Parametric regression model approach for CT-based prediction of stopping power ratio for a Hounsfield look-up table

This study proposes a computed tomography (CT) number (CTN)-to-stopping power ratio (SPR) calibration method that can be directly combined with the established CTN-to-density conversion curves used in photon therapy. Furthermore, it evaluates the method's robustness against tissue variations and assesses dose differences compared with current calibration methods for scanned proton therapy. To develop the CTN-to-SPR calibration curve, relative electron densities (REDs) derived from a CIRS Model 062 M electron density phantom with tissue-substitute inserts were applied to a parametric regression model (PRM) constructed from 53 standard human tissues (including both male and female reference data from ICRP Publication 110). PRM stability was tested by varying the mass density and elemental composition of the reference tissues and recalculating SPRs from RED and CTN. For dose evaluation, we analyzed six representative patients and an expanded cohort of 30 additional patients (10 each for the head-and-neck, chest, and pelvis) whose CT datasets were retrospectively acquired at 80 or 120 kV. For the calibration, phantom CT datasets were acquired using Siemens SOMATOM X.cite and GE Revolution CT ES scanners at 100 kV. Treatment plans were recalculated using the PRM-derived Hounsfield look-up table (HLUT; CTN-to-SPR_PRM) without re-optimization. The dose impact and consistency were then evaluated against the current clinical standard (CTN-to-SPR_clin) using dose-volume histograms (DVHs). RED-based SPR was expressed asSPR=-0.0173ρe3+0.0057ρe2+1.0163ρe, (root-mean-square error (RMSE) ∼0.82% relative to the theoretical SPR for 53 ICRP reference tissues, 0.68% without air). The CTN-to-SPR_PRMand CTN-to-SPR_clincalibration curves showed consistent profiles; therefore, unified curves using averaged CTNs were applied due to the negligible differences between head and body phantom sizes. Compared to CTN-to-SPR_clin, the proposed CTN-to-SPR_PRMyielded higher SPR values in the lung and bone regions, but lower SPR values in the soft tissue regions. The relative differences between the two calibration curves were within 3% between -200 and +200 Hounsfield unit (HU). In the extreme density regions, the differences were >25% at -800 HU, and 1.23% and 0.11% at 1000 HU for the Siemens and GE scanners, respectively. Hydrogen and calcium variations caused the largest RMSE increases (1.58% and 6.00%), and trends were consistent across sexes. In the representative cases, DVH deviations were generally within 3%, except -5% in left parotid and -4.2% in lung Dmean. Across the expanded 30-case cohort, the PRM maintained robust target coverage and organs at risk sparing with minimal relative dose differences. Our findings suggest the potential of the CTN-to-SPR_PRMapproach for SPR estimation, providing a basis for discussing more flexible and streamlined CT calibration protocols.