Score-Based Generative Null-Space Shuttle for the Field-of-View of STCT Expansion

Abstract

Micro computed tomography (Micro-CT) is widely used across various fields for high-resolution imaging. Recently, our previous work developed a source translation-based computed tomography (STCT) model to achieve high-resolution imaging for large objects. However, when the sample size exceeds the field-of-view (FOV) of STCT, the traditional algorithms cannot recover the invisible null-space information from incomplete projection data. To address this issue, we propose the score-based generative null-space shuttles (SGNS) algorithm, which employs score-based generative models to learn prior information and restores missing null-space information through a null-space shuttle approach during the sampling process. To ensure consistency in the generated results, the measured data are introduced as ground truth information during the sampling phase. The numerical and physical experiments demonstrate our algorithm can effectively eliminate artifacts caused by insufficient projection data and recover more detailed image information. In addition, by using range-null space hallucination maps, we demonstrate the proposed algorithm can reliably and stably reconstruct cross-sectional images of objects beyond the FOV.