Accuracy of Relative Electron Density and Stopping Power Ratio Estimation Using Dual-Energy Cone-Beam CT Virtual Monoenergetic Images
Abstract
Purpose
Accurate tissue characterization is critical for adaptive proton therapy. This study evaluates the performance of relative electron density (RED) and stopping power ratios (SPR) predicted by virtual monoenergetic images (VMIs) derived from dual-energy CBCT (DE-CBCT), benchmarking them against conventional DECT and Twin-Beam DECT (TB-DECT).
Methods
Two GAMMEX phantoms (RMI-467 and MECT-1472) and a CIRS head phantom were scanned using DECT, TB-DECT, and DE-CBCT. VMIs (40-150 keV) were generated via energy-dependent weighting factors. We implemented an image-based approach using standard electron density phantoms to estimate these factors, eliminating the need for specialized dual-material phantoms. Stoichiometric calibration was performed to derive Hounsfield look-up tables for RED and SPR conversion. An open-source tool was developed to automate phantom-based calibration, reducing manual workload.
Results
Phantom size significantly influenced CT number (CTN) stability due to scatter; high-density bone regions showed deviations of 100-200 HU, while soft tissue remained within 50 HU. Despite higher inherent noise in DE-CBCT, an 80/140 kVp acquisition pair effectively mitigated this. For VMIs ≥ 70 keV, DE-CBCT achieved SPR prediction accuracy comparable to TB-DECT. Notably, DE-CBCT yielded the lowest root-mean-square error (RMSE = 0.005) for SPR estimation, likely due to optimized calibration alignment between the phantom and imaging modality.
Conclusion
This study validates DE-CBCT-derived VMIs for accurate RED and SPR prediction. This approach offers a hardware-accessible alternative to DECT for adaptive radiotherapy, with potential for further AI-enhanced image optimization.