Evaluation of Radiomic Feature Repeatability from a Novel CBCT System
Abstract
Purpose
The novel Varian TrueBeam HyperSight-CBCT (HS-CBCT) aims to overcome image non-uniformity and artifacts in conventional CBCT. While radiomic feature repeatability has been investigated for conventional CBCT, the performance of HS-CBCT remains under-evaluated. This study investigated the intra-day and inter-day repeatability of radiomic features in on-board volumetric images acquired with HS-CBCT in comparison with CT.
Methods
A CT-ED phantom (Sun Nuclear) equipped with 16 tissue-equivalent inserts was utilized. Images were acquired in triplicate for intra-day analysis and repeated on different days for inter-day analysis. Both HS-CBCT and CT were scanned at 100 and 120 kV. For HS-CBCT, to evaluate reconstruction dependency, three tube current settings and three reconstruction algorithms were applied: iterative CBCT (iCBCT), fast iCBCT, and Feldkamp (FDK). A total of 837 radiomic features, including original and wavelet-filter-based features, were extracted per the 16 inserts. Feature stability was quantified using the intraclass correlation coefficient (ICC) and coefficient of variation (CV). Features with ICC ≥ 0.90 were defined as highly stable, and CV < 5% indicated high reproducibility.
Results
Focusing on the clinical 120 kV protocol and iCBCT, inter-day reproducibility was superior in CT compared to HS-CBCT (mean ICC: 0.84 vs. 0.72; median CV: 4.06% vs. 4.97%). Consequently, the proportion of highly repeatable features (ICC ≥ 0.90) was higher in CT (62.4%) than in HS-CBCT (45.2%). Regarding intra-day repeatability, while HS-CBCT demonstrated excellent stability with a lower median CV (2.43%) than CT (2.93%), CT retained higher overall feature reliability (mean ICC: 0.86 vs. 0.74; high-repeatability: 70.0% vs. 50.0%). Trends at 100 kV were consistent with these findings.
Conclusion
HS-CBCT (iCBCT) demonstrated high intra-day feature stability comparable to CT, suggesting sufficient reliability for daily trend analysis. However, inter-day reproducibility was inferior to CT. These findings suggest that while TB-HS is promising, rigorous feature pre-selection is essential for longitudinal delta-radiomics studies.