Purpose: Most radiomics studies performed to date were assessed in a mostly pre-treatment setting using either MRI and CT. Unlike MR and CT images that are used for diagnostic and treatment planning purposes, CBCT images are used for daily patient setup prior to radiation delivery and may provide day-to-day radiomic feature changes of the prostate. While the predictive value of multiparametric (mp)MRI in prostate cancer (pCa) treatment is well-known, the potential of imaging features extracted from CBCT is novel for patients with pCa. The purpose of this study was to determine if there is any correlation between the prostate radiomic features between MRI, CT, and CBCT and to determine the mechanism through which they correlate.
Methods: Twenty patients receiving radiotherapy for pCa who were treated on a variety of dose fractionations as part of IRB-approved protocols were included. Forty-two radiomic features were extracted from the prostate on CT, first fraction CBCT, and mpMRI sequences, including Apparent Diffusion Coefficient, Early and Late Dynamic Contrast Enhanced, T2-weighted, and T2 Fat-Saturated MRI. First-order intensity and second order statistical radiomic features from four gray-level matrix classes were considered. The correlation between radiomic features and prostate volume was also calculated. The Spearman correlation (R) was used to investigate the associations between the radiomics features of the aforementioned imaging modalities. Statistical significance was p<0.05. The threshold for well-correlated radiomic features was R>0.85.
Results: CBCT and CT radiomic features were correlated with features from 5 and 3 mpMRI, respectively. Most well-correlated radiomic features were also well-correlated with volume. After volume-normalization, these radiomic features were no longer well-correlated between different modalities.
Conclusion: The majority of radiomic features that correlated between CT, CBCT, and mpMRI was due to the strong volume-dependence of these features. This work highlights the importance of volume-normalization when considering radiomic features between different modalities.
Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by a research grant from Varian Medical Systems, Palo Alto, CA. (GR013242). Dr. Abramowitz has received an honorarium from Varian.
Not Applicable / None Entered.