Author: Cem Altunbas, Farhang Bayat, Roy Bliley, Brian Kavanagh, Uttam Pyakurel, Tyler Robin, Ryan Sabounchi 👨🔬
Affiliation: Department of Radiation Oncology, University of Colorado School of Medicine, Taussig Cancer Center, Cleveland Clinic 🌍
Purpose: The use of image features extracted from serial CBCT images to assess radiotherapy response and toxicity is an active research area. However, poor image quality often compromises reliability of CBCT features, limiting effectiveness in predicting treatment outcomes. This study evaluates the impact of 2D antiscatter grid-based quantitative CBCT (qCBCT) method on the robustness of image features in CBCT images of prostate cancer patients.
Methods: Nine prostate cancer patients were enrolled in an IRB-approved study. Each was scanned using the Varian TrueBeam’s iCBCT protocol and qCBCT with identical acquisition parameters. Prostate, rectum, bladder, and seminal vesicles (SVs) were delineated in CBCT and planning CT (pCT) images. A total of 315 radiomic features, including first-order statistics and texture-based features (GLCM, GLSZM, NGTDM), were extracted. Agreement between CBCT and pCT features was assessed using Concordance Correlation (CCC) and Pearson Correlation Coefficients (PCC). Features with PCC and CCC values above 0.6 were deemed robust.
Results: PCC analysis showed that 114 and 76 features were robust in the prostate for qCBCT and iCBCT, respectively. Number of robust features in qCBCT and iCBCT were 117 and 58 in the bladder, 106 and 64 in SVs, and 3 and 9 in the rectum. In general, features extracted from qCBCT had larger PCC and CCC values than iCBCT, implying that the level of robustness is higher in qCBCT. Mean PCC values for iCBCT and qCBCT were 0.66 and 0.74 (p<0.001) in the bladder. Mean CCC values in SVs were 0.63 and 0.80 (p<0.001) for iCBCT and qCBCT, respectively.
Conclusion: qCBCT images of prostate cancer patients showed up to 200% increase in the number of robust radiomic features and had higher robustness in general. This suggests qCBCT may enhance the correlation between treatment outcomes data with CBCT image features, increasing its potential in treatment response and toxicity assessment.