Comparison of CT-number parameterization models for stoichiometric CT calibration in proton therapy

•A novel X-ray spectrum based model for use in the stoichiometric CT calibration.•Intra-model comparison and robustness evaluation of five parameterization models.•All models perform comparably well if tissue substitutes contain elements with Z ≤ 20.•High-Z content may affect calibration accuracy and robustness for some models. This study compares the predictions of three parameterization models used in previously published works, implementing the stoichiometric CT calibration for proton therapy, and a further two alternative parameterizations suggested here. Stoichiometric calibrations of patient CT-number to stopping-power ratio (SPR) were performed for four CT protocols using tissue substitutes supplied by CIRS (CIRS Inc., Norfolk, VA, USA). To evaluate robustness of the five models (Sch96/Sch00/Mar12/Karol/Spek), the calibration was repeatedly simulated by randomly perturbing the measured CT-numbers of the tissue substitutes (1σ:10 HU). The impact of high-Z content was assessed through calibrations where the two substitutes with barium content were replaced by hypothetical materials without barium. The stoichiometric calibrations generally agreed within 1% between the models, for non-bony tissues. For higher CT-numbers, a well-known 2-parameter model (Sch00) generated larger SPRs compared to the other models, with inter-model discrepancies of up to 3%. The 95% coverage interval of the calibrations obtained from the robustness analysis varied substantially. The well-known 2- and 3-parameter models (Sch00/Sch96) had the largest intervals. However, the partly-hypothetical (i.e. no barium) input data generated calibrations that agreed within 1% over the whole CT scale for all models and improved the 95% coverage interval of the well-known models (Sch00/Sch96). All parameterization models performed comparably if the scanned materials only contained elements with Z ≤ 20. However, the two alternative models proposed here (Karol/Spek), together with a previously published 1-parameter model (Mar12), generated robust calibrations in close agreement even when tissue substitutes contain elements with higher atomic number.