-convex-sets (ASD-POCS), and maximum-likelihood expectation-maximization (ML-EM)—were applied to LAT of thin-slab objects, such as printed circuit boards, as representative algebraic, least-squares, compressed-sensing, and statistical reconstruction approaches, respectively. The contrast-to-noise performance was evaluated on the in-focus plane of a thin aluminum disc phantom, while ghosting artifacts were assessed via artifact-spread functions on out-of-focus planes. The IR algorithms demonstrated improved contrast-to-noise performance under sparse data conditions. Although the statistical ML-EM and compressed-sensing ASD-POCS algorithms reduced ghosting artifacts compared to FBP, increasing the total scan angle was found to be the most effective strategy to mitigate artifacts, highlighting the fundamental limitation of LAT. This study discusses the principles of LAT and algorithmic strategies for restoring missing data.