Influence of scan mode, tilt, and radiation dose on CT radiomic metrics.

backgroundRadiomic features derived from computed tomography (CT) are highly susceptible to variations in acquisition parameters, which can introduce confounding effects in multicenter research and reduce diagnostic accuracy. While the effects of parameters such as scanning mode and dose have been studied, the impact of gantry tilt-despite its routine clinical use-remains underexplored in radiomics literature. PURPOSE: To systematically evaluate how scan mode (axial vs. helical), gantry tilt (0° vs. 5°), and radiation dose affect CT-based radiomic metrics using an anthropomorphic liver phantom containing six 3D-printed texture inserts, with special emphasis on the novel inclusion of tilt.

methodsTwelve unique image acquisition configurations were scanned on a GE Revolution Apex CT scanner, with each configuration repeated once. Manual segmentation of volumes of interest (VOIs) was performed, and 93 radiomic features spanning six texture families were extracted using PyRadiomics. First-order dispersion metrics (standard deviation, interquartile range, and coefficient of variation) were analyzed alongside higher-order features via regression with heatmap visualization, and repeatable, robust, and calibratable features were identified.

resultsHelical scans without tilt generally exhibited lower first-order dispersion than axial scans. Introducing a 5° tilt reduced dispersion in axial scans but had inconsistent effects in helical scans, with no coherent trend observed. Radiation dose demonstrated an expected inverse relationship with dispersion metrics. Intraclass correlation coefficient (ICC) analysis revealed that 34% of radiomic metrics exhibited good or excellent repeatability across all trials (ICC ≥ 0.6), but only 13% demonstrated good or excellent robustness, highlighting the sensitivity of radiomic metrics to scanning conditions. Regression analysis yielded 31 metrics (33%) that can be calibrated using their significant linear relationships with the parameters varied in this study, thereby allowing researchers to correct for variations in acquisition settings.

conclusionsThese findings underscore the importance of accounting for acquisition variability-including less frequently examined parameters such as tilt-when designing radiomic studies, selecting robust features, and interpreting results in clinical and multicenter studies. This approach helps distinguish meaningful biological variation from imaging artifacts, thereby improving the reliability of radiomic analysis in personalized medicine.