3598 - Repeatability Assessment of Radiomic Features Using Enhanced Cone Beam CT for Liver Tumors
Presenter(s)
C. Claunch1, P. Pathak1, S. K. Kamel2, O. Awad1, A. N. Hanania3, Z. A. Siddiqui4, S. S. Desai1, and A. S. Mohamed1; 1Department of Radiation Oncology, Dan L. Duncan Comprehensive Cancer Center, Baylor College of Medicine, Houston, TX, 2Department of Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, 3Department of Radiation Oncology, Dan. L. Duncan Cancer Center, Baylor College of Medicine, Houston, TX, 4University of Pittsburgh Medical Center, Pittsburgh, PA
Purpose/Objective(s): To evaluate the test-retest repeatability of radiomic features extracted from an enhanced cone beam computed tomography (CBCT) imaging platform for liver tumors and surrounding normal tissue, testing the hypothesis that this technology enables highly repeatable quantitative imaging features for adaptive radiotherapy applications.
Materials/Methods: Five patients with liver tumors underwent two CBCT scans per day during the course of five liver stereotactic body radiotherapy (SBRT) treatments using an enhanced CBCT platform. Gross tumor volumes (GTV) and normal liver tissue were contoured each fraction, and the structures were co-registered to repeated CBCT’s. Images unable to be co-registered due to patient positioning or excess artifact were excluded. Radiomic features were extracted using open source software, including shape metrics, first-order statistics, and texture features. Repeatability was assessed using Concordance Correlation Coefficient (CCC), Coefficient of Variation (CV), and Intraclass Correlation Coefficient (ICC). Features meeting all criteria of ICC/CCC > 0.9 and CV < 0.3 were classified as excellent repeatability metrics.
Results: A total of 32 CBCT were included in this analysis. Of 103 total extracted features, shape-based features demonstrated the highest repeatability with 86% (12/14) meeting excellent criteria for both tumor and normal liver. First-order features showed more variability, with 19% (3/16) meeting excellent criteria. Among texture features, 27% (22/73) achieved excellent repeatability across all categories. Volume measurements showed perfect repeatability (ICC = 1.0) with low variability (CV < 0.2). Normal liver tissue exhibited similar patterns, with notably high stability in volume-based metrics (ICC = 0.993) and shape features. Overall, 36% (37/103) of all extracted features met excellent repeatability criteria.
Conclusion: Enhanced CBCT demonstrates excellent repeatability for shape-based radiomic features and select texture metrics, outperforming previously reported values for conventional CBCT imaging. While first-order and some texture features show greater variability, the repeatability of shape and volume measurements approach values near those of diagnostic CT. These findings support the feasibility of adaptive radiotherapy strategies based on repeatable radiomic feature changes, though larger externally validated studies are needed.