Repeatability and Reproducibility of Radiomic Features Produced over Time by the Fan-Beam kV-CT on a Novel Ring Gantry-Based PET/CT Linear.

The RefleXion® X1 is a new external-beam radiotherapy delivery system on a ring gantry with dual imaging systems fan-beam kV-CT and PET. The imaging system has the potential to be used for cancer prognosis and treatment response assessment using the radiomic features obtained from daily treatment images. The scanning variability in radiomics features and the differences between scanning protocols must be evaluated to warrant the accuracy of the prediction model. This study aims to establish the technical validation of the repeatability and reproducibility of radiomic features produced by the X1 kV-CT. The Credence Cartridge Radiomics phantom was scanned 10 times on the X1 system over a 3-month time frame using the two most frequently used scanning protocols: head/medium dose/fast couch and body/medium dose/slow couch. A cylindrical region of interest of 8.0cm in diameter x 1.5cm in thickness was contoured for each of the 6 cartridges. The 6 cartridges are each comprised of different materials using acrylonitrile butadiene styrene, acrylic beads, polyvinyl chloride, hemp seeds encased in polyurethane, shredded rubber and dense cork. These materials were chosen to produce a range of radiomics feature values. 66 radiomic features were extracted using a widely available, IBSI-compliant radiomics software package. Linear regression analysis was performed to evaluate the repeatability and reproducibility over time. T-test for the difference in the mean and F-test for the difference in standard deviation was used to compare the radiomics features between scanning techniques. There was no significant variation of radiomic features over time. Based on linear regression analysis between the radiomic features and number of elapsed days since the initial scan, the fitted coefficient was less than 0.05 for all the radiomic features (scaled so that the mean value was unity). There was no strong linear relationship between radiomic features and elapsed days (R2 <= 0.02 in all regression analysis results). As reported in other studies, a comparison of radiomic features within each of the six cartridges demonstrated that up to 38 of the 66 radiomic features showed statistically significant differences between the fast scan and the slow scan protocols (two-tailed p value < 0.05 in paired t-tests) and up to 25 radiomic features had statistically significant differences in standard deviations between the fast and slow scan protocols (p value < 0.05 in F-tests), indicating the necessity of using consistent scan protocols to monitor radiomic features throughout the treatment course accurately. The clinically useful CT radiomic features produced by the X1 are reproducible over time, demonstrating its utility as a quantitative imaging platform. The CT radiomic features produced by the X1 vary significantly with the scanning protocols. Evaluating radiomics feature changes over time requires image acquisition using the same scanning protocol. [ABSTRACT FROM AUTHOR]