Exploiting advances in quantitative computed tomography to improve dose calculation in radiotherapy

<p dir="ltr">The work of this thesis had the overall aim to help improve dose-calculation accuracy in individualized radiotherapy treatment planning. To achieve this, we explored computed tomography (CT), using single-energy CT (SECT) and multi- energy CT (MECT) methods to derive patient material properties used for the treatment-planning dose calculations, and also compared results using analytical versus Monte Carlo (MC) dose algorithms. We focused on proton therapy as these aspects are generally more influential for charged-particle radiation, although some findings will transfer to radiotherapy more broadly. All standard energy-integrating CT scans were performed on a SOMATOM Definition AS+ scanner (Siemens Healthineers, Forcheim, Germany). The calibration phantom used was a 062M and the anthropomorphic head phantom was a 731-HN, both manufactured by CIRS (CIRS/SunNuclear, Norfolk, VA, USA).</p><p dir="ltr">In the first study, we evaluated three published and two newly proposed parametrization models for stoichiometric SECT calibration of CT number to proton stopping-power ratio (SPR). Each model was calibrated on the Definition scanner using the calibration phantom and 11 tissue substitute materials, two of which included trace amounts of barium (Z=56). Model performance was assessed under two scenarios: (1) standard calibrations for each model and protocol using the measured CT numbers, and (2) simulations of 10 000 calibrations per model and protocol where the input CT numbers were randomized around the measured values. Both scenarios were also repeated with virtual removal of barium content from the calibrators. The results showed that all models yielded similar outcomes in both scenarios when barium was excluded. However, when barium was present, differences emerged between the models. One model in particular exhibited larger SPR deviations in scenario 1 and greater variability in scenario 2 compared to the reference model. Notably, the two newly proposed models demonstrated the least sensitivity to the presence of barium across both scenarios, indicating improved robustness to trace high-Z elements in calibration materials.</p><p dir="ltr">Noisy dual-energy CT (DECT) images have been reported to compromise the accuracy of proton dose calculations; the second study aimed to develop an iterative DECT reconstruction algorithm that simultaneously produces accurate material properties and supresses image noise. A 140 kV/80 kV protocol combination was used on the Definition scanner, with each protocol calibrated using the same phantom and methods as in study 1. We solved for relative electron density (RED) and effective atomic number (EAN), from which proton SPR could be derived, denoting the model K-DECT. The iterative algorithm developed to solve K-DECT, denoted prior-image constrained denoising (PIC-D), incorporated penalty functions to supress image noise. PIC-D demonstrated efficient noise suppression, and with well-tuned penalties, introduced only minor bias in absolute quantities - evaluated using both the calibration phantom and the anthropomorphic head phantom.</p><p dir="ltr">Study 3 investigated whether first-generation clinical photon-counting CT (PCCT) could improve material characterization in a human-like head geometry compared with DECT and SECT. Both the calibration phantom and the anthropomorphic head phantom were scanned on a NAEOTOM Alpha (Siemens Healthineers Forchheim, Germany) PCCT scanner as well as on the Definition scanner using SECT and DECT protocols. A method was developed to derive RED/EAN from a set of eight virtual monoenergetic images (VMIs). For PCCT and DECT, SPR was computed from RED/EAN, while SECT relied on a stoichiometric look-up table. The results showed that PCCT generally produced narrower pixel value distributions within each material and enabled superior material characterization compared to both DECT and SECT - though this was not consistent across all combinations of quantities and phantom materials. Additionally, the findings suggested that SPR may not be the most suitable metric for evaluating performance in image-derived material properties.</p><p dir="ltr">In Study 4, we evaluated the dosimetric impact of two types of cranial titanium implants on proton beam dose distributions, with two primary aims: (1) to quantify the limitations of an analytic dose-calculation algorithm in a clinical treatment planning system (TPS) when implants are present, and (2) to assess the impact when the same algorithm uses SPR maps derived from standard clinical SECT images to represent the test geometry. The test geometry consisted of the calibration phantom combined with physical samples of two implants - a CranioFix (Aesculap AG, Tuttlingen, Germany) and a titanium mesh implant (OssDsign, Uppsala, Sweden) - scanned using SECT on the Definition scanner. A virtual ground-truth geometry was constructed using CAD models of the implants, with material properties derived from tabulated compositions. Proton treatment plans were generated and calculated using the pencil beam algorithm in the Eclipse v16.1 TPS (Varian, Siemens Healthineers, Palo Alto, USA) for both the ground-truth and CT-based geometries. Additionally, Monte Carlo simulations of the plans were performed using TOPAS for the ground-truth geometry. The results showed that the Eclipse pencil beam algorithm failed to accurately model dose distributions in the shadow regions of the implants, as validated against the Monte Carlo results. Dose inhomogeneities were consistently underestimated for both implant types, and a shorter end-of-range was predicted behind the CranioFix. Furthermore, poor CT representation of the implants, and coarse dose-matrix resolution, significantly compromised the accuracy of TPS-based dose calculations.</p><h3>List of scientific papers</h3><p dir="ltr">I. Ödén J, <b>Zimmerman J,</b> Poludniowski G. Comparison of CT-number parameterization models for stoichiometric CT calibration in proton therapy. Physica Medica-European Journal of Medical Physics. 2018 Mar;47:42-9. <a href="https://doi.org/10.1016/j.ejmp.2018.02.016" rel="noreferrer" target="_blank">https://doi.org/10.1016/j.ejmp.2018.02.016</a></p><p dir="ltr">II. <b>Zimmerman J,</b> Thor D, Poludniowski G. Stopping-power ratio estimation for proton radiotherapy using dual-energy computed tomography and prior-image constrained denoising. Medical Physics. 2023;50(3):1481-95. <a href="https://doi.org/10.1002/mp.16063" rel="noreferrer" target="_blank">https://doi.org/10.1002/mp.16063</a></p><p dir="ltr">III. <b>Zimmerman J,</b> Poludniowski G. Assessment of Photon-Counting Computed Tomography for Quantitative Imaging in Radiation Therapy. International Journal of Radiation Oncology*Biology*Physics. 2025 Apr 1;121(5):1316-27. <a href="https://doi.org/10.1016/j.ijrobp.2024.11.069" rel="noreferrer" target="_blank">https://doi.org/10.1016/j.ijrobp.2024.11.069</a></p><p dir="ltr">IV. <b>Zimmerman J,</b> Carlsson Tedgren Å, Poludniowski G. The influence of metal cranial implants on proton therapy dose distributions: a simulation study. [Manuscript]</p>