Fig 1.
Positioning system on a scanner bed.
The system consists of a plastic holder (A), linear stages (B), an L-shaped plastic rod (C), a 20-liter water tank (D), and a 22Na point source (E).
Table 1.
Detailed characteristics of the measurement.
Fig 2.
Lesions selected from the online database.
Lesion 1, Lesion 2, and Lesion 3 in different orthogonal orientations. Since the current linear stages are only able to cover the size of a 5x5x5 cm3 cube, the lesion sizes had to be customized according to this dimension.
Table 2.
PET/CT scanners and the reconstruction settings.
Fig 3.
Reconstructed views of the lesions in three orthogonal views.
The format and the labelling are described in the main body of the manuscript. The images were obtained with the same window level and width to ensure uniformity in visualization. The values of the colorbar are expressed in SUV. In the last row the corresponding original lesions are depicted.
Fig 4.
Relative difference values between the radiomics parameters of the applied imaging settings.
The horizontal axis represents the imaging settings (without reference D) and the discretization methods for each of the three lesions, while the 46 radiomics indices are shown on the vertical axis. The colors corresponding to the percentage level of the relative difference value are the following, blue: 0–5%, light blue: 5–10%, green: 10–15%, yellow: 15–20%, orange: 20–25% and red: >25%.
Fig 5.
Coefficient of variation values (CVs) of the Radiomics parameters.
We present CV values between 0–25%, where values <10%, 10–25%, and >25% represent low, moderate, and high variability, respectively.
Fig 6.
Coefficient of variation (CV) values of the Conventional parameters.
The actual value of a CV is also shown in the cells.
Fig 7.
Effect of spatial resampling on the radiomics parameters.
The heat map indicates the coefficient of variation (CV) of the radiomics parameters. After displaying the colors of the original data (without spatial resampling), the CVs obtained with voxel sizes of 4×4×4, 3×3×3, 2.5×2.5×2.5 and 1×1×1 mm3 are shown. Here, we presented those results that were calculated using absolute discretization.
Fig 8.
P-values derived from Wilcoxon signed-rank test.
The colormap contains the p values, representing the results of the statistical tests performed between the different lesions, evaluating the data of all eight cameras. Values smaller than p = 5% are considered statistically different.
Fig 9.
Intraclass correlation coefficient (ICC) of the calculated radiomics parameters.
Based on the value of the ICC, radiomics indices can be characterized by excellent (ICC>0.9, marked with dark blue), good (0.75<ICC≤0.9, marked with light blue), moderate (0.5<ICC≤0.75, marked with green) and poor (ICC≤0.5, marked with yellow, orange, and red) repeatability. Values marked with white have ICC<0.