Author: Caroline Chung, Michael Knopp, Stephen F. Kry, Hunter S. Mehrens, John Rong 👨🔬
Affiliation: The University of Texas MD Anderson Cancer Center, UT MD Anderson Cancer Center, University of Cincinnati 🌍
Purpose: To evaluate the variability of CT dose index (CTDIvol) and radiomics features across a large cohort of radiotherapy simulation CT scans from multiple institutions.
Methods: Three IROC phantoms were irradiated by institutions as part of radiotherapy credentialing for NCI’s clinical trial programs between 2013-2024. (1) Head and neck (HN; N=483) (2) Stereotactic head (SRS; N=384), and (3) Thoracic (N=247). Each phantom underwent a "standard of care" simulation CT at the respective institution. CTDIvol values were either abstracted from the DICOM header or calculated using DICOM header parameters. Additionally, 100 radiomics features were calculated for each phantom's primary target(s) and select organ-at-risk (OAR) structures.
Results: The stereotactic head phantom had the highest mean CTDIvol (54.6±31.8 mGy) compared to the head and neck (39.1±28.1 mGy) and thoracic (22.3±26.2 mGy) phantoms, with all differences statistically significant (p < 0.001). Subgroup analysis by manufacturer and model showed a similar trend across phantoms. Many radiomics features would not be influenced by CTDIvol and, correspondingly, Pearson’s correlation coefficients between radiomics features and CTDIvol were generally weak (|rs|<0.5), implying other factors such as reconstruction kernels, filters, and algorithms could also contribute to CTDIvol. However, some metrics did show strong correlation such as inverse variance for head and neck’s primary target. Of note, while 40% of features for the head and neck targets showed statistically significant correlations (p < 0.05) with CTDIvol, less than 30% of features for the stereotactic head and thoracic phantom targets did.
Conclusion: Significant variations in CTDIvol values were observed across different institutions; an observation that spanned different anatomical sites. Many radiomics features correlated with CTDIvol, which highlights the potential for inconsistency in these metrics in multi-institutional situations. Further effort to standardize and harmonize imaging, including RT simulations, is necessary to achieve consistent imaging data and maximize the potential value of multi-institutional data.