UniSino: Physics-Driven Foundational Model for Universal CT Sinogram Standardization

-- During raw - data acquisition in CT imaging, diverse factors can degrade the co llected sinograms, with un-dersampling and noise leading to severe artifacts and noise in reconstructed images and compromising diagnostic accuracy. Conventional correction methods rely on manually designed algorithms or fixed empirical parameters, but thes e approaches often lack generalizability across heterogeneous artifact types. To address these limitations, we propose UniSino, a foundation model for universal CT sino-gram standardization. Unlike existing foundational models that operate in image domain, UniSino directly standardizes dat a in the projection domain, which enables stronger generalization across diverse undersampling scenarios. Its training framework incorporates the physical characteristics of sinograms, enhancing generalization and enabling robust performance across mul tiple subtasks spanning four benchmark datasets. T he code is available at: https://github.com/yqx7150/UniSino . In CT imaging, the sinogram represents the raw pr ojection data before image reconstruction . However, in practical clinical data acquisition, raw sinogram data are frequently degraded by a multitude of factors -- including hardware limitations, environmental variability, and patient - induced factor s -- resulting in complex and heterogeneous data corruption [2]. Without effective preprocessing, such imperfections are readily amplified through the reconstruction process, which lead to severe image artifacts [3], including detector - induced ring patterns [4], beam ha rdening from metal implants [5], geometric distortions from miscalibration [6], and motion - induced inconsistencies [7, 8]. These artifacts not only degrade visual quality but critically compromise diagnostic reliability.