Investigating the Effects of Computed Tomography Reconstruction Settings from Multiple Scanners on IBSI Compliant Radiomic Features

Radiomic metrics offer promising tools for characterizing medical image textures. Yet, radiomics features are sensitive to reconstruction algorithm variation. Using International Biomarker Standardization Initiative (IBSI) compliant radiomic features, this study aimed to investigate the effect of reconstruction settings on radiomic features. A physical phantom, mimicking biological textures and densities, was scanned on six different computed tomography (CT) scanners. Radiomic features were then calculated for the six regions of interest (ROIs) of the phantom. To assess distribution, z-scores for each feature were calculated, identifying similar and dissimilar reconstruction setting profiles. This was then summarized in a series of heatmaps, identifying both stable reconstruction settings and outliers. To determine feature robustness to scanner and reconstruction kernel variation, coefficient of variation (CoV) was calculated. A majority (67.2%) of radiomic features yielded high CoVs, thus are not robust to inter-scanner variability. 4.1% of radiomic features were robust with a CoV less than 5%. A breakdown of the 463 IBSI radiomic features and which are robust is elucidated. Following these calculations, recommendations for which features to include or omit are made.