Introduction Combined thermal and fluid modeling is useful for design and optimization of cyclotron water targets. Previous heat transfer models assumed either a distribution of void under saturation conditions [1] or a static volumetric heat distribution [2]. This work explores the coupling of Monte Carlo radiation transport and Computation Fluid Dynamics (CFD) software in a computational model of the BTI Targetry visualization target [3]. In a batch water target, as the target medium is heated by energy deposition from the proton beam, a non-uniform density distribution develops. Production target operation is ultimately limited by the range thickness of the target un-der conditions of reduced water density. Since proton range is a function of target density, the system model must include the corresponding change in the volumetric heat distribution. As an initial attempt to couple the radiation transport and fluid dynamics calculations, the scope of this work was limited to subcooled target conditions. With the increasing availability of multi-phase CFD capabilities, this work provides the basis for extending these calculations to boiling targets where the coupling of the radiation transport and fluid dynamics is expected to be much stronger. Material and Methods The Monte Carlo radiation transport code MCNPX was used to create energy deposition data tallies from proton interaction with the target water and beam window. The beam was modeled as a Gaussian distribution with 50% transmission through a 10 mm diameter collimator. The energy deposition tally was translated into a 3-dimensional, point-wise heat generation table and supplied as an input to the CFD code ANSYS CFX. An iterative method was developed to couple the volumetric heat distribution from MCNPX to the fluid density distribution computed within ANSYS CFX. A 3-dimensional table of water density was exported from ANSYS CFX and imported into MCNPX. MCNPX was then used to calculate the heat generation rate (due to proton interactions) based on the assumed density profile. Applying the new heat generation profile to the ANSYS CFX model resulted in changes to the beam shape and penetration depth. The iterative scheme continued until converged values for density and heat generation rate were achieved. Monte Carlo methods are computationally ex-pensive due to the large number of particle histories needed to generate accurate results. CFD simulations are also computationally expensive due to the large number of mesh elements needed. Optimization methods were used for both MCNPX and ANSYS CFX to result in achievable solution times and memory requirements. Local mesh refinement in the beam strike area was necessary for convergence. This was achieved by extending the boundary layer of the mesh within the target water domain deeper into the fluid. This allowed for better resolution within the beam strike area without significantly increasing the expense in the remainder of the fluid domain. Additionally, direct simulation of the cooling water domain was decoupled from the computational model during the iterative process. Heat transfer coefficients from the first iteration were applied as a boundary condition for subsequent iterations. Once the beam and density distributions reached convergence, the beam data was applied to a high fidelity “full” model, which included the cooling water domain as well as increased particle histories in MCNPX. Results and Conclusions The target was initially modeled assuming a 10 μA beam of 18 MeV protons into uniform density target water with operating pressure of 400 psi. These conditions resulted in predicted maximum temperatures below the saturation temperature. The final converged beam data was compared to the original (uniform density) beam data. As expected, the density-dependent beam penetrates farther into the target water than when a uniform density is assumed. The density-dependent beam has a broader Bragg peak region with a lower maximum heat generation rate than the original beam. A line plot of the volumetric heat generation rate through the center of the beam is shown in FIG. 2. Even though the maximum volumetric heat generation rate was lower, the density-dependent beam resulted in a higher maximum fluid temperature. Experiments were performed with the visualization target on an IBA 18/9 cyclotron, and video was recorded for a range of target operating conditions. Analysis of the video recordings from the experiment gives a peak fluid velocity in the target chamber of roughly 5–10 centimeters per second with a 10 A beam current. The velocities predicted by the CFD model are within the same range. There is also good agreement be-tween proton beam range between the experiment and model. The effective proton range can be seen in FIGURES 3 and 4. Future work will include applying the coupling technique for two-phase boiling conditions and to gas targets. If successful, this method should be a powerful tool for design and optimization of liquid and gas targets.

Introduction Combined thermal and fluid modeling is useful for design and optimization of cyclotron water targets. Previous heat transfer models assumed either a distribution of void under saturation conditions [1] or a static volumetric heat distribution [2]. This work explores the coupling of Monte Carlo radiation transport and Computation Fluid Dynamics (CFD) software in a computational model of the BTI Targetry visualization target [3]. In a batch water target, as the target medium is heated by energy deposition from the proton beam, a non-uniform density distribution develops. Production target operation is ultimately limited by the range thickness of the target un-der conditions of reduced water density. Since proton range is a function of target density, the system model must include the corresponding change in the volumetric heat distribution. As an initial attempt to couple the radiation transport and fluid dynamics calculations, the scope of this work was limited to subcooled target conditions. With the increasing availability of multi-phase CFD capabilities, this work provides the basis for extending these calculations to boiling targets where the coupling of the radiation transport and fluid dynamics is expected to be much stronger. Material and Methods The Monte Carlo radiation transport code MCNPX was used to create energy deposition data tallies from proton interaction with the target water and beam window. The beam was modeled as a Gaussian distribution with 50% transmission through a 10 mm diameter collimator. The energy deposition tally was translated into a 3-dimensional, point-wise heat generation table and supplied as an input to the CFD code ANSYS CFX. An iterative method was developed to couple the volumetric heat distribution from MCNPX to the fluid density distribution computed within ANSYS CFX. A 3-dimensional table of water density was exported from ANSYS CFX and imported into MCNPX. MCNPX was then used to calculate the heat generation rate (due to proton interactions) based on the assumed density profile. Applying the new heat generation profile to the ANSYS CFX model resulted in changes to the beam shape and penetration depth. The iterative scheme continued until converged values for density and heat generation rate were achieved. Monte Carlo methods are computationally ex-pensive due to the large number of particle histories needed to generate accurate results. CFD simulations are also computationally expensive due to the large number of mesh elements needed. Optimization methods were used for both MCNPX and ANSYS CFX to result in achievable solution times and memory requirements. Local mesh refinement in the beam strike area was necessary for convergence. This was achieved by extending the boundary layer of the mesh within the target water domain deeper into the fluid. This allowed for better resolution within the beam strike area without significantly increasing the expense in the remainder of the fluid domain. Additionally, direct simulation of the cooling water domain was decoupled from the computational model during the iterative process. Heat transfer coefficients from the first iteration were applied as a boundary condition for subsequent iterations. Once the beam and density distributions reached convergence, the beam data was applied to a high fidelity “full” model, which included the cooling water domain as well as increased particle histories in MCNPX. Results and Conclusions The target was initially modeled assuming a 10 μA beam of 18 MeV protons into uniform density target water with operating pressure of 400 psi. These conditions resulted in predicted maximum temperatures below the saturation temperature. The final converged beam data was compared to the original (uniform density) beam data. As expected, the density-dependent beam penetrates farther into the target water than when a uniform density is assumed. The density-dependent beam has a broader Bragg peak region with a lower maximum heat generation rate than the original beam. A line plot of the volumetric heat generation rate through the center of the beam is shown in FIG. 2. Even though the maximum volumetric heat generation rate was lower, the density-dependent beam resulted in a higher maximum fluid temperature. Experiments were performed with the visualization target on an IBA 18/9 cyclotron, and video was recorded for a range of target operating conditions. Analysis of the video recordings from the experiment gives a peak fluid velocity in the target chamber of roughly 5–10 centimeters per second with a 10 A beam current. The velocities predicted by the CFD model are within the same range. There is also good agreement be-tween proton beam range between the experiment and model. The effective proton range can be seen in FIGURES 3 and 4. Future work will include applying the coupling technique for two-phase boiling conditions and to gas targets. If successful, this method should be a powerful tool for design and optimization of liquid and gas targets.

Die Protonentherapie zeichnet sich durch steile Dosisgradienten und damit einen gut lokalisierbaren Energieübertrag aus. Um dieses Potential voll ausschöpfen zu können, werden weltweit Möglichkeiten erforscht, die Dosisdeposition und insbesondere die Reichweite der Protonen im Patienten zu verifizieren. Eine vielversprechende, erst im letzten Jahrzehnt entdeckte Methode ist das Prompt Gamma-Ray Timing (PGT), das auf der Abhängigkeit der detektierten Flugzeitverteilung prompter Gammastrahlung von der Transitzeit der Protonen im Patienten beruht. In dieser Arbeit wird eine Geant4-Simulation zur Vorhersage der PGT-Spektren bei Bestrahlung eines PMMA-Phantoms entwickelt und durch den Vergleich mit experimentellen Daten validiert. Sowohl die Emissionsausbeute prompter Gammastrahlung im Phantom als auch die Detektionsrate werden abhängig von der Protonenenergie analysiert. Zur Vergleichbarkeit mit den gemessenen Spektren wird eine mehrschrittige Prozessierung der Simulationsergebnisse vorgestellt. Schließlich wird die Simulation genutzt, um die Sensitivität der PGT-Methode auf Reichweitenänderungen zu demonstrieren. Dafür können in das Phantom Cavitäten unterschiedlicher Dicke und verschiedenen Materials eingefügt werden. Für geeignet gewählte Verteilungsparameter der simulierten PGT-Spektren wird deren detektierte Änderung mit der bekannten induzierten Reichweitenänderung ins Verhältnis gesetzt. Die so bestimmte Sensitivität ist mit früheren Ergebnissen für gemessene Spektren im Rahmen der Unsicherheiten in Übereinstimmung.:1 Einleitung und Motivation 1 2 Theoretische Grundlagen 5 2.1 Wechselwirkung von Protonen mit Materie 5 2.1.1 Bethe-Bloch-Gleichung 6 2.1.2 Reichweite im CSDA-Modell 9 2.1.3 Tiefendosiskurve und Bragg-Peak 10 2.2 Prompt Gamma-Ray Timing 11 2.2.1 Emission prompter Gammastrahlung 11 2.2.2 Korrelation zur Protonen-Reichweite und Dosisdeposition 11 2.2.3 Idee des Prompt Gamma-Ray Timings 14 3 Material und Methoden 17 3.1 Dresdner IBA-Protonentherapie 17 3.1.1 Beschleunigungsprinzip des Isochronzyklotrons 17 3.1.2 Zeitliche Struktur der Protonen-Pakete 18 3.2 Teilchentransportrechnungen mit Geant4 20 3.3 PLD-Format für Pencil-Beam-Scanning-Pläne 21 3.3.1 Geometrische Definition der Spots 21 3.3.2 Dosimetrische Definition der Spots 23 3.3.3 Verwendete Bestrahlungspläne 24 3.4 Messaufbau zur experimentellen Validierung 26 3.4.1 Target – PMMA-Phantom mit verschiedenen Cavitäten 27 3.4.2 Detektoren – CeBr3-Szintillatoren mit Photomultipliern 27 4 Ergebnisse und Diskussion 29 4.1 Simulierte Emission prompter Gammastrahlung 29 4.1.1 Simulierte Emissionsspektren 29 4.1.2 Simulierte Emissionsprofile 30 4.1.3 Totale Emissionsausbeute 31 4.2 Simulierte Detektion prompter Gammastrahlung 33 4.2.1 Detektionsrate und Raumwinkeleffekt 33 4.2.2 Simulierte PGT-Spektren 35 4.2.3 Simulierte Energiespektren 37 4.3 Vergleich simulierter und gemessener Spektren 39 4.3.1 Nachverarbeitung der Simulationsergebnisse 40 4.3.2 Auswahl des Energiefensters 45 4.3.3 Empirisches Modell zur Beschreibung der Zeitspektren 47 4.3.4 Diskussion systematischer Abweichungen 49 4.4 Sensitivität der Simulation gegenüber induzierten Reichweitenänderungen 51 5 Zusammenfassung und Ausblick 59 Anhang 61 A Parameter des Messaufbaus 61 B Angepasste Modellparameter aus Abbildung 4.12 62 C Sensitivität auf Reichweitenänderung bei 162 MeV 63 Literaturverzeichnis 69 / Proton therapy is characterized by steep dose gradients and thus a well-localizable energy transfer. To fully harness this potential, possibilities are being explored worldwide to verify the dose deposition and especially the range of protons in the patient. A promising method discovered only in the last decade is prompt gamma-ray timing (PGT), which relies on the dependence of the detected time-of-flight distribution of prompt gamma radiation on the transit time of protons in the patient. In this study, a Geant4 simulation is developed to predict PGT spectra during irradiation of a PMMA phantom and validated by comparison with experimental data. Both the emission yield of prompt gamma radiation in the phantom and the detection rate are analyzed depending on the proton energy. For comparability with the measured spectra, a multi-step processing of the simulated results is presented. Finally, the simulation is used to demonstrate the sensitivity of the PGT method to changes in range. For this purpose, cavities of different thicknesses and materials can be inserted into the phantom. For appropriately chosen distribution parameters of the simulated PGT spectra, their detected change is compared to the known induced change in range. The sensitivity determined in this way is consistent with previous results for measured spectra within the uncertainties.:1 Einleitung und Motivation 1 2 Theoretische Grundlagen 5 2.1 Wechselwirkung von Protonen mit Materie 5 2.1.1 Bethe-Bloch-Gleichung 6 2.1.2 Reichweite im CSDA-Modell 9 2.1.3 Tiefendosiskurve und Bragg-Peak 10 2.2 Prompt Gamma-Ray Timing 11 2.2.1 Emission prompter Gammastrahlung 11 2.2.2 Korrelation zur Protonen-Reichweite und Dosisdeposition 11 2.2.3 Idee des Prompt Gamma-Ray Timings 14 3 Material und Methoden 17 3.1 Dresdner IBA-Protonentherapie 17 3.1.1 Beschleunigungsprinzip des Isochronzyklotrons 17 3.1.2 Zeitliche Struktur der Protonen-Pakete 18 3.2 Teilchentransportrechnungen mit Geant4 20 3.3 PLD-Format für Pencil-Beam-Scanning-Pläne 21 3.3.1 Geometrische Definition der Spots 21 3.3.2 Dosimetrische Definition der Spots 23 3.3.3 Verwendete Bestrahlungspläne 24 3.4 Messaufbau zur experimentellen Validierung 26 3.4.1 Target – PMMA-Phantom mit verschiedenen Cavitäten 27 3.4.2 Detektoren – CeBr3-Szintillatoren mit Photomultipliern 27 4 Ergebnisse und Diskussion 29 4.1 Simulierte Emission prompter Gammastrahlung 29 4.1.1 Simulierte Emissionsspektren 29 4.1.2 Simulierte Emissionsprofile 30 4.1.3 Totale Emissionsausbeute 31 4.2 Simulierte Detektion prompter Gammastrahlung 33 4.2.1 Detektionsrate und Raumwinkeleffekt 33 4.2.2 Simulierte PGT-Spektren 35 4.2.3 Simulierte Energiespektren 37 4.3 Vergleich simulierter und gemessener Spektren 39 4.3.1 Nachverarbeitung der Simulationsergebnisse 40 4.3.2 Auswahl des Energiefensters 45 4.3.3 Empirisches Modell zur Beschreibung der Zeitspektren 47 4.3.4 Diskussion systematischer Abweichungen 49 4.4 Sensitivität der Simulation gegenüber induzierten Reichweitenänderungen 51 5 Zusammenfassung und Ausblick 59 Anhang 61 A Parameter des Messaufbaus 61 B Angepasste Modellparameter aus Abbildung 4.12 62 C Sensitivität auf Reichweitenänderung bei 162 MeV 63 Literaturverzeichnis 69

Derzeitige Reichweiteunsicherheiten in der Protonentherapie verhindern das vollständige Ausschöpfen ihrer physikalischen Vorteile. Ein wesentlicher Anteil ist dabei auf die Vorhersage der Reichweite mittels Röntgen-Computertomographie (CT) zurückzuführen. Um die CT-bezogene Unsicherheit zu verringern, wird die Zwei-Spektren-Computertomographie (DECT) als vielversprechend angesehen. Innerhalb dieser Arbeit wurde die Anwendbarkeit von DECT in der Protonentherapie untersucht. Zunächst wurde ein CT-Scanprotokoll für die Strahlentherapie hinsichtlich Bildqualität und Konstanz der CT-Zahlen für verschiedene Körperregionen und -größen optimiert. Anschließend wurde die patientenindividuelle DECT- basierte Reichweitevorhersage kalibriert und ihre Genauigkeit in zwei Experimenten mit bekannter Referenz unter Verwendung eines anthropomorphen Phantoms und von homogenen biologischen Geweben verifiziert. Die klinische Relevanz von DECT wurde in einer retrospektiven Analyse von Krebspatienten mit Tumoren im Kopf, Becken oder Thorax nachgewiesen. Die systematischen Reichweiteunterschiede zwischen DECT und dem klinischen Standardverfahren konnten durch die Optimierung der Standardmethode basierend auf zusätzlichen mit DECT erworbenen Patienteninformationen reduziert werden. Somit wurde DECT erstmalig klinisch genutzt, um die Reichweiteberechnung zu verbessern. Die patientenindividuelle DECT-basierte Reichweitevorhersage kann zusätzlich Gewebevariabilitäten innerhalb eines und zwischen Patienten berücksichtigen, wie für Kopftumorpatienten gezeigt wurde. Dies legt den Grundstein für eine genauere Reichweiteberechnung und eröffnet neue Möglichkeiten für die Reduktion klinischer Sicherheitssäume, in denen die CT-bezogenen Unsicherheiten berücksichtigt sind.:1 Introduction 2 Physical Principles of Computed Tomography 2.1 Image Acquisition 2.2 Image Reconstruction 2.3 Dual-Energy Computed Tomography 3 Physical Principles of Proton Therapy 3.1 Treatment Techniques 3.2 Uncertainties in Proton Therapy 4 Principles of Stopping-Power Prediction from Computed Tomography 4.1 Single-Energy Computed Tomography 4.2 Dual-Energy Computed Tomography 5 Experimental Calibration of Stopping-Power Prediction 5.1 Scan Protocol Optimisation in Computed Tomography 5.2 Characterisation of Pseudo-Monoenergetic CT Calculation 5.3 Determination of Proton Stopping Power 5.4 Calibration of Stopping-Power Prediction Methods 6 Experimental Verification of Stopping-Power Prediction 6.1 Anthropomorphic Head Phantom 6.2 Homogeneous Biological Tissue Samples 7 Clinical Translation and Validation of Dual-Energy Computed Tomography 7.1 Feasibility of Dual-Spiral Dual-Energy CT 7.2 Range Prediction in Cerebral and Pelvic Tumour Patients 7.3 Tissue Variability in Brain-Tumour Patients 7.4 Feasibility of 4D Dual-Spiral Dual-Energy CT 7.5 DECT-Based Refinement of the Hounsfield Look-Up Table 8 Summary 9 Zusammenfassung / Range uncertainty in proton therapy currently hampers the full exploitation of its physical advantages. A substantial amount of this uncertainty arises from proton range prediction based on X-ray computed tomography (CT). Dual-energy CT (DECT) has often been suggested as a promising imaging modality to reduce this CT-related range uncertainty. Within this thesis, the translation of DECT into application in proton therapy was evaluated. First, a CT scan protocol was optimised for radiotherapy considering the image quality and CT number stability for various body regions and sizes. The patient-specific DECT-based range prediction was then calibrated and its accuracy validated in two ground-truth experiments using an anthropomorphic phantom and homogeneous biological tissues. Subsequently, the clinical relevance of DECT was demonstrated in a retrospective cohort analysis of cerebral, pelvic and thoracic tumour patients. The systematic range deviations between the DECT and state-of-the-art approach were then reduced by adapting the standard method utilizing additional patient information obtained from DECT. Hence, DECT was clinically applied for the first time to refine proton range calculation. As a further step, the use of patient-specific DECT-based range prediction also considers intra- and inter-patient tissue variabilities as quantified in brain-tumour patients. A future implementation will be an important cornerstone to improve proton range calculation and might open up the possibility to reduce clinical safety margins accounting for the CT-related range uncertainty.:1 Introduction 2 Physical Principles of Computed Tomography 2.1 Image Acquisition 2.2 Image Reconstruction 2.3 Dual-Energy Computed Tomography 3 Physical Principles of Proton Therapy 3.1 Treatment Techniques 3.2 Uncertainties in Proton Therapy 4 Principles of Stopping-Power Prediction from Computed Tomography 4.1 Single-Energy Computed Tomography 4.2 Dual-Energy Computed Tomography 5 Experimental Calibration of Stopping-Power Prediction 5.1 Scan Protocol Optimisation in Computed Tomography 5.2 Characterisation of Pseudo-Monoenergetic CT Calculation 5.3 Determination of Proton Stopping Power 5.4 Calibration of Stopping-Power Prediction Methods 6 Experimental Verification of Stopping-Power Prediction 6.1 Anthropomorphic Head Phantom 6.2 Homogeneous Biological Tissue Samples 7 Clinical Translation and Validation of Dual-Energy Computed Tomography 7.1 Feasibility of Dual-Spiral Dual-Energy CT 7.2 Range Prediction in Cerebral and Pelvic Tumour Patients 7.3 Tissue Variability in Brain-Tumour Patients 7.4 Feasibility of 4D Dual-Spiral Dual-Energy CT 7.5 DECT-Based Refinement of the Hounsfield Look-Up Table 8 Summary 9 Zusammenfassung

La proton thérapie est une modalité de traitement du cancer qu’utilise des faisceaux de protons. Les systèmes de planification de traitement actuels se basent sur une image de l’anatomie du patient acquise par tomodensitométrie. Le pouvoir d’arrêt des protons relatif à l’eau (Stopping Power Ratio en Anglais, SPR) est déterminé à partir des unités Hounsfield (Hounsfield Units en Anglais, HU) pour calculer la dose absorbée au patient. Les protons sont plus vulnérables que les photons aux modifications du SPR du tissu dans la direction du faisceau dues au mouvement, désalignement ou changements anatomiques. De plus, les inexactitudes survenues de la CT de planification et intrinsèques à la conversion HU-SPR contribuent énormément à l’incertitude de la portée des protons. Dans la pratique clinique, au volume de traitement s’ajoutent des marges de sécurité pour tenir en compte ces incertitudes en détriment de perdre la capacité d’épargner les tissus autour de la tumeur. L’usage de l’imagerie bi-énergie en proton thérapie a été proposé pour la première fois en 2009 pour mieux estimer le SPR du patient par rapport à l’imagerie mono-énergie. Le but de cette thèse est d’étudier la potentielle amélioration de l’estimation du SPR des protons en utilisant l’imagerie bi-énergie, pour ainsi réduire l’incertitude dans la prédiction de la portée des protons dans le patient. Cette thèse est appliquée à un nouveau système d’imagerie, l’Imaging Ring (IR), un scanner de tomodensitométrie conique (Cone-Beam CT en Anglais, CBCT) développé pour la radiothérapie guidée par l’image. L’IR est équipé d’une source de rayons X avec un système d’alternance rapide du voltage, synchronisé avec une roue contenant des filtres de différents matériaux que permet des acquisitions CBCT multi-énergie. La première contribution est une méthode pour calibrer les modèles de source et la réponse du détecteur pour être utilisés en simulations d’imagerie X. Deuxièmement, les recherches ont évalué les facteurs que peuvent avoir un impact sur les résultats du procès de décomposition bi-énergie, dès paramètres d’acquisition au post-traitement. Les deux domaines, image et basée en la projection, ont été minutieusement étudiés, avec un spéciale accent aux approches basés en la projection. Deux nouvelles bases de décomposition ont été proposées pour estimer le SPR, sans avoir besoin d’une variable intermédiaire comme le nombre atomique effectif. La dernière partie propose une estimation du SPR des fantômes de caractérisation tissulaire et d’un fantôme anthropomorphique à partir d’acquisitions avec l’IR. Il a été implémentée une correction du diffusé, et il a été proposée une routine pour interpoler linéairement les sinogrammes de basse et haute énergie des acquisitions bi-énergie pour pouvoir réaliser des décompositions en matériaux avec données réelles. Les valeurs réconstruits du SPR ont été comparées aux valeurs du SPR expérimentales déterminés avec un faisceau d’ions de carbone. / Proton therapy is a promising radiation treatment modality that uses proton beams to treat cancer. Current treatment planning systems rely on an X-ray computed tomography (CT) image of the patient's anatomy to design the treatment plan. The proton stopping-power ratio relative to water (SPR) is derived from CT numbers (HU) to compute the absorbed dose in the patient. Protons are more vulnerable than photons to changes in tissue SPR in the beam direction caused by movement, misalignment or anatomical changes. In addition, inaccuracies arising from the planning CT and intrinsic to the HU-SPR conversion greatly contribute to the proton range uncertainty. In clinical practice, safety margins are added to the treatment volume to account for these uncertainties at the expense of losing organ-sparing capabilities. The use of dual-energy (DE) in proton therapy was first suggested in 2009 to better estimate the SPR with respect to single-energy X-ray imaging. The aim of this thesis work is to investigate the potential improvement in determining proton SPR using DE to reduce the uncertainty in predicting the proton range in the patient. This PhD work is applied to a new imaging device, the Imaging Ring (IR), which is a cone-beam CT (CBCT) scanner developed for image-guided radiotherapy (IGRT). The IR is equipped with a fast kV switching X-ray source, synchronized with a filter wheel, allowing for multi-energy CBCT imaging. The first contribution of this thesis is a method to calibrate a model for the X-ray source and the detector response to be used in X-ray image simulations. It has been validated experimentally on three CBCT scanners. Secondly, the investigations have evaluated the factors that have an impact on the outcome of the DE decomposition process, from the acquisition parameters to the post-processing. Both image- and projection-based decomposition domains have been thoroughly investigated, with special emphasis on projection-based approaches. Two novel DE decomposition bases have been proposed to estimate proton SPRs, without the need for an intermediate variable such as the effective atomic number. The last part of the thesis proposes an estimation of proton SPR maps of tissue characterization and anthropomorphic phantoms through DE-CBCT acquisitions with the IR. A correction for X-ray scattering has been implemented off-line, and a routine to linearly interpolate low-energy and high-energy sinograms from sequential and fast-switching DE acquisitions has been proposed to perform DE material decomposition in the projection domain with real data. DECT-derived SPR values have been compared with experimentally-determined SPR values in a carbon-ion beam.