Monte Carlo particle transport codes for ion beam therapy treatmentplanning – Validation, development and applications

Till Tobias Böhlen

Monte Carlo particle transport codes forion beam therapy treatment planningValidation, development and applications

Till Tobias Böhlen

Sammanfattning

Extern strålbehandling med proton- och jonstrålar kräver noggranna verktygför dosimetrisk karakterisering av strålfälten. Monte Carlo (MC)-program förpartikeltransport, t.ex. FLUKA och GEANT4, kan utgöra en värdefull metodför att öka noggrannheten i dosberäkningar och för att stödja olika aspekter avjonstrålterapi, som dosplanering och behandlingsövervakning. En av förutsät-tningarna för sådana tillämpningar är att MC-programmen kan modellera rel-evanta fysikaliska processer tillförlitligt och noggrant. Som ett första fokus iavhandlingen utvecklas och utvärderas de fysikaliska modellerna för jonstrål-terapi i MC-programmen och valideras mot experimentella data. Lämpligamodeller och konfigurationer i MC-koderna för jonstrålbehandling kan därmedfastställas. Noggrannheten i hur MC-koderna FLUKA och GEANT4 beskrivernukleära fragmenteringsprocesser och produktion av sekundära laddade kärn-fragment undersöks för strålbehandling med koljoner. Förmågan hos MC-koden FLUKA att beskriva egenskaperna hos blandade strålfält skapade avjonstrålar studeras också genom att simulera mikrodosimetriska storheter hosjonstrålar och sedan jämföra med experimentella data. En korrekt beskrivn-ing av mikrodosimetriska storheter är viktig för att förutsäga den biologiskaeffektiviteten hos partikelstrålar.

Vidare har två modeller som beskriver Compton-spridning och transport avtvå gammafotoner skapade genom annihilering av positroner i vila utvecklats,validerats och integrerats i FLUKA modellerna. En detaljerad beskrivning avbåda dessa processer är viktig för en noggrann simulering av positronemis-sionstomografi (PET) och prompt gamma-avbildning. Dessa tekniker utgör al-ternativ för att, i klinisk drift, övervaka dosfördelningen vid cancerbehandlingmed jonstrålar.

Den andra inriktningen av avhandlingen är att bidra till utvecklingen avett MC-dosplaneringsverktyg för protoner och joner med atomnummer Z ≤ 8,baserat på FLUKA. Till skillnad från tidigare kliniska FLUKA-baserade MCimplementationer för jonterapi där redan färdiga dosplaner simuleras, möjlig-gör det utvecklade verktyget att optimering av absorberad dos och RBE-viktaddos görs samtidigt. I avhandlingen undersöks robustheten i jonstrålfält medavseende på osäkerheter i RBE-värden, samtidigt som olika optimeringsstrate-gier jämförs.

Abstract

External radiotherapy with proton and ion beams needs accurate tools for thedosimetric characterization of treatment fields. Monte Carlo (MC) particletransport codes, such as FLUKA and GEANT4, can be a valuable method toincrease accuracy of dose calculations and to support various aspects of ionbeam therapy, such as treatment planning and treatment monitoring. One ofthe prerequisites for such applications is however that the MC codes are ableto model reliably and accurately the relevant physics processes. As a first focusof this thesis work, physics models of MC codes with importance for ion beamtherapy are developed and validated with experimental data. As a result suit-able models and code configurations for applications in ion beam therapy areestablished. The accuracy of the MC codes FLUKA and GEANT4 in describ-ing nuclear fragmentation processes and the production of secondary chargednuclear fragments is investigated for carbon ion therapy. As a complementaryapproach to evaluate the capability of the MC code FLUKA to describe thecharacteristics of mixed radiation fields created by ion beams, simulated mi-crodosimetric quantities for ion beams are compared with experimental data.The correct description of microdosimetric quantities is also important whenthey are used to predict values of relative biological effectiveness (RBE) forion beams.

Furthermore, two models describing Compton scattering and the acollinear-ity of two-quanta positron annihilation at rest in media were developed, vali-dated and integrated as native models in FLUKA. The detailed description ofthese processes is important for an accurate simulation of positron emissiontomography (PET) and prompt-γ imaging. Both techniques are candidates tobe used in clinical routine to monitor dose administration during cancer treat-ments with ion beam therapy.

The second objective of this thesis is to contribute to the development ofa MC-based treatment planning tool for protons and ions with atomic num-ber Z ≤ 8 using FLUKA. In contrast to previous clinical FLUKA-based MCimplementations for ion beam therapy which only re-calculate a given treat-ment plan, the developed prototype features inverse optimization of absorbeddose and RBE-weighted dose for single fields and simultaneous multiple-fieldoptimization for realistic treatment conditions. In a study using this newly-developed tool, the robustness of ion therapy treatment fields to uncertaintiesin the prediction of RBE values is investigated, while comparing different op-timization strategies.

c© Till Tobias Böhlen, Geneva 2012

ISBN 978-91-7447-551-7

Printed in Sweden by Universitetsservice US-AB, Stockholm 2012

Distributor: Department of Physics, Stockholm University

To Ilaria, for all the wonderful time

List of papers

The following papers are included in this thesis. They are referred to in thetext by their roman numerals.

I Böhlen TT, Cerutti F, Dosanjh M, Ferrari A, Gudowska I, Mairani Aand Quesada JM 2010 Benchmarking nuclear models of FLUKA andGEANT4 for carbon ion therapy, Phys. Med. Biol. 55 5833–5847.DOI: 10.1088/0031-9155/55/19/014

II Böhlen TT, Dosanjh M, Ferrari A, Gudowska I and Mairani A 2011 FLUKAsimulations of the response of tissue-equivalent proportional counters toion beams for applications in hadron therapy and space, Phys. Med. Biol.56 6545–6561. DOI: 10.1088/0031-9155/56/20/002

III Böhlen TT, Dosanjh M, Ferrari A and Gudowska I 2012 Simulations ofmicrodosimetric quantities with the Monte Carlo code FLUKA for carbonions at therapeutic energies, Int. J. Radiat. Biol. 88(1-2) 176–182.DOI: 10.3109/09553002.2011.620062

IV Böhlen TT, Ferrari A, Patera V and Sala PR 2012 Describing Comptonscattering and two-quanta positron annihilation based on Compton pro-files: Two models suited for the Monte Carlo method, J. Instrum. 7(07)P07018. DOI: 10.1088/1748-0221/7/07/P07018

V Böhlen TT, Brons S, Dosanjh M, Ferrari A, Fossati P, Haberer T, Patera Vand Mairani A 2012, Investigating the robustness of ion beam therapytreatment plans to uncertainties in biological treatment parameters, Phys.Med. Biol. 57 7983–8004. DOI: 10.1088/0031-9155/57/23/7983

VI Mairani A, Böhlen TT, Schiavi A, Tessonnier T, Battistoni G, Parodi Kand Patera V 2012 A Monte Carlo-based treatment planning tool for pro-ton therapy, manuscript.

VII Sihver L, Lantz M, Böhlen TT, Mairani A, Cerutti AF and Ferrari A 2012A comparison of total reaction cross section models used in FLUKA,GEANT4 and PHITS, Proceedings IEEE Aerospace 2012, Big Sky, Mon-tana, March 3-10, 2012. DOI: 10.1109/AERO.2012.6187014

Reprints were made with permission from the publishers.

Related papers which are not included in the thesis.

I∗ FIRST Collaboration (Golosio B et al.) 2011, The FIRST experimentfor nuclear fragmentation measurements at GSI, Proceedings NuclearScience Symposium and Medical Imaging Conference (NSS/MIC), 2011IEEE, Oct., 2277–2280.DOI: 10.1109/NSSMIC.2011.6153861

II∗ FIRST Collaboration (Pleskac R et al.) 2012, The FIRST experiment atGSI, Nucl. Instrum. Meth. A 678, 130–138.DOI: 10.1016/j.nima.2012.02.020

III∗ FIRST Collaboration (Abou-Haidar Z et al.) 2012, Performance of up-stream interaction region detectors for the FIRST experiment at GSI, J.Instrum. 7 P02006.DOI: 10.1088/1748-0221/7/02/P02006

Author’s contribution

My contribution to the papers which are included in the thesis is as follows.For paper I-III and V, I did all simulations and analyses and the papers are de-signed, structured and written by me, except for the appendix of paper II. Forpaper IV, I developed the model for two-quanta positron annihilation togetherwith Alfredo Ferrari. I evaluated the presented models and performed relatedsimulations and analyses. I designed, structured and wrote all the text of thepaper. I developed and validated the optimization code used for inverse treat-ment planning in paper VI. In addition, I performed parts of the calculationsand analysis for the presented treatment plans and contributed to the designand structuring of the paper. I wrote about 50% of the text. For paper VII, allFLUKA and GEANT4 simulations were performed by me and all discussionsrelated to FLUKA and GEANT4 were written by me. Further, I contributedactively to the design, structuring and proofreading of the paper.

Contents

Sammanfattning v

Abstract vi

List of papers xi

Author’s contribution xiii

Abbreviations xvii

1 Project context 11.1 Cancer treatment with ion beams . . . . . . . . . . . . . . . . 1

1.1.1 Physical and biological aspects . . . . . . . . . . . . . 21.1.2 Clinical aspects and status . . . . . . . . . . . . . . . 41.1.3 Accelerators and beam delivery techniques . . . . . . 61.1.4 Treatment planning . . . . . . . . . . . . . . . . . . . 81.1.5 Imaging techniques for treatment monitoring and ver-

ification . . . . . . . . . . . . . . . . . . . . . . . . . 131.2 Monte Carlo codes for ion beam therapy . . . . . . . . . . . . 15

1.2.1 Principal features of the Monte Carlo technique fordose calculation . . . . . . . . . . . . . . . . . . . . . 15

1.2.2 Multi-purpose Monte Carlo codes . . . . . . . . . . . 171.3 Thesis objectives and outline . . . . . . . . . . . . . . . . . . 19

2 Validation of nuclear interaction models 212.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 212.2 Hadronic interaction models . . . . . . . . . . . . . . . . . . 21

2.2.1 Total nuclear reaction cross sections . . . . . . . . . . 232.2.2 Modelling of hadronic interactions . . . . . . . . . . . 24

2.3 Benchmarking nuclear models . . . . . . . . . . . . . . . . . 282.3.1 Reaction cross sections . . . . . . . . . . . . . . . . . 282.3.2 Charged fragments . . . . . . . . . . . . . . . . . . . 29

2.4 Discussion and perspective . . . . . . . . . . . . . . . . . . . 29

3 Simulation of microdosimetric quantities 333.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 333.2 Linear energy transfer and microdosimetry . . . . . . . . . . . 333.3 Measurement devices for microdosimetry . . . . . . . . . . . 353.4 Microdosimetry applications . . . . . . . . . . . . . . . . . . 363.5 Accuracy of FLUKA simulations . . . . . . . . . . . . . . . . 39

4 Compton scattering and e+e− annihilation 414.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414.2 Model applications . . . . . . . . . . . . . . . . . . . . . . . 42

5 Monte Carlo treatment planning 455.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 455.2 Workflow and components of the MCTP tool . . . . . . . . . 45

5.2.1 Conversion of CT data for MC calculations and rangecalibration . . . . . . . . . . . . . . . . . . . . . . . 47

5.2.2 Conversion of dose-to-medium to dose-to-water . . . . 475.3 Treatment planning for carbon ions . . . . . . . . . . . . . . . 48

5.3.1 Examples of carbon ion treatment plans . . . . . . . . 485.4 Biological robustness of treatment plans . . . . . . . . . . . . 51

6 Conclusions and perspectives 53

Acknowledgements I

References III

Abbreviations

CNAO Centro Nazionale di Adroterapia Oncologica, Pavia, Italy

CT Computerized tomography

CTV Clinical target volume

DNA Deoxyribonucleic acid

DVH Dose-volume histogram

FWHM Full-width at half-maximum

GPU Graphics processing unit

GSI Helmholtzzentrum für Schwerionenforschung (formerly Gesellschaftfür Schwerionenforschung, Society for Heavy Ion Research), Darm-stadt, Germany

GTV Gross target volume

HIT Heidelberger Ionenstrahl-Therapiezentrum, Heidelberg, Germany

HT Hadron therapy

HU Hounsfield unit

IMPT Intensity-modulated particle therapy

IMRT Intensity-modulated radiotherapy

INFN-LNS Istituto Nazionale di Fisica Nazionale - Laboratori Nazionali del Sud,Catania, Italy

LEM Local effect model

LET Linear energy transfer

LOR Line of response

MC Monte Carlo

MCTP Monte Carlo-based treatment planning

MKM Microdosimetric kinetic model

MRI Magnetic resonance imaging

MV Megavoltage

NIRS National Institute of Radiological Sciences, Chiba, Japan

PET Positron emission tomography

PMMA Polymethyl methacrylate

PSI Paul Scherrer Institut, Villigen, Switzerland

PTV Planning target volume

RBE Relative biological effectiveness

RMS Root mean square (also quadratic mean)

RT Radiotherapy

SFUD Single-field uniform dose

SOBP Spread-out Bragg peak

TEPC Tissue-equivalent proportional counter

WEPL Water-equivalent path length

1. Project context

1.1 Cancer treatment with ion beams

The probability of dying from a cancer (cause of death) is about 25% in in-dustrialized countries (IAEA, 2008; Jemal et al., 2008). Radiotherapy is, to-gether with surgery, the most successful and most frequently-used treatmentfor cancer, with more than 50% of all cancer patients being treated with ra-diotherapy in industrialized countries (sometimes in combination with othertreatment modalities) (IAEA, 2008). The aim of curative radiotherapy is to de-liver an amount of radiation dose to solid tumours that causes reproductive celldeath to all cancer stem cells. However, achievable tumour doses are generallylimited by radiosensitive organs and normal tissues surrounding the tumour.External radiotherapy with protons and ions, so-called hadron therapy or ionbeam therapy, is becoming increasingly used in recent years with an ever grow-ing number of treatment facilities being operational or in construction, see fig-ure 1.1. This can mostly be attributed to the fact that ion beam therapy allowsto target the tumour effectively while sparing surrounding normal tissues.

0

5

10

15

20

25

30

35

40

1950 1960 1970 1980 1990 2000 2010 2020 0

20000

40000

60000

80000

Active H

T c

entr

es

Tre

ate

d p

atients

Year

Hadron therapy (HT) centres and patients

Active centresTreated patients

Figure 1.1: World-wide number of operational hadron therapy centres andtreated patients. Data from PTCOG (2012).

1

1.1. CANCER TREATMENT WITH ION BEAMS

1.1.1 Physical and biological aspects

The main advantage of protons and light ions (atomic number Z . 8) comparedto conventional radiation (photons and electrons) when used for radiotherapy istheir inverse depth-dose profile, known as Bragg curve, which allows to placethe highest energy release per path length (i.e. stopping power) just beforestopping, inside the tumour. Furthermore, the increased mass of ions results,compared to light charged particles, such as electrons, in substantially reducedmultiple Coulomb scattering and leads to an improved lateral dose-gradient aswell as well-defined particle ranges connected with a sharp dose fall-off whenstopping. This allows to deliver treatments with lower integral doses to thepatient for the same or an escalated dose in the tumour (St. Clair et al., 2004;Widesott et al., 2011). Apart from the already mentioned features of the doseprofile which are mainly due to electromagnetic processes, nuclear fragmenta-tion reactions, leading to the production of energetic secondary fragments withcharges lower or equal to the charge of the primary beam particles, have an im-portant impact on the dose profiles and radiation fields created by ion beams.Figure 1.2 shows the primary and charged fragment yields versus depth, theso-called attenuation and build-up curves, for a carbon ion beam initially at400MeV per nucleon for water, water-equivalent bone, and water-equivalentair. The contribution of fluences of charged secondary fragments to the ra-

0.01

0.1

1

0 5 10 15 20 25 30 35

N/N

0

depth in water [cm]

Water (solid) - Air (dashed) - Bone, cortical (dotted)CB

BeLi

HeH

Figure 1.2: Fragment yields of hydrogen, helium, lithium, beryllium, boron,and carbon ions versus depth as a fraction of primary carbon ions initially at400MeV per nucleon. Fragment yields for the materials water (solid lines), air(dashed lines), and cortical bone (dotted lines) with water-equivalent densitiesare presented. The inelastic scattering length for carbon ions in water-equivalentbone and air is larger compared to the one for water. Consequently, primarycarbon ions are found to be attenuated more slowly and fluences of secondarycharged fragments are lower for bone and air.

2

CHAPTER 1. PROJECT CONTEXT

diation field of an ion beam increases with increasing ion mass and atomicnumber Z (see chapter 2). Secondary charged fragments lead to a significantalteration of the dose profile and give rise to the characteristic ‘fragment tail’observed for ions with Z ≥ 2 (see also figure 1.3) and a slightly increased an-gular spread. Far travelling neutrons and light charged fragments produced innuclear interactions are the main origin of out-of-field doses and are of impor-tance for studies assessing the secondary cancer risk of ion beams (Hultqvistet al., 2010; Jarlskog and Paganetti, 2008b; Yonai et al., 2012).

In addition to the advantageous physical properties, an increased biologicaleffectiveness at depths close to the Bragg peak is the main asset for using ionsheavier than protons, that are in terms of clinical experience most notably car-bon ions. Photon and electron beams are sparsely ionizing and exhibit an ap-proximately uniform radiation quality in the patient when used therapeutically.This leads to a uniform biological effectiveness for a given biological systemand end-point (for instance cell surviving fraction) throughout the treatmentfield for the same absorbed dose Dabs levels. Instead, low-energy ions with ahigh linear energy transfer (LET) are considered to be densely ionizing. Theycause more severe and complex damages to cells for the same absorbed dosecompared to low-LET or sparsely ionizing radiation. The difference in bio-logical effectiveness of a type of radiation (Dtest) with respect to a referenceradiation (Dref), which is normally taken to be 60Co γ rays (IAEA, 2008), isdescribed by the concept of relative biological effectiveness (RBE), defined asthe absorbed dose ratio of both radiations which achieve in a biological systemfor a given end-point an isoeffect

WRBE =

(Dref

Dtest

)isoeffect

. (1.1)

Since these RBE values can vary strongly for ions as a function of depth, ab-sorbed dose Dabs in a given region is no longer an adequate measure for suchradiation to evaluate the expected effect of a patient treatment and is replacedby RBE-weighted dose

DRBE =WRBE ·Dabs . (1.2)

RBE values for protons are generally approximated for clinical purposesby a constant RBE of 1.0 or 1.1 for the whole treatment field (IAEA, 2008;ICRU 78, 2007) as they are varying to a lesser degree for proton fields com-pared to fields delivered with higher-Z ions. While this may be a clinicallyacceptable approximation for proton beams, it is clearly not advisable for car-bon ion beams which exhibit a much stronger RBE increase towards the endof their ranges. Figure 1.3 shows the typical features of a spread-out Braggpeak (SOBP) used for treating patients with carbon ions. The RBE varies no-tably over the treatment volume and increases towards the distal edge of the

3

1.1. CANCER TREATMENT WITH ION BEAMS

SOBP. To compensate for this increase, a lower absorbed dose is given towardsthe distal end of the SOBP in order to achieve a constant RBE-weighted dosein the treatment volume. This makes a systematic derivation of RBE values

Figure 1.3: Absorbed dose Dabs, RBE WRBE and RBE-weighted dose (biologicaldose) DRBE for a SOBP created with carbon ions. The RBE was calculated forhuman salivary gland cells with respect to cell survival. The ‘fragment tail’,created by long-ranged charged nuclear fragments, is marked with an arrow.

necessary for the clinical application of carbon ions.The two RBE models which are currently the most advanced for their clin-

ical application to patient treatment with scanned carbon ions are the localeffect model (LEM) which was developed by Scholz and co-workers (Elsässeret al., 2010; Scholz et al., 1997) and the microdosimetric kinetic model (MKM)(Hawkins, 1994, 2004). By assuming a target geometry and accounting for theradiation track structure, both of the models aim to translate a given responseof a biological system to photon radiation to a response to ion radiation with arestricted number of parameters. Although both models are based on the sameprincipal constituents, the translation approach is largely different. For a moredetailed description of the MKM see section 3.4.

1.1.2 Clinical aspects and status

Depending on the site and tumour target, proton therapy is applied with dif-ferent treatment objectives. These include currently the following: ablativetherapy1, organ preservation therapy (i.e. most notably choroidal melanoma),dose escalation around critical structures, and reduction in acute and long-termmorbidity (Linz et al., 2012, Ch. 2). The last point includes importantly pae-diatric tumours, where integral dose reduction in normal tissues is of major

1Ablative intent with single or hypofractionated therapy.

4

CHAPTER 1. PROJECT CONTEXT

concern to reduce both acute and long-term toxicities (Hall, 2006; Merchant,2009; St. Clair et al., 2004).

Light ions with 1 < Z . 8 exhibit compared with protons a notably en-hanced radiobiological effectiveness in the tumour region due to their higherLET. This is advantageous, especially for treating radio-resistant hypoxic tu-mours. Moreover, they possess a reduced penumbra which translates in asharper lateral dose fall-off. Currently, most clinical experience for ions withZ > 1 exists for carbon ions. Carbon ions and protons are used for simi-lar indications. In clinics, carbon ions and protons are applied to skull-basechordomas and chondrosarcomas, uveal melanomas, head and neck cancers,non-small cell lung cancer (NSCLC), liver tumours, prostate carcinomas andrenal-cell carcinomas (Linz et al., 2012; Okada et al., 2010; Schulz-Ertner,2009). According to a recent study, current clinical data does not allow to de-termine the clinical efficacy of carbon ion therapy compared with proton ther-apy yet (Suit et al., 2010). This is principally due to different dose fractionationschemes which are used. Randomized trials evaluating efficacy of proton andcarbon ion treatments for chondrosarcomas, chordomas and prostate cancersare ongoing (Combs et al., 2010).

The potential use of ion beam therapy for treating non-cancerous disor-ders, such as trigeminal neuralgia, arterial aneurism, atrial fibrillation, renalsympathetic denervation, and arteriovenous malformations (AVM), is investi-gated (Bert et al., 2012). For such treatments, the Bragg peak of particle beamsis used as ‘remote scalpel’ for radiosurgery. The treatment of AVM with pro-tons has been performed by several proton therapy centres (Bert et al., 2012).Photon beams are already applied for the treatment of such non-cancerous dis-orders.

With in total about 100 000 patients treated (as of 2012, see also figure 1.1),ion beam therapy is, since the establishment of the first hospital-based treat-ment facility in the early 90’s, still an exotic treatment form. This can mainlybe attributed to the technical challenges which result in a considerable com-plexity of the treatment delivery, necessitating specifically trained personnelas well as a larger initial investment and maintenance cost of ion beam therapyfacilities compared to conventional radiotherapy, that is mostly Megavoltage(MV)-photon radiotherapy (Linz et al., 2012, Ch. 3). For radiotherapy withions heavier than protons, a notably different radiation quality, compared tosparsely ionizing radiation for which most clinical experience exists, and thestrong variation of the radiation quality in the treatment field (mostly as a func-tion of depth) poses an additional challenge and complication for appropriateclinical application. On the other hand, clinical results of ion beam therapyare promising, particularly for paediatric patients and for inoperable locally-advanced and radio-resistant tumours which are often considered incurable.

5

1.1. CANCER TREATMENT WITH ION BEAMS

Epidemiological studies estimate that ion beam therapy is beneficial for a cer-tain fraction of cancer treatments (e.g. ENLIGHT1 studies: Baron et al., 2004;Mayer et al., 2004; Orecchia et al., 2004).

1.1.3 Accelerators and beam delivery techniques

Accelerators designed for the delivery of radiotherapy treatments have the fo-cus set on the reliability of the operation, beam control, maximum time ofbeam availability, and low maintenance. All these are key points for treat-ing patients safely, effectively and efficiently in a clinical environment. Incontrast, accelerators which are used primarily for research purposes, as wereused for many ion beam therapy pilot projects, are generally designed with afocus on a maximum flexibility of the machine. Therapeutically-used energiesfor protons are usually between 50 and 250 MeV. For carbon ions, energiesused for patient treatment are between 80 and 400 MeV per nucleon. Beamintensities for treatment are generally in the range of 1 · 106 to 4 · 1010 par-ticles per second (synchrotron). Accelerators used for treatment delivery en-compass both cyclotrons and synchrotrons of typical diameter ranges between5-25 m for proton and carbon ion beams. Cyclotrons have a stable and easilyadjustable beam intensity and are generally considered to be easy-to-operate,reliable and compact. However, they have a fixed extraction energy and thebeam energy variation for treatments is currently only achieved by passive de-grading systems in the beam line. Apart from the problem of unwanted activa-tion, this technique bears additional difficulties for carbon ions. The reason forthis being the fragmentation of carbon ions into alpha particles with the samecharge-mass ratio as the primaries and mostly also a similar velocity. Thismakes the cleaning of degraded carbon beams difficult to achieve (Schardtet al., 2010). Synchrotrons allow for extraction energy variation. However,they have a larger diameter than cyclotrons and need an injection linac of sev-eral meters length. In addition, they have a delicate extraction system and aregenerally more complex to operate. Both cyclotrons and synchrotrons are usedfor proton therapy with most commercial systems built by IBA Inc. (Belgium),Accel Varian (USA) (both cyclotrons) and Hitachi Ltd. (Japan) (synchrotrons)(Schardt et al., 2010). Only synchrotrons are used in carbon ion therapy centresor dual and multiple ion centres which are currently in operation. Nonetheless,there are efforts to design and built superconducting cyclotrons also for carbonion therapy.

Patients are generally preferably treated in horizontal position, fixed on apatient table. To irradiate patients from different directions, patient tables areusually mounted on devices, such as robotic arms, which allow some rotation

1http://enlight.web.cern.ch

6

CHAPTER 1. PROJECT CONTEXT

about the vertical axis. Rotating gantries which steer the beam magneticallyallow to access and treat the patient in a more flexible way with fields froman enlarged solid angle compared to treatments with fixed beam lines and areclinically preferred. If only fixed beam lines are available the angular restric-tions of the beam delivery are often compensated for by patient tables tiltingabout one or two axes (typically up to±15◦) and by treating patients in supine,prone or sitting positions (Linz et al., 2012, Ch. 30).

There are two principal techniques applied to deliver therapeutic treatmentfields with hadrons. The broad beam technique (often scattering technique)uses scattering elements or magnetic wobblers to broaden the pristine pencilbeam delivered by the accelerator transversally (Schardt et al., 2010). Colli-mation devices are then used to conform the widened beam to the target shape.The beam energy is degraded by introducing a range modulator in form ofa turning wheel of varying thickness or by so-called ripple filters in order toproduce a SOBP suitable to target the whole tumour in depth with a homo-geneous RBE-weighted dose. A patient-specific range compensator shapesthe distal part of the SOBP to be conform with the planning target volume(PTV). While most pioneering work for the scattering technique was carriedout in the 60’s, the scanning technique (also pencil beam technique) was de-veloped mostly in the 80’s and 90’s with pioneering work being done by PSI(Paul Scherrer Institut, Switzerland) for protons (Pedroni et al., 1995) and byPSI (Helmholtzzentrum für Schwerionenforschung, Germany) for carbon ions(Haberer et al., 1993) in the flavours of spot scanning and raster scanning. Thescanning technique employs thin accelerator pencil beams which are laterallydeflected by magnets and varied in energy in order to scan the PTV. Scan-ning is generally performed by starting to scan the transverse layer definedby the largest beam energy and then decreasing the beam energy step-wise,always when the scanning of such an ‘energy slice’ (slice with equal radio-logical depth) is completed. While the scanning technique allows to deliverin a very flexible manner more conformal treatment fields which are less con-taminated by unwanted secondary fragments (in particular neutrons) producedin passive beam elements (Brenner and Hall, 2008; Hall, 2006), it is still farmore sensitive to target motion: an interplay between patient/tumour motionand beam scanning results in a degradation of the conformity and homogeneityof delivered fields. Hot and cold spots in dose delivery can be avoided by av-eraging motion-effects with so-called fast repainting or rescanning techniques(Bert and Durante, 2011). For these techniques the PTV is scanned many timesrapidly resulting in an averaged and hence more homogeneous dose distribu-tion.

7

1.1. CANCER TREATMENT WITH ION BEAMS

1.1.4 Treatment planning

The design and optimization of single or multiple fields of therapeutic beamsto deliver a conformal dose or RBE-weighted dose to a target structure whilesparing organs-at-risk (OAR) and normal tissues is referred to as treatmentplanning. Treatment planning for therapeutic ion beams proceeds in severalsuccessive steps. In the following an overview and outline of the most essen-tial steps in the planning chain is given. In many respects, treatment planningfor radiotherapy with photons and protons or ions are alike. The main differ-ences between treatment planning for conventional radiotherapy with photonsand ion beam therapy arise from the differences in their depth-dose profilesand differences in radiation quality, which have to be accounted for (see sec-tion 1.1.1). Depending on the ion beam delivery technique, the treatment plan-ning process varies. The description below focuses on treatment planning forthe more flexible scanning technique as employed for instance at HIT (Heidel-berger Ionenstrahl-Therapiezentrum, Germany) and CNAO (Centro Nazionaledi Adroterapia Oncologica, Italy). In the beginning, the target volume(s) forradiotherapeutic treatment as well as critical structures are delineated on theplanning computerized tomography (CT) image of the patient. Planning CTdata maps the photon attenuation in typical voxel dimensions of 1 mm in thedimensions within a CT slice and with 2-3 mm slice thickness. Manual delin-eation is preformed on the basis of the CT slices to create volumes relevant forfurther planning and optimization. Information from the CT data sets is oftencomplemented by images from magnetic resonance imaging (MRI) (Schlegelet al., 2006, pp. 99–111) and positron emission tomography (PET) to allowfor a better definition of the target volume (Levy, 2007). The planning CT datawith the outlined structures is the basis which is then used as a representationof the patient for the subsequent planning procedure which is specific to the ra-diotherapy modality. A target RBE-weighted dose per fraction (together witha fractionation scheme) is prescribed to the target structure(s) and maximumRBE-weighted doses1 are prescribed to the OARs, according to the clinicaltreatment protocol. For proton therapy, it is common clinical practice to usea fixed RBE of 1.0 or 1.1 (IAEA, 2008) whereas for scanned therapeutic car-bon ions, RBE values are determined by biophysical models as discussed insection 1.1.1. Latter RBE values are dependent in particular on the specific tu-mour and normal tissue types involved in the treatment. The more pronouncedvariation of biological effectiveness for carbon ions compared to protons whichmakes it necessary to account for the variation of RBE clinically is the maindifference arising for treatment planning for the two modalities. In the next

1In modern treatment planning systems also constraints on the dose-volume his-tograms (DVH) are possible.

8

CHAPTER 1. PROJECT CONTEXT

step, a suitable number of beams (typically 1-3) and beam directions are cho-sen for the dose delivery to the target. Beam directions are selected based onseveral criteria. These include notably (1) the angular restrictions of the acces-sibility of the patient due to the beam delivery system (i.e. due to fixed-beamline, gantry, limited rotation of patient table), (2) good target coverage andavoidance of critical structures, (3) avoidance of uncertainties in the delivereddose and (4) restricted total number of fields and overall delivery time. Point(3) aims to increase the robustness of the treatment plan heuristically and in-cludes the avoidance of large heterogeneities in the beam entrance path (e.g.nasal cavities) which result in larger range uncertainties, as well as avoidanceof areas with larger movements and beam angles orthogonal to organ motionsas they will generally give larger dose deviations than beams parallel to themotion direction (Linz et al., 2012, Ch. 30, p. 508f). For most treatments, anopposed field geometry is preferred (Jäkel et al., 2001a). Alternative irradia-tion geometries include single fields, two orthogonal fields and patched fields.The PTV is obtained in the treatment planning process by enlarging the clin-ical target volume (CTV) with appropriate margins, specific to the treatmentmodality and field directions, to ensure target coverage. Aspects specific to thetreatment modality that are considered for margin selection include notablyset-up uncertainties, range uncertainties, motion effects as well as limitationsdue to possible adjacent OARs and uncertainties in the beam delivery system.

Conversion of CT information for dose calculation Dose and fluence cal-culations are performed on the CT images. For this purpose Hounsfield units(HU), the native unit of CT information which measures photon attenuationcoefficients µx and which is closely linked to electron densities, have to beconverted into quantities significant for the stopping of protons and ions inmatter. The HU of a given CT voxel is defined as

HU =

(µx

µH2O−1)·1000 , (1.3)

where µx and µH2O are the mean values of the photon attenuation coefficientsof tissue and water, respectively. The typical range for HUs is between -1000 and 3500. The calibration of HUs is fixed with air (-1000) and water(0) and is dependent on the CT scanner and scanning protocol. Dose enginesfor treatment planning systems (TPS) compute doses traditionally in a water-equivalent representation of the patient’s anatomy in the beam’s eye view.Hence, HUs are mapped to water-equivalent path lengths (WEPL) or relativerange in water (Jäkel et al., 2001b; Rietzel et al., 2007; Schaffner and Pedroni,1998) and result in a CT-to-range calibration which is mostly empirical. The

9

1.1. CANCER TREATMENT WITH ION BEAMS

relation depends on the x-ray energy spectrum of the CT and beam harden-ing effects (Schaffner and Pedroni, 1998). Figure 1.4 shows the calibration ofWEPLs and corresponding HUs at CNAO. TPSs use such a calibration curveto compute particle ranges in the patient geometry. Range uncertainties due

Figure 1.4: Calibration curve of WEPLs and corresponding HUs for planningCT scans at CNAO.

to this method are estimated to be generally in the order of 3-5% (Linz et al.,2012, Ch. 29, p. 492). For special situations, such as in the presence of metals(e.g. dental fillings, prostheses or neurosurgical implants) or in regions withvery low or high densities (e.g. lung or bone), larger errors are found (Jäkel,2006). Inaccuracies are mostly due to limited validity of using HUs as a sur-rogate to deduce ion stopping powers. Monte Carlo (MC) dose calculationsare naturally performed in non-distorted geometries taking into account thereal composition of the patient’s anatomy. Therefore, the previously discussedapproach is not directly applicable to MC-based dose calculations. Details ofa HU conversion scheme for MC-based dose computations are given in sec-tion 5.2.1.

Dose- and fluence-calculation algorithms Approaches used to obtain iondose and fluence distributions in ion beam therapy can be divided into analyti-cal algorithms and algorithms using the MC method for particle transport, alsocalled MC codes. Due to their meticulous and faithful modelling of physicalinteractions, MC codes have inherently the potential to provide the most accu-rate algorithms for dose and fluence computations. However, long calculationtimes still largely impede their use routinely in the clinics for ion beam ther-apy. Nevertheless, MC codes are well-established to support numerous aspectsof ion beam therapy as discussed in more detail in section 1.2.

10

CHAPTER 1. PROJECT CONTEXT

Analytical dose engines used for treatment planning are specific and tai-lored to the beam delivery technique. The following discussion will focus onanalytical algorithms used for scanning techniques. So-called ray tracing al-gorithms (also ray casting) were developed for broad beam techniques (Chenet al., 1979; Endo et al., 1996) and built the foundation for the development ofpencil beam algorithms which are used nowadays for treatment planning withscanned beams. Pencil beam algorithms model the laterally integrated dosedistribution Ddepth(z,E0) with depth z and lateral dose distribution Dlat(z,r,E0)separately and provide thereby a simple parametrization of dose distributionsin homogeneous water phantoms (Linz et al., 2012, Ch. 30, p. 513)

P(z,r,E0) = Ddepth(z,E0) ·Dlat(z,r,E0) , (1.4)

where E0 is the initial beam energy and r is the lateral distance from the beamaxis. The general shape of the depth-dose distribution Ddepth(z,E0) is givenby the Bethe-Bloch equation, but includes usually several modifications to ac-count for the beam energy spread, energy straggling, scattering effects and nu-clear fragmentation. Dlat(z,r,E0) is generally parametrized by a single Gaus-sian or the sum of several Gaussians (Parodi et al., 2010). The lateral full-widthat half-maximum (FWHM) of the beam increases with depth mostly due to theeffects of multiple Coulomb scattering and nuclear fragments.

The accurate and facility-specific parametrizations of the dose distribu-tions are generally obtained from fitted measurements (Schaffner et al., 1999),from MC-computed dose distributions validated with measurements (see sec-tion 1.2) or by hybrid approaches. To account for the patient geometry, the dosedistribution in water is scaled for treatment planning according to the water-equivalent density integrated along the central beam axis in the patient. Such atreatment neglects all off-axis density variations which lead to altered particleranges and an increased beam energy spread. For typical scanned beams withlateral sizes in the order of 5-10 mm (FWHM), this approximative treatmentcan lead to significant misestimations in clinical dose distributions for hetero-geneous regions (Schaffner et al., 1999). To improve accuracy of dose pre-dictions for heterogeneous regions, a commonly-used approach is to composescanned pencil beams used for treatment of multiple elemental pencil beams(also called beamlets or infinitesimal pencil beams) (Schaffner et al., 1999;Soukup et al., 2005). Schaffner et al. (1999) investigated various analyticaldose models with improved modelling of heterogeneities and concluded thatnone of these approaches can predict doses correctly in all situations. Furtherapproximations are inherent to such analytical approaches. These include theeffects of nuclear fragmentation with depth and their influence on the lateralbeam penumbra, which can be only empirically accounted for by pencil beamalgorithms (Krämer et al., 2000).

11

1.1. CANCER TREATMENT WITH ION BEAMS

For treatment planning which considers variations of biological effective-ness of particle beams, as is necessary for treatments with carbon ions, not onlyintegral doses but also relative dose contributions from charged fragments andtheir energy distributions have to be known. This is necessary in order to modelthe RBE values at every point of the radiation field. Fragment fluences can bepredicted either by empirical models (e.g. Krämer et al., 2000) or using MCcodes (see chapter 2).

Treatment plan optimization In the following, aspects of treatment planoptimization for scanned ion beams are briefly discussed. Treatments usingthe broad beam technique follow a different approach which is by far less flex-ible. The SOBPs for broad beam techniques are determined by the hardwarewhich modulates the beam energy in a fixed way in order to achieve a homo-geneous RBE-weighted dose with depth in the target volume. In contrast, thescanning technique allows for a given field direction to vary particle numbersof single pencil beams nearly freely to optimize the resulting dose distribu-tion for a given treatment plan. The segmentation into PTVs and OARs isthe starting point for the optimization procedure. Ideally, the PTV should becovered uniformly with the prescribed dose while sparing as much as possi-ble OARs and normal tissues. Also additional optimization criteria may beconsidered in the optimization process. These may include robustness to phys-ical uncertainties (Inaniwa et al., 2011; Pflugfelder et al., 2008; Unkelbachet al., 2009), such as range uncertainties, possible patient set-up errors, andpatient motion, as well as uncertainties in biological treatment parameters, asdiscussed in paper V. Computation-intensive iterative optimization proceduresare used to find a compromise between homogeneous PTV coverage, sparingof OARs and possibly additional optimization goals for a given set of pencilbeams with scalable particle fluences. For typical patient treatment cases, be-tween 10 000 and 50 000 of such beam scanning spots have to be optimizedfor each field. Optimizations should take into account limitations of the beamdelivery system, i.e. minimum particle number deliverable per beam spot, andoverall time for treatment delivery, i.e. avoidance of too many scanning spotsand scanning spots with only low doses.

For optimization of RBE-weighted doses, as it is necessary for therapeuticcarbon ions, the RBE values of the treatment fields have to be evaluated forthe optimization process. Since RBE values have a non-linear dependence ondose, dose distributions have to be calculated on an absolute scale. This alsoimplies an added complexity for the optimization process in general (Krämeret al., 2000; Krämer and Scholz, 2006). Separate optimization of single treat-ment fields to achieve a uniform physical or RBE-weighted dose is referred toas single-field uniform dose (SFUD) optimization. Instead, simultaneous op-

12

CHAPTER 1. PROJECT CONTEXT

timization of multiple ion fields which includes all scanning spots of all fieldsis commonly referred to as intensity-modulated particle therapy1 (IMPT) (Lo-max, 1999). In paper VI, implementations of two established optimizationalgorithms are described and used for optimizing absorbed doses and RBE-weighted doses for ion beam treatment plans using MC dose kernels.

1.1.5 Imaging techniques for treatment monitoring and verification

Increased conformity in dose delivery of radiotherapy treatments allows tospare healthy tissues by achieving steeper dose gradients at the PTV borders.At the same time, this progress in dose conformity requires an augmented pre-cision and control in the dose delivery since, due to the steep dose gradients,even slight deviations in dose delivery can lead to severe tumour misses orover-exposure of OARs and may lead to a worse therapeutic outcome. Hence,the development of highly conformal radiotherapy treatments, such as ionbeam therapy, went along with an increased use of imaging techniques whichhelp to assure correct treatment delivery. This includes imaging which allowsto control the correct patient positioning and anatomy for each treatment ses-sion, such as cone-beam CT and, at an investigational stage, proton and heavyion CT (Rinaldi, 2011). Recently, also imaging techniques which aim at an in-dependent dose verification by detecting possible deviations of the in vivo de-livered doses from the prescribed doses, according to the treatment plan, havebeen increasingly a focus of research. For ion beam therapy, several imagingmethods are explored which use secondary radiation emerging from the patientthat originates from nuclear interactions during the treatment delivery. To date,in vivo range monitoring of ion beam therapy using PET is the only imagingtechnique which has been successfully applied in clinical scenarios in variouspilot studies (Enghardt et al., 2004b; Linz et al., 2012, Ch. 31). This monitor-ing technique is based on positron emitters, such as 11C, 10C and 15O, whichare produced during treatment delivery with proton or carbon ion beams in thepatient. PET scanners are used to measure photons pairs from two-quanta po-sition annihilations either during or shortly after the treatment (Parodi et al.,2008b). These measurements allow the reconstruction of a three-dimensionalactivity map which can then be visually compared to the expected activity fromsimulations to verify the treatment and detect possible delivery errors, such asrange deviations (Parodi et al., 2008a). If critical, found discrepancies triggeran investigation of the origin of the differences, for example by the acquisi-tion of a new patient CT to check for possible anatomical changes (such ascavity filling, tumour shrinkage, oedema, etc.). Patient replanning can then

1Even though SFUD modulates beam fluences it is generally not considered to bethe equivalent of intensity-modulated radiotherapy (IMRT) for photons.

13

1.1. CANCER TREATMENT WITH ION BEAMS

correct previous errors in treatment delivery. In order to allow a better cor-relation of the activity distribution with the dose profile and increase the lowβ+-activity (approximately 200 Bq/Gy/cm3 for 12C and 600 Bq/Gy/cm3 forprotons), which is considered to be one of the main limiting factors for PET(Enghardt et al., 2004b), patient irradiation with β+-radioactive ion beams hasbeen explored (Enghardt et al., 2004a).

Prompt-γ imaging and prompt-charged particle imaging are two methodswhich have been suggested as alternatives for in vivo ion beam treatment mon-itoring in recent years. Measurements of proton and carbon ion beams in ho-mogeneous phantoms have demonstrated that the Bragg peak position can becorrelated with the prompt-γ emission profile in beam direction (Min et al.,2006; Testa et al., 2008, 2009). The prompt-γ profile exhibits a characteris-tic fall-off at the end of the primary ion range. This decrease in prompt-γemission is caused by the drop of the nuclear reaction cross sections for lowenergies due to the Coulomb barrier (see section 2.2). As a consequence of thehigh background from prompt-radiation, PET monitoring can be performedcurrently only during pauses of beam delivery (as they occur for synchrotronsbetween the extraction phases) or, as off-line PET, shortly after the deliveryof a treatment fraction. Instead, prompt-γ imaging potentially allows for real-time monitoring during the treatment. However, poor photon counting statis-tics is an issue. This is currently addressed by developments of new tailoreddetection devices, such as Compton cameras and collimated prompt-γ cam-eras (Testa et al., 2010). The detection of energetic light charged fragments(i.e. mostly protons) is proposed to be used in a similar fashion to the previ-ously mentioned technique to obtain information about the Bragg peak position(Dauvergne et al., 2009; Henriquet et al., 2012). Similarly to prompt-γ’s, theemission profile of these particles in the patient decreases towards the end ofthe primary ranges and allows for a correlation with the in vivo particle ranges.Interaction vertex imaging can be used to reconstruct trajectories of detectedsecondary charged particles and obtain their point of production by intersect-ing their trajectory with the extrapolated trajectory of the primary ion or, incase of larger multiplicities, by intersection with the trajectories of the othersecondary charged particles. Compared to prompt-γ imaging, charged parti-cle imaging has the advantage of an easier detection of the charged particlesand an increased counting rate. The feasibility of both approaches in a clinicalscenario still remains to be demonstrated.

14

CHAPTER 1. PROJECT CONTEXT

1.2 Monte Carlo particle transport codes and their use forion beam therapy

1.2.1 Principal features of the Monte Carlo technique for dose calcu-lation

Particle transport codes using the MC method are considered to be the gold-standard for dosimetric calculations in conventional radiotherapy and are em-ployed for various aspects in the field of conventional radiation therapy sincedecades (Chetty et al., 2007; Rogers, 2006). Over the last decade MC codesare also increasingly being used at ion beam therapy facilities. For pencilbeam scanning techniques, inverse treatment planning and dose computationis generally performed based on analytical pencil beam algorithms (discussedin section 1.1.4). They allow for fast computation times while providing anacceptable accuracy for clinical needs in most cases. However, their inherentapproximations may show shortcomings for some critical treatment cases in-volving for instance regions with large heterogeneities or treatments in regionswith metallic implants (Parodi et al., 2007b). The high dose gradients whichare achieved with particles underline the need for accurate dose calculations.Due to the high treatment conformity obtained with ion beam therapy, dosedistributions are compared to conventional radiotherapy more sensitive to pa-tient and organ motion, density variations and heterogeneity effects. Figure 1.5illustrates the difference in effect for the same range offset ∆x for photon andion beam therapy. The success of an ion beam therapy treatment dependstherefore crucially on the accuracy of the patient representation used for dosecalculations (i.e. CT data), the precision of the patient set-up and beam deliv-ery, the accuracy of the dose calculation algorithm, the accounting for possibleinter- and intra-fractional motion, etc. (see also discussion in section 1.1.5).MC codes can help to address one link in this interleaved chain relevant for thecorrect and successful treatment delivery by facilitating an increased accuracyin dose and fluence computations in the patient. This may allow, for example,to decrease treatment margins for complex geometries, reducing thereby theintegral dose received by the patient as discussed by Paganetti (2012). Com-pared to analytical algorithms, full MC particle transport codes are superior, inthat they can:

• reproduce faithfully physical interactions1,

• take into account the real atomic composition of tissue (opposed to awater-equivalent approach),

1Limited only by the knowledge about physics processes themselves and the avail-ability of models describing them which are suitable for MC methods.

15

1.2. MONTE CARLO CODES FOR ION BEAM THERAPY

Figure 1.5: Resulting changes in depth-dose distributions are shown schemati-cally for a MV photon field and a SOBP delivered with ions for the same mis-estimation in range ∆x. Under-dosage of the tumour and over-dosage of normaltissues and critical structures close to the tumour are much more severe for theSOBP due to the conformity of the treatment and the maximum dose difference∆Dmax is generally much larger.

• are inherently capable of describing three dimensional mixed radiationfields in complex geometries and

• include naturally effects of heterogeneities.

Due to their rigorous description of physics in nearly arbitrary geometries, theyalso have an increased flexibility in application range compared to analyticalapproaches, which are usually tailored to a specific problem. For these reasons,MC codes are used at ion beam treatment facilities to support various aspectsof treatment planning.

A noteworthy alternative to the multi-purpose MC codes (also full MC),which are discussed in this document, are simplified MC implementations tai-lored to treatment planning for ion beam therapy (Fippel and Soukup, 2004;Hotta et al., 2010; Tourovsky et al., 2005). These codes are based on the MCtechnique but they use largely simplified physics models and parametrizationsfor calculating dose deposition, with an accuracy in the intermediate range be-tween detailed MC particle transport codes and analytical algorithms. Theyare promising for performing time-efficient calculations and are mostly aimedto be used in routine clinical treatment planning. Simplified MC codes arealready increasingly available in commercial TPSs for conventional radiother-apy. Also MC-based methods, such as track-repeating algorithms (Li et al.,2005; Yepes et al., 2010), can be used for fast dose computation with a highaccuracy. On the contrary full MC codes have an increased computation time,

16

CHAPTER 1. PROJECT CONTEXT

but introduce much less approximations in the physics description. Therefore,they offer the highest flexibility and predictive capability in nearly arbitrarilycomplex situations.

1.2.2 Multi-purpose Monte Carlo codes

Multi-purpose MC particle transport codes such as FLUKA (Battistoni et al.,2007; Ferrari et al., 2005), GEANT4 (Agostinelli et al., 2003; Allison et al.,2006), MCNPX (Hughes et al., 1997), SHIELD-HIT (Dementyev and Sobo-levsky, 1999), and PHITS (Niita et al., 2006) have shown to provide detailedsimulations of fluences and dose distributions in patient or patient-like ge-ometries for proton and/or ion beams. Even though they are generally notused for routine treatment planning dose calculations and optimization as theyare currently too time-consuming, they are prominent candidates for numer-ous applications in ion beam therapy. In addition to the MC codes whichprovide fundamental modelling of materials, geometry and physics, there areseveral dedicated MC-based frameworks, such as GAMOS1, GATE2, PTsim3,TOPAS4 (all based on GEANT4) and FICTION (Sommerer et al., 2012, basedon FLUKA), which are specifically tailored to the needs of ion beam therapyapplications. Even though being partly still under development at the time ofwriting, they enjoy increased popularity in recent years.

Clinical implementations of MC codes which allow to re-calculate treat-ment fields obtained and optimized by TPSs using pencil beam algorithmsexist at several treatment facilities (Paganetti et al., 2008; Parodi et al., 2009a;Peterson et al., 2009; Sommerer et al., 2012; Tourovsky et al., 2005). Besidesfacilitating a retrospective accurate analysis of dose delivery to patients, theyare also employed for a prospective ‘second opinion’ and improvement of thetreatment process for critical cases prior to the patient treatment (Linz et al.,2012, Ch. 7, p. 113). Some authors argue that they should become a standardtool for quality assurance for ion beam therapy (Tourovsky et al., 2005). With-out entering into such a debate it is clear that MC implementations providean independent dosimetric verification tool, which can be particularly usefulin the commissioning and initial stage of a facility (Parodi et al., 2012). Theycan also reduce the necessity of experimental dosimetric treatment verifica-tions which are both more time-consuming and expensive and are currentlyperformed routinely for each treatment field prior to the first patient treatment(Linz et al., 2012, Ch. 7, p. 106). Furthermore, MC codes can be used to

1http://fismed.ciemat.es/GAMOS2http://www.opengatecollaboration.org3http://wiki.kek.jp/display/g4med/PTsim4http://www.slac.stanford.edu/∼perl

17

1.2. MONTE CARLO CODES FOR ION BEAM THERAPY

complement experimental measurements and generate systematically data ofphysical quantities as input for analytical treatment planning codes. For thefirst time, the MC code FLUKA was used to generate systematically physicalinput data for the generation of pencil beam and fluence kernels of a commer-cial TPS for scanned protons and carbon ions at HIT (Parodi et al., 2012) (andmore recently also at CNAO). In principle, all of the necessary input data couldbe obtained by measurements. However, obtaining measurements for 255 en-ergy steps (both with and without ripple filter), for four beam focus levels andten intensities delivered at HIT, would be in practice very time-consuming.Hence, it was decided to use FLUKA for the interpolation of these quanti-ties. The INFN-TPS project also aims at developing a TPS for proton andcarbon ion therapy which employs such MC-generated data from FLUKA andGEANT4 (Agodi et al., 2008). Additional areas of MC applications for ionbeam therapy encompass: beam line modelling (Paganetti et al., 2008), sup-port for dosimetric measurements and biological experiments, support for invivo treatment verification (see also section 1.1.5 and chapter 4) (Parodi et al.,2007a; Sommerer et al., 2009), risk-estimation for secondary cancer induction(Hultqvist et al., 2010; Jarlskog and Paganetti, 2008b; Yonai et al., 2012) andradioprotection. With ever-increasing computational possibilities and comput-ing power the use of MC in routine treatment planning, such as fully MC-optimized treatment planning becomes more and more feasible and reality.For instance, a recent GPU-based MC implementation reports a decrease ofcomputing time of as much as a factor of 5000 (Jahnke et al., 2012).

The MC codes which are presently employed for applications in ion beamtherapy were mostly developed for user communities which have distinct setsof requirements in terms of interaction regimes and accuracy, such as the high-energy physics or the radioprotection community. In view of their clinicalapplication, physics models of these MC codes need to be well tested andmeet the accuracy requirements necessary for their application in this domain.This concerns specially a very selected set of physics models in an energyrange which is relevant for ion beam therapy and includes the electromagneticmodels which describe the energy losses of ions slowing down and producedsecondary delta rays as well as nuclear interactions. The latter give rise tofluences of secondary charged fragments, neutrons, β+-emitters and prompt-γ’s, which modify the primary radiation field in terms of dose and its biologicaleffectiveness, result in out-of-field doses relevant for secondary cancer riskestimations, and allow for treatment monitoring and verification.

It is essential to note that even thorough benchmarking and validation ofphysics models cannot eliminate the need for fine-tuning certain parameters ofthe MC simulations in order reproduce optimally the delivered dose profiles fora specific facility. This is necessary since some parameters, such as the ioniza-

18

CHAPTER 1. PROJECT CONTEXT

tion potential, beam energy, momentum spread, and divergence, are generallynot precisely enough known (Parodi et al., 2012). It is practice for clinicalapplications to verify particle ranges and the conversion scheme of HUs byin-house measurements with real tissues and adjust simulations, if necessary(Jäkel et al., 2001b; Rietzel et al., 2007; Schaffner and Pedroni, 1998). Ab-solute range precisions for analytical treatment planning of about 1-3mm arefound, including also uncertainties from the CT images and CT conversionschemes. Analogous to analytical TPSs which calculate dose distributions inthe patient geometry with the water-equivalent approach, MC codes are val-idated and adjusted to dose measurements in water, tissue-like plastics, andtissues (Jarlskog and Paganetti, 2008a; Paganetti et al., 2008; Parodi et al.,2009a).

1.3 Thesis objectives and outline

MC codes are a valuable and versatile tool for ion beam therapy. Applicationsof MC codes to ion beam therapy pose a high demand in terms of accuracyfor the description of interactions for a very selected set of projectile-targetcombinations and projectile energies up to about 500MeV per nucleon, as dis-cussed in the previous section. This has triggered previous and ongoing effortsfor the improvement and validation of physics models implemented in the MCcodes as well as measurement campaigns for the collection of experimentaldata needed for their validation (Agodi et al., 2008; Linz et al., 2012, Ch. 7,p. 100). The first objective of this thesis work is to contribute to these de-velopment and validation efforts. For carbon ion therapy, charged fragmentsnotably modify dose distributions and RBE of the radiation field with respectto the dose and RBE distributions due to the primary carbon ions only (seealso figures 1.2 and 1.3). Hence, their fluences in the patient need to be pre-dicted correctly. For their use in carbon ion therapy, the accuracy of nuclearmodels of FLUKA and GEANT4 is assessed in paper I for the description ofcharged fragments. In paper VII, the current accuracy of reaction cross sectionmodels for ions in several popular MC codes is investigated more generallyfor a larger set of projectile-target systems and energies. Reaction cross sec-tions scale the occurrence of fragmentation reactions in the first place (see sec-tion 2.2.1). The microscopic patterns with which particles release their energyare, in addition to the macroscopic quantity absorbed dose, decisive for thebiological effects of a radiation field. Microscopic energy deposition patternsof the primary ions and secondary fragments produced in nuclear reactionsvary depending on the particle’s energy and its charge. Microdosimetric mea-surements are compared with FLUKA predictions using a configuration of thesimulation which was optimized regarding delta-ray production and transport

19

1.3. THESIS OBJECTIVES AND OUTLINE

settings (paper II and III). This comparison allows to evaluate the adequacyof the FLUKA physics models to describe electromagnetic processes for ionsand secondary delta-rays on the scale of microns. At the same time, it facili-tates the evaluation of the description of mixed radiation fields in phantoms ascreated by carbon ion beams. In a complementary way to measuring particlefluences, such a comparison allows to discriminate energy depositions fromprimary ions and secondary charged fragments to the spectra. Paper IV pro-poses and validates two original models describing the processes of Comptonscattering and positron annihilation at rest in media. These processes are rel-evant for treatment monitoring techniques using PET and prompt-γ imaging.Both models were developed as native FLUKA models and are included in theofficial release.

The second objective of this thesis is to contribute to the development ofthe prototype of a MC-based treatment planning (MCTP) tool (paper VI). Us-ing FLUKA, this research tool allows to perform single-field and simultaneousmultiple-field optimization for realistic treatment conditions using the patientCT as well as phantom geometries as used for dosimetric quality assurance.Treatment planning can be performed for ions with charges from 1 to 8 byeither optimizing absorbed dose or RBE-weighted dose. In the frame of thisthesis work, the focus was set mainly on: the development and implementa-tion of optimization algorithms, the extension of the tool from protons to otherions, and the testing and evaluation of its performances. Test plans were cre-ated and evaluated using a proton case for treatment conditions at the CNAOfacility and using a carbon ion case for treatment conditions at the HIT facility.For the latter case also a direct comparison with the certified TPS used at HITis presented. The MCTP tool is used also in a study which investigates thebiological robustness of ion therapy treatment plans (paper V).

This thesis is organized as follows: chapter 2 is dedicated to nuclear inter-action models used in MC codes and their validation for ion beam therapy; theimportance of microdosimetry for applications in ion beam therapy togetherwith the evaluation of the accuracy of FLUKA for such applications is dis-cussed in chapter 3; the applications of the newly-developed physics modelsfor Compton scattering and positron annihilation at rest are outlined in chap-ter 4; chapter 5 presents the prototype of the MCTP tool, detailing aspects ofthe implementation as well as example cases. Conclusions of this thesis andfuture perspectives can be found in chapter 6.

20

2. Validation of nuclearinteraction models implementedin MC codes

2.1 Introduction

The main reasons for the interest in the accurate modelling of nuclear frag-mentation reactions for clinical proton and ion beams include, as discussedin chapter 1: (1) the dependence of RBE on the ion charge, its energy anddelivered dose, (2) the modification of the dose distribution due to secondarycharged fragments (most notably in the tail after the Bragg peak), (3) the cor-rect prediction of out-of-field fluences of light charged fragments and neutronsfor secondary cancer risk estimations, (4) the use of products from nuclearreactions for in vivo treatment monitoring as well as (5) accurate dosimetry(correct prediction of fluence correction factors) and (6) shielding calculations(Lühr et al., 2012). For radiotherapy with carbon ions, the accurate modellingof the production and transport of secondary particles has an increased impor-tance as the radiation field created by carbon ion beams contains, contrary tothe case for proton beams, a large fraction of differently-charged fragmentswhich contribute to fluences and to the delivered dose. The dose fraction de-livered during carbon ion therapy by charged fragments in the PTV is of theorder of some tens of percents (Kempe et al., 2007; Matsufuji et al., 2003)and in the tail after the SOBP even all of the dose is delivered by fragments.An accurate prediction of these fragment fluences by MC simulation tools istherefore necessary. This chapter outlines MC modelling approaches for thedescription of nuclear interactions, using FLUKA as representative example,and discusses the current accuracy of the nuclear models implemented in MCcodes for ion beam therapy.

2.2 Hadronic interaction models

Since most simulations in the frame of this thesis are performed using theFLUKA code, the following description of hadronic interaction models for

21

2.2. HADRONIC INTERACTION MODELS

MC techniques concentrates on the models from the FLUKA code. However,it can be considered to be representative also for other codes, as most state-of-the-art multi-purpose MC implementations follow a similar modelling schemefor the description of the hadronic cascade and subsequent reaction processes(even though the actual models, their implementation and validation can bevery different). For a description of the hadronic models of GEANT4, thereader is referred to the GEANT4 physics reference manual (Geant4, 2012)and paper I as a starting point. Due to the large complexity of the hadronicinteractions and used modelling approaches, the present section aims merelyto give an overview of main aspects of the approach pursued for MC codesand outlines some of the modelling concepts used, largely on a phenomeno-logical basis. A more detailed description can be found in literature such asBallarini et al. (2004), Ferrari and Sala (1998), Ferrari and Sala (2002), Fer-rari et al. (2005) and references therein. The approach which is used gen-erally by MC codes to describe hadronic interactions, FLUKA being no ex-ception, is referred to as ‘microscopic’ approach, with each single step hav-ing a physics-motivated basis. As other physics models of FLUKA, also thehadronic FLUKA models follow the modelling strategy to use, whenever pos-sible, the approach: ‘theory-driven and benchmarked with data at the single-interaction level’ and predictions are obtained generally with a minimal set offree parameters which are fixed for all energies and target-projectile combina-tions (Ferrari and Sala, 2002). Hence, differential hadronic cross sections arenot explicitly tabulated, except for neutrons with E < 20 MeV (see below). In-stead, reaction channels and energies are sampled from physics models servingas event generators. Elementary hadron-hadron scatterings (including elastic,charge and strangeness exchanges) are described on the basis of phase-shiftanalysis (Ferrari et al., 2005). For the description of inelastic hadron-nucleoninteractions, depending on the projectile energy two model families are used:models based on individual resonance production and decays (isobar model, upto about 3-5 GeV) and those based on parton/quark string models (up to sometens of TeV) (Ferrari and Sala, 1998, 2002). In the case of FLUKA beyond afew tens of TeV, the DPMJET model based on the Dual Parton Model (DPM)is used (Roesler et al., 2001). The following description will concentrate onthe lower energetic nuclear interactions models with relevance for ion beamtherapy. For nucleon-nucleon interactions at these energies, pion production,as lowest-energy hadron production channel, is negligible. Pion production innuclei can be observed as low as 150 MeV (due to Fermi motion) but becomesimportant mostly only for higher energies (larger than 700 MeV). Hence, nu-clear interactions consist at their basis of elastic nucleon-nucleon scatterings,which can result in nucleon and light fragment emissions while leaving theremanent concerned nuclei in excited states. The general scheme adopted for

22

CHAPTER 2. VALIDATION OF NUCLEAR INTERACTION MODELS

the simulation of hadron-nucleus interactions in FLUKA can be described bythe following chain of stages which switch from an initially dynamical to astatistical treatment (Ferrari and Sala, 2002):

• Cascade stage: Glauber-Gribov cascade (at high energies > 1 GeV) andintra-nuclear cascade (INC)

• Pre-equilibrium stage

• Equilibrium de-excitation stage: evaporation, fission, fragmentation(Fermi break-up) and γ de-excitations

• Radioactive decays

For the proper description of nucleus-nucleus (A-A) interactions, as impor-tant for ion beam therapy, the last three stages are essentially the same. Instead,for the initial cascade stage an extension of the INC approach or the use of al-ternative models, such as the quantum molecular dynamics (QMD) models ormodels based on the Boltzmann-Uehling-Uhlenbeck equation (Aichelin, 1991;Bertsch and Gupta, 1988) is required. In the case of FLUKA, A-A interactionsare treated using an adaption (Andersen et al., 2004) of the relativistic QMD(rQMD-2.4) model (Sorge et al., 1989a,b). The rQMD approach is applied inFLUKA for projectile energies in the range from about 100 MeV per nucleonto 5 GeV per nucleon. At lower energies, FLUKA uses a model based on theBoltzmann Master equations (BME) theory (Cerutti et al., 2006) for the ini-tial description of A-A interactions. The non-elastic interaction probability forA-A interactions is given by the total reaction cross sections σR. Phenomeno-logical data-driven models are generally used for the computation of σR in MCcodes.

It is important to underline the mutual influence of the different stages inthe whole chain with which the description of hadronic interactions are mod-elled in MC codes. Adequate predictions of resulting quantities of interest areobtained only in a synergistic interaction of all or some of these stages andtheir respective models. The following two sections describe the total reactioncross sections and introduce the modelling stages and models which are usedby FLUKA to describe hadronic interactions.

2.2.1 Total nuclear reaction cross sections

The total nuclear reaction cross section σR is defined as the total nuclear crosssection σtot minus the elastic nuclear cross section σel of two colliding nucleiat a given energy

σR = σtot−σel . (2.1)

23

2.2. HADRONIC INTERACTION MODELS

For energies below a few GeV down to the Coulomb barrier, FLUKA usesa parametrization of total reaction cross sections σR (Andersen et al., 2004)which is loosely based on the semi-empirical model from Tripathi et al. (1999).This parametrization includes various extensions to a simple geometrical de-scription of reaction cross sections, given by σR = πr2

0 · (A(1/3)P +A(1/3)

T )2. Inthe approach by Tripathi et al., σR is modelled as

σR = πr20 ·(

A(1/3)P +A(1/3)

T +δE

)2·(

1−RcB

Ecm

)Xm , (2.2)

with r0 = 1.1 fm, the projectile AP and target AT mass number, Rc being aCoulomb multiplier and the Coulomb barrier B, given by

B =α hcZPZT

R, (2.3)

with the atomic number of the projectile ZP and target ZT and R being a rep-resentative nuclear radius at which the Coulomb barrier height is evaluated.Furthermore, Ecm is the kinetic energy in the centre-of-momentum system, Xmis a low energy optical model multiplier, h is the reduced Planck constant, α

is the fine-structure constant and c is the speed of light. δE is an expressiondepending on the kinetic energy of the projectile energy E that includes theeffects of Pauli blocking and transparency. The original model parameters byTripathi et al. were modified and improved for FLUKA to match newer andoften more accurate sets of cross section measurements and, at the high en-ergy end above 3 GeV per nucleon, Glauber model predictions computed byDPMJET. The lighter system is by convention always chosen to be the projec-tile, using, if needed, inverse kinematics for the calculation of σR. Figure 2.1shows σR as predicted by FLUKA for carbon ions for the most frequent ele-ments present in the human body.

For completeness, it should be noted that at relativistic energies (& 1 GeV)and for heavy systems, mutual electromagnetic excitations by the exchange ofvirtual photons between the colliding nuclei start to become important (Bal-larini et al., 2004; Hill et al., 1988). These Coulomb excitations may result inthe emission of particles for one or both of the nuclei, referred to as electro-magnetic dissociation. The cross section for this process has to be added tothe nuclear reaction cross section described above. However, electromagneticdissociation can be neglected for energies and light systems which are relevantfor ion beam therapy.

2.2.2 Modelling of hadronic interactions

Hadron-nucleus interactions FLUKA describes hadron-nucleus interact-ions for energies up to 5 GeV and down to the reaction threshold (down to

24

CHAPTER 2. VALIDATION OF NUCLEAR INTERACTION MODELS

0 200 400 600 800

1000 1200 1400 1600 1800 2000 2200 2400

50 100 150 200 250 300 350 400 450 500

Nu

clea

r re

acti

on

cro

ss s

ecti

on

[m

b]

Energy [MeV/n]

Carbon beamHCNOP

Ca

Figure 2.1: Reaction cross sections σR of carbon ions at therapeutic energies aspredicted by FLUKA for the most frequent elements (by weight) present in thehuman body.

20 MeV for neutrons, see below) with the pre-equilibrium-cascade model PEA-NUT (Pre-Equilibrium Approach to NUclear Thermalization) (Fassò et al.,1995a,b; Ferrari and Sala, 1994, 1998). Higher energetic hadrons (> 5 GeV)are treated using the Glauber-Gribov cascade (Battistoni et al., 2006). Thetwo essential components of the PEANUT model at lower energies are a gen-eralized intra-nuclear cascade (GINC) for the description of the initial h-Anon-elastic interaction stage and the subsequent pre-equilibrium stage modelwhich de-excites the hot nuclear components by emission of nucleons and lightnuclei (A < 5) until thermal equilibrium is reached (see next paragraph). TheGINC is based on the original INC modelling approaches but includes severalimprovements, specifically for the classically unreliable treatment of the lowenergy region (< 100−200 MeV) and the higher energy region (> 2−3 GeV).Some of the basic concepts and features of INC models can be summarized asfollows1. INC models are based on a series of two-body collisions with semi-classical trajectories. Trajectories can be curved when taking into account theeffects of projectile- and energy-dependent nuclear and Coulomb mean fields(referred to as refraction and reflection effects). The two-body collisions them-selves are described based on free hadron-nucleon cross sections, while takinginto account the local density of the nuclear medium. Secondary particles pro-duced in the cascade of collisions are treated the same way as primaries. Thecascade simulation proceeds until all particles have interacted, escaped or arebelow a certain energy cut-off. The target nucleon motion in the nuclear well isdescribed by a Fermi gas. This has an effect on both the centre-of-mass energyand momentum when calculating hadron-nucleon interaction cross sections as

1A more detailed description can be found in Ferrari and Sala (1998) and Battis-toni et al. (2007).

25

2.2. HADRONIC INTERACTION MODELS

well as the production of the final state particles by the selection of nucleonswhich are close in momentum space. In spite of the modelling of nucleoncollisions being based on the use of semi-classical trajectories, quantum me-chanical effects, such as Pauli blocking and formation zone (materializationtime) as well as nucleon anti-symmetrization, coherence length and hardcorenucleon correlations are accounted for by the PEANUT GINC model (Fassòet al., 1995a; Ferrari and Sala, 2002). In addition, nuclear densities specific toprotons and neutrons are chosen, binding energies are updated after each parti-cle emission and energy and momentum conservation is ensured including therecoil of the residual nucleus (Battistoni et al., 2007).

Pre-equilibrium stage The pre-equilibrium stage starts when all non-nucleonshave been emitted or decayed and all nucleons are below 50 MeV. It usesa statistical exciton-based modelling approach in which the state of the nu-clear configuration is characterized by the number of protons and neutrons, byparticle-like (nucleons excited above the Fermi level) and hole-like (holes inthe Fermi sea) excitons and the compound nucleus excitation energy and mo-mentum (Ferrari and Sala, 2002). A step-wise evolution (Markovian process)of this system redistributes the available excitation energy among the excitonswhile increasing the total number of excitons. Excitons of a certain type andenergy, have at each step a certain probability to be emitted in the continuum.

De-excitation stage At the end of the pre-equilibrium stage the residual nu-cleus is assumed to be in an equilibrium state, with the excitation energy beingshared among a large number of nucleons. The compound nucleus is now char-acterized by its mass, charge and excitation energy (typically from few MeV toseveral tens of MeV). The emission of nucleons and light fragments is possi-ble and can be described for heavier compound nuclei as an evaporation froma hot system based on the original statistical approach by Weisskopf-Ewing(Weisskopf, 1937; Weisskopf and Ewing, 1940). The FLUKA implementa-tion of this model is extended to allow the emission of about 600 possiblenucleons and nuclei in various excited states with A < 25. Evaporation is incompetition with fission. Fission is the consequence of a collective deforma-tion of the compound nucleus due to the excitation energy (Bohr and Wheeler,1939). However, it is only important for heavy remnants (Z > 70). For lightnuclei, the statistical assumptions and the description by sequential emissionused for evaporation are less sound and the fragmentation of the compoundnucleus is described by FLUKA using the so-called Fermi break-up model in-stead (Epherre et al., 1969; Fermi, 1950). In the frame of this model, lightexcited nuclei with A ≤ 17 can break into two or more fragments. The prob-ability for the branching ratios into all possible stable and unstable fragments

26

CHAPTER 2. VALIDATION OF NUCLEAR INTERACTION MODELS

including excited states is derived based on phase space, spin and Coulombbarrier considerations (Ferrari and Sala, 2002).

De-excitation by the emission of prompt-γ rays is in competition with theprocesses described above (to be precise, this is already the case for the pre-equilibrium stage). The branching ratio for γ de-excitation is generally negli-gible as long as evaporation of nucleons and nucleon-based fragments is en-ergetically possible, but it becomes important for dissipating the remainingexcitation energy once other processes are energetically no longer possible.FLUKA describes prompt-γ de-excitation as emission in the continuum and,in the case of lower excitation energies, accounting for discrete nuclear levels.The de-excitation stage ends leaving the residual nuclei in their ground state.

Radioactivity Radioactive decays of the residual nuclei are performed ac-cording to tabulated data of decay modes. The decay modelling is based mostlyon tabulations from ENSDF1 (Evaluated Nuclear Structure Data File) (Bhat,1987).

Nucleus-nucleus interactions Interactions for which the potential generatedby the projectile hadron can be neglected in first order, can be generally welldescribed by the INC approach, especially when the target nucleus is large.For such cases the projectile can be thought to move in the static mean fieldof the target nucleus while considering short-range correlations between thenucleons. The extension of the GINC approach to light projectile nuclei is rel-atively straightforward for energies higher than the binding energy of the lightnucleus (Ferrari and Sala, 1998). To the contrary, for the description of A-Ainteractions including heavier systems for both projectile and target nuclei, thepotential field of the projectile nucleus should be taken into account. Hence, adescription in which each nucleon of the colliding nuclei feels a global forcedue to the nucleons of both nuclei is more appropriate. The QMD model familyaccounts for this effect. On the basis of the quantum mechanical Hamiltonianequations, QMD models use a dynamic modelling of the nuclear field amongthe nucleons while describing each nucleon in form of a Gaussian wave func-tion. For the description of A-A interactions, FLUKA is currently interfacedwith a modified version of the rQMD-2.4 code (in cascade mode) (Andersenet al., 2004) which is based on the original work by Sorge et al. (1989a; 1989b).Remaining excited pre-fragment leftovers are treated using the PEANUT de-excitation models. The QMD approach can provide reliable results down toapproximately 100 MeV per nucleon. In the region below this threshold2 a

1http://www.nndc.bnl.gov/ensdf/2As for other FLUKA models no sharp energy cut is used in the FLUKA produc-

tion version, but rather an energy range in which a mixed sampling from both model

27

2.3. BENCHMARKING NUCLEAR MODELS

model based on the BME theory is used to describe partial or complete fu-sion of two colliding ions into a composite nucleus and the subsequent nuclearthermalization process (Andersen et al., 2004; Cerutti et al., 2006). The BMEsdescribe the time evolution of the momentum distributions of the nucleons inthe composite nucleus based on two-body elastic scattering and particle emis-sion in the continuum (Cerutti et al., 2006). Since emission probabilities haveto be determined by computational intensive integrations of the BMEs, a pre-computed database of ejectile probabilities for the complete fusion of differentnuclei combinations is used for fast MC sampling in FLUKA by interpola-tion between the pre-computed systems whenever possible. If interpolationfrom the pre-computed database is not feasible for a given composite nucleus,the PEANUT pre-equilibrium model is invoked for thermalization. Once theresidual nuclei are in an equilibrium state, the PEANUT de-excitation modelsare applied.

Neutron transport at low energies The description of transport and inter-actions of neutrons below 20 MeV down to thermal energies is based on amulti-group approach which uses cross section data for each of the 260 groups(including 31 groups for thermal neutrons) starting from standard evaluateddatabases (ENDF/B, JEF, JENDL, etc.) (Battistoni et al., 2007; FLUKA Col-laboration, 2012).

2.3 Benchmarking nuclear models implemented in MonteCarlo codes

2.3.1 Reaction cross sections

Paper VII compares predictions of reaction cross sections σR by FLUKA,GEANT4 and PHITS to a large set of experimental data for a variety of sys-tems. The presented FLUKA cross sections include some recent modellingimprovements for systems with light projectiles or targets (1 < A ≤ 4). Thelargest differences in predictions of σR for the tested MC codes are found forthe region below 100 MeV per nucleon. These differences are partially due tothe different accounting for Coulomb effects in the codes. Fragmentations atthese lower energies contribute however only with a small fraction to fluencesin the patient. For higher energies, σR varies more slowly with energy. Still,discrepancies for some projectile-target combinations up to some tens of per-cents are observed. Often there is no or not accurate-enough experimental data

approaches is performed to allow for the smooth transition between the predictions ofthe QMD model and the BME model.

28

CHAPTER 2. VALIDATION OF NUCLEAR INTERACTION MODELS

to estimate the adequacy of the models in regions where the model predictionsdiffer. Comparisons of σR predictions of FLUKA and GEANT4 models forcarbon ions at therapeutic energies on light targets show differences mostlywithin about 10% and are in good agreement with measurements, see paper I.For carbon ions on hydrogen, larger discrepancies were found for energiesbelow 20 MeV per nucleon. The GEANT4 cross section does not decrease to-wards lower energies, as would be expected due to the Coulomb barrier. Thesedifferences were subsequently reduced by the GEANT4 developers by modi-fying the modelling of Coulomb effects in the code (Quesada, 2010).

2.3.2 Charged fragments

The prediction of charged fragment production for carbon ion beams by FLUKAand GEANT4 was examined and compared with available experimental dataof both thin and thick target measurements of integral and differential (angleand energy) quantities, see paper I. Differences between measurements andMC predictions of some tens of percents were found for integral quantities(such as integral charge-changing cross sections and fragment yields at differ-ent depths in a water phantom) for both codes. FLUKA reproduces the overallshape of the experimental angular yields rather well. Larger angles (4− 10◦)of fragments with Z > 1 tend to be slightly underestimated and there are somediscrepancies in the forward angles. Angular yields predicted by GEANT4 nu-clear models tend to underestimate experimental data for small angles whereaslarger angles tend to be overestimated for fragments with Z > 2. Disregardingthe integral yield, the experimental energy distribution of the produced frag-ments is generally well described by models from both codes for the angleswhich contribute most to the integral fragment yields.

2.4 Discussion and perspective

The performed benchmarks demonstrate that the nuclear models of both FLUKAand GEANT4 can predict the attenuation of the primary carbon ion beam andsecondary charged fragment yields in reasonable agreement with experimentaldata. It should be noted here that the discussed differences are expected to berelevant for imaging applications, for studies examining cancer risk due to out-of-field doses by light charged fragments and neutrons and, to some extend,for the computation of RBE in the treated volume (Mairani et al., 2010). In-stead, the observed differences have little impact in terms of the total absorbeddose delivered by a therapeutic carbon ion beam and it was shown that thedepth and lateral dose profiles of carbon ion beams can be generally simulatedwithin the therapeutically required accuracy by MC codes like FLUKA, see

29

2.4. DISCUSSION AND PERSPECTIVE

for instance Parodi et al. (2012). According to a simple calculation, uncertain-ties in charged fragment fluences can be estimated to introduce an uncertaintyin RBE-weighted doses in the high dose region of a carbon ion treatment fieldof a few percents at most (paper I). A recent study performing a more detailedevaluation also concludes that there is a mild impact on the RBE in the highdose region for typical uncertainties in the nuclear fragment spectrum (Lühret al., 2012). Uncertainties in the prediction of RBE values for patient treat-ment can be assumed to be generally much larger for carbon ion therapy, seepaper V.

Even though the attenuation of the primary carbon ion beam and integralyields of charged fragments with depths are reasonably well predicted, thereare still substantial differences between the measurements and MC predictions.This is on one hand due to the limited accuracy of the MC models describingthe complex chain of nuclear interactions (see section 2.2) for the relevantenergy regime and projectile-target combinations. There have been and areseveral efforts to improve the accuracy of nuclear models implemented in MCcodes for ion beam therapy, for example by tuning them based on experimentaldata (e.g. Hansen et al., 2012). However, such efforts are presently hindered bythe limited availability of adequate experimental data for systems and energiesof interest for ion beam therapy and the often sub-optimal accuracy of existingmeasurements. As pointed out earlier, the validation and tuning of nuclearmodels implemented in MC codes is preferably performed using data at the‘single interaction level’ (i.e. cross section data) and not integral data, such ascoming from thick target experiments. Since the therapeutic application of ionbeams other than protons and carbon ions is expected to have a clinical revivalin near future1, the existing measurement data base should also be enrichedby measurements of other light ion beams, such as lithium and oxygen. Thiswould allow to validate the applicability of nuclear models implemented inMC codes for these ions more accurately.

As a contribution to complement the existing experimental data base offragmentation measurements, the FIRST (Fragmentation of Ions Relevant forSpace and Therapy) experiment aims to obtain accurate cross section measure-ments with a sophisticated detector set-up for projectile-target combinationsand energies relevant for ion beam therapy (Abou-Haidar et al., 2012; Golo-sio et al., 2011; Pleskac et al., 2012). Using thin targets, the detector set-upis designed to measure cross sections of charged fragments, differential in en-ergy and angle, with a large angular acceptance (≈ 60◦). For forward angles,also neutrons are measured and fragment masses can be resolved. The authorcontributed to this project by developing the MC simulation of the complete

1Initial clinical use of some of these ions was conducted at Lawrence BerkeleyLaboratory (Berkeley (CA), USA) several decades ago (Petti and Lennox, 1994).

30

CHAPTER 2. VALIDATION OF NUCLEAR INTERACTION MODELS

experimental set-up of FIRST (Pleskac et al., 2012) and parts of the recon-struction software for particle tracking in magnetic fields and track matchingas well as by assisting detector test measurements and data taking. The FIRSTexperiment took data in summer 2012 at GSI for C-C at 400 MeV per nu-cleon and C-Au at 400 MeV per nucleon. Data analysis and evaluation of theMC simulation and reconstruction software are currently still ongoing. Furthermeasurements including other ions and energies are intended in future.

Based on an accurate experimental data base, nuclear models implementedin MC codes should in the years to come continue, alongside with analyticalmodelling approaches, to improve the quality of their predictive capabilities inthe description of nuclear fragmentation processes for their application to ionbeam therapy.

31

2.4. DISCUSSION AND PERSPECTIVE

32

3. Simulation of microdosimetricquantities for ion beams

3.1 Introduction

The spatial distribution of energy deposition events of radiation in biologicaltargets is a factor influencing the complexity of the resulting damages and thecapacity of the cells to repair them correctly. Hence, the biological effective-ness of an ion for cell inactivation is closely related to its track structure andthe spatial distribution of energy deposition events along a primary track iscrucial for quantifying damage induced in biological targets, as mentioned insection 1.1.1. For ions, energy deposition occurs in the track core (radius onthe nm-scale) by the primary ion and by secondary electrons (delta-rays, Augerand Coster-Kronig electrons) which travel larger distances from the track core.Energy loss of the primary ion and production cross sections of secondaryelectrons depend on energy and charge of the primary ion. This causes a trackstructure which is specific to the energy and charge of the primary ion.

This chapter outlines the microdosimetric approach for measuring radia-tion quality and spatial distribution of energy deposition events together withaspects of commonly-used microdosimetric measurement devices and radiobi-ological models for radiotherapy based on microdosimetric quantities. In thelast part of the chapter, the capacity of FLUKA for the simulation of micro-dosimetric quantities for ion beams is discussed.

3.2 Linear energy transfer and microdosimetry

In principle, track structure provides a very detailed basis for radiobiologicalmodelling and there are efforts to model radiobiological effects through the de-tailed description of the track structure and subsequent chemical and biologicalprocesses (e.g., Chauvie et al., 2006; Nikjoo et al., 2008, 2006). However forclinical applications, most radiobiological modelling approaches try to reducethe great amount of information and parameters to a more easily manageableform and biological effects are described based on concepts which aim to givea measure of characteristics of the track structure, such as LET, radial dose

33

3.2. LINEAR ENERGY TRANSFER AND MICRODOSIMETRY

profiles (amorphous track structure) and lineal energy (e.g., Cucinotta et al.,1999; Elsässer et al., 2010; Hawkins, 2004; Paganetti et al., 1997; Rollet et al.,2010; Wedenberg et al., 2010). LET and lineal energy serve also as quantitieswhich are used to estimate the biological effectiveness of measured radiationdoses for radioprotection (ICRP, 2007).

The unrestricted LET is defined as electronic stopping power (ICRU 16,1970) and describes the energy loss due to collisions with atomic electronsoccurring along the trajectory of a primary charged particle. The LET conceptwas developed to serve as an approximate description of the energy transfer bycharged particles while recognizing at the same time its serious limitations fordescribing biological effects (ICRU 16, 1970; ICRU 36, 1983, Ch. 4):

• Particles with different charges and velocities can have the same LET.However, the energy distribution of the delta-rays is dependent on thevelocity of the particle and can result in a significantly different spatialenergy distribution along the primary track.

• The LET, being a mean value, does not consider the random nature ofenergy losses along a track. Consequently, it cannot alone describe theenergy deposition in a small volume in which the number of interactionsis small and thus fluctuating.

• LET does not consider the length of a primary track relative to a finitetarget structure. The limited range of charged particles influences theenergy deposition in finite sites as the LET may change significantlyalong the path through the site and as the track can begin or end in thesite.

It was experimentally shown that biological effectiveness has an inverse rela-tionship with velocity of an ion (Goodwin et al., 1996) and that the RBE at agiven LET decreases as a function of increasing ion mass and charge for en-ergies of interest for ion beam therapy (Furusawa et al., 2000). An attempt toaccount for radial dimensions of a particle track was carried out by introduc-ing the concept of restricted linear energy transfer (LET∆) which includes onlyenergy losses due to delta-rays with an energy smaller than ∆.

The field of microdosimetry, with pioneering works from Rossi and Rosen-zweig (1955) and co-workers, sought to overcome some limitations of the LETconcept in describing biological effects by defining stochastic quantities re-lated to energy depositions in microscopic volumes with materials relevantfor biological considerations (site concept) (Guetersloh et al., 2004; ICRU 16,1970; ICRU 36, 1983; Kellerer and Chmelevsky, 1975, and references therein).Tissue-equivalent proportional counters (TEPC) are detectors which mimic di-mensions and medium characteristics similar to a mammalian cell nucleus.

34

CHAPTER 3. SIMULATION OF MICRODOSIMETRIC QUANTITIES

They respond to particles passing the sensitive volume of the TEPC as well aspenetrating delta-rays from particles passing close to the sensitive TEPC vol-ume. TEPCs measure energy imparted per event and derived quantities suchas the lineal energy y = ε/l, which is defined as the quotient of energy im-parted to matter in a volume by a single energy-deposition event ε and themean chord length in that volume l (ICRU 36, 1983). Similarly, the specificenergy is defined by z = ε/m, where m is the mass of the microscopic volume.

Two commonly used averages of lineal energy spectra are the frequencymean lineal energy

yF =∫

∞

0y f (y)dy , (3.1)

where f (y) is the (normalized) frequency probability density of y, and the dosemean lineal energy

yD =∫

∞

0yd(y)dy =

1yF

∫∞

0y2 f (y)dy , (3.2)

where d(y) is the (normalized) dose probability density. Averages for specificenergy z are defined analogously and mean specific energy in a volume is givenby

zD =

∫∞

0 z2 f (z)dz∫∞

0 z f (z)dz, (3.3)

where f (z) is the frequency probability density of z.

3.3 Measurement devices for microdosimetry

TEPCs are the principal instrument used for measuring microdosimetric quan-tities (ICRU 36, 1983). They generally mimic volumes of tissue with dimen-sions on the µm-scale by a cavity with a size typically in the cm-range which isfilled with a low-density tissue-equivalent gas. However, measurements withTEPCs down to tens of nanometres have been reported (Cesari et al., 2002)and using the variance-covariance method (Kellerer and Rossi, 1984), mea-surements down to 10 nm have been performed (e.g. Lillhök et al., 2007). Inrecent years silicon solid state detectors are increasingly used as alternativedevice for the measurement of microdosimetric quantities (Agosteo and Pola,2011). For the commonly used walled TEPC, the gas cavity is surrounded bya layer of tissue-equivalent plastic. It is known that the use of these walledTEPCs leads to distortions of experimentally obtained microdosimetric distri-butions with respect to distributions as expected in a microscopic tissue volumewhich are due to the density differences between the cavity and the surround-ing wall and the large size of the cavity. The effects leading to these distortions

35

3.4. MICRODOSIMETRY APPLICATIONS

are termed wall effects and can be classified in four different types accordingto ICRU 36 (1983, Ch. 5.2.2).

Delta-ray effect: Delta-rays, produced by a charged primary particle outsideof the cavity, enter the cavity together with the particle itself. Some ofthese delta-rays would miss the cavity if it were of microscopic dimen-sions or only some of the delta-rays and not the charged particle wouldproduce an event in the cavity. This effect was found to be important forheavy charged particles. For protons at 5MeV, the frequency of suchdouble events due to delta-rays is about 15%.

Re-entry effect: An electron re-enters the cavity due to scattering processes.This would not necessarily be the case for a microscopic region. Thiseffect is only relevant for primary electrons up to about 1MeV.

V effect: Several charged particles which are produced by a nuclear inelasticinteraction, typically at forward angles, impinge simultaneously on thecavity. Again, this might not be the case if the considered region wereof microscopic dimensions since the respective angles would have to bemuch smaller. The ‘V effect’ can lead to very large energy depositionsand may therefore have a noticeable impact on yD.

Scattering effect: Uncharged primary particles may produce charged parti-cles in interactions which are close enough so that several of these par-ticles enter the cavity. Probably only one of these particles would enterthe region for the same event if it were of microscopic dimensions.

All of the wall effects have in common that they increase the energy deposi-tion per event in the TEPC by the superposition of energy deposition eventswhich would not occur simultaneously in a medium of uniform density. Theimportance of the ‘V effect’ for lineal energy spectra of TEPC measurementsin phantoms irradiated with a carbon ion beam at therapeutic energies is in-vestigated in paper III. Wall effects can be avoided when using micrometricsensitive volumes. This is the case for solid state detectors (Agosteo and Pola,2011). Alternatively, validated simulation tools can be used to disentangleartefacts due to wall effects in the lineal energy spectra as shown in paper III.

3.4 Microdosimetry for radiobiological modelling and ap-plications in radiotherapy

Microdosimetric quantities such as lineal energy y and the momenta of thespectra yF and yD can be used to estimate RBE for therapeutic radiation fields.

36

CHAPTER 3. SIMULATION OF MICRODOSIMETRIC QUANTITIES

Several approaches have been proposed in the last decades that relate single-event spectra to biological effectiveness. A method which was originally de-veloped for neutron therapy by Menzel, Pihet, Wambersie and co-workers(Loncol et al., 1994; Menzel et al., 1990) determines RBE with a biologicalweighting function r(y) which characterizes the RBE as a function of y. Ac-cording to this approach RBE is calculated by

WRBE =∫

r(y)d(y)dy , (3.4)

where d(y) is the single-event dose probability density, see (3.2). The approachcan be also employed to predict RBE for therapeutic proton and ion fields(Gerlach et al., 2002; Paganetti et al., 1997; Rollet et al., 2010).

Lindborg and Grindborg (1997) suggested an approximative relation be-tween the ratios of the linear terms α of the linear-quadratic (LQ) model andthe yD-ratios for a reference radiation x and a test radiation i for early reactingtissues and tumours

WRBE ≈αi

αx· 1

1+ dx(αx/βx)

≈yD,i

yD,x· 1

1+ dx(αx/βx)

, (3.5)

with the fractionation dose dx. They found this relation to be approximatelyfulfilled for yD measured for spheres of a few nanometres (≈ 10nm). Themodel was applied to neutron, proton and carbon therapy beams (Hultqvistet al., 2010; Lillhök et al., 2007). A more recently developed model attemptspredictions of cell survival by a nanodosimetric-based LQ model for mixedradiation qualities (Wang et al., 2007).

The MKM (Hawkins, 1994, 2004), previously mentioned in section 1.1.1,provides a mechanistic description of cell damage by different radiation quali-ties. It relates cell survival for different radiation qualities to stochastic energydepositions in domains with a dimension of some microns. The cell nucleusis divided in microdomains with radius RD (∼ 5 µm) and the sum of lesionsin the microdomains is related to cell survival: Lnucl = − ln(S). The cellular-population-averaged number of lesions in the nucleus Lnucl is obtained by

Lnucl = (α0 +βxzD)D+βxD2 (3.6)

where D is the dose and the mean specific energy of the domain zD, see (3.3).As a result of the derivation of the MKM formalism, βx is constant for differ-ent radiation qualities and is set to the quadratic parameter of the LQ modelfor the reference radiation, that is x-rays. α0 denotes the linear parameter ofthe LQ model in the limit of LET→ 0. For practical purposes, α0 can be as-sumed to be approximately equal to the αx for x-rays. zD can be obtained by

37

3.4. MICRODOSIMETRY APPLICATIONS

microdosimetric measurements with TEPCs (Kase et al., 2006), by amorphoustrack structure codes (Kase et al., 2008) or by MC simulations (Sato et al.,2009). In the recently improved version of the MKM model (Kase et al., 2008,2006), zD is replaced by the saturation-corrected specific energy z∗D (ICRU 36,1983) and accounts for the decrease of RBE due to the overkill effect. Thisextension of the MKM was validated recently for a large range of LETs anddifferent ions with a good agreement for in vitro cell experiments (Kase et al.,2008; Sato et al., 2011). Further work with the aim of a clinical applicationof the MKM is ongoing at NIRS (National Institute of Radiological Sciences,Japan). A challenge for clinical application might be the fact that the domainsize is treated as a free RD parameter (Linz et al., 2012, Ch. 8).

Some controversial issues of microdosimetric approaches for the radiobi-ological modelling of RBE and related quantities for ion beam therapy havebeen raised.

• Biological response to radiation is assumed to involve mechanisms inresponse to radiation damages on the scales down to nanometres (e.g.DNA damage in form of single-strand breaks and double-strand breaks).Spatial energy depositions in micrometre dimensions can be assumednot to appreciate those effects adequately (IAEA, 2008, Annex II).On the other hand, the success in describing cellular responses of thepreviously-mentioned radiobiological modelling approaches, which useenergy deposition patterns on the scale of microns, indicate that impor-tant effects for biological response can be reasonably accounted for.

• Vice versa, one can argue that aforementioned radiobiological modellingapproaches which rely only on nanodosimetric quantities for the pre-diction of biological responses, fail to appreciate mechanisms on largerscales, such as the importance of giant loops, as proposed by the frameof LEM-IV (Linz et al., 2012, Ch. 8).

• An argument concerning the practicability of microdosimetric approachesusing biological weighting functions is that, since those weighting func-tions are generally obtained by experimental data and are strictly onlyvalid for a given biological endpoint and dose level, it would require atremendous effort to establish databases required for all clinically rele-vant situations (IAEA, 2008, Annex II).

Apart from their usefulness for radiobiological modelling, microdosimet-ric quantities, such as lineal energy, can serve as a measure for homogeneityof radiation quality for therapeutic ion beam treatment fields. As suggested inpaper V, they could be of use for optimization procedures in ion beam therapy.

38

CHAPTER 3. SIMULATION OF MICRODOSIMETRIC QUANTITIES

Microdosimetric quantities are superior to LET specifically for the character-isation of radiation fields which consist of mixed ion species, as discussed insection 3.2. A further application of microdosimetric measurements is thatthey can provide a complementary approach to particle fluences to charac-terize the radiation field produced by ions. Contributions of various chargedfragments to the lineal energy spectra can be observed in measurements (Endoet al., 2007, 2010; Martino et al., 2010; Matsufuji et al., 2003) and simulations(paper III).

3.5 Accuracy of FLUKA simulations of lineal energy forion beams

In order to use microdosimetry for ion beam therapy and radiobiological mod-elling, physics models and computational tools for obtaining microdosimetricquantities are needed. Besides simplified track descriptions, such as amor-phous track structure models and parametrizations based on measurements,MC codes can be of use to obtain these quantities (Rollet et al., 2010; Satoet al., 2009). Additionally, MC simulations are also flexible enough to supportthe design optimization, development and test measurements of new micro-dosimetric measurement devices (e.g. Agosteo and Pola, 2011; Rollet et al.,2012). For such applications, paper II investigates the accuracy and adequateconfiguration of FLUKA for simulating lineal energy distributions and theirmomenta, yF and yD, for ion beams for volumes with dimensions of the order ofone micron. The prediction of these quantities requires the correct modellingof various electromagnetic processes including notably ionization and excitat-ion of atomic electrons, which give rise to the electronic stopping power andenergy loss straggling, as well as the production and transport of secondaryelectrons. This work analyses the accuracy and optimal configuration withwhich FLUKA can be used to predict energy imparted on microscopic scalesby ions, such as measurable by TEPCs. It demonstrates that FLUKA can beused to describe energy deposition patterns of ions in TEPC cavities and that itprovides an adequate description of the main features of lineal energy spectra.According to the results one can expect to reproduce yF and yD within ±10%of the measured values for ions with atomic numbers ranging from 1 to 26 andenergies from 80 up to 1000 MeV per nucleon. The root mean square (RMS)deviation between FLUKA and measurements is 4.7% for yF

1 and 4.4% for yD.It is also demonstrated that radial dose profiles of ions can be expected to be ac-curate up to a sub-micron level. Predictions of microdosimetric quantities in ir-

1Excluding measurements with no or a very low minimum energy threshold εmin,for details see paper II.

39

3.5. ACCURACY OF FLUKA SIMULATIONS

radiation fields clinically relevant for ion beam therapy do not only require thecorrect simulation of energy depositions by ions in microscopic volumes. Theyalso need a correct description of the fluences of primary and secondary ions ata given position in the patient (see chapter 2). Paper III investigates thereforemicrodosimetric quantities of mixed radiation fields as created by therapeuticcarbon ion beams in phantoms using FLUKA. For the tested measurements,simulations of yF and yD were found to be respectively always within ±21%(RMS 11%) and ±11% (RMS 5%) of the measured values. Contributionsof charged fragments to the lineal energy spectra were found to be generallyslightly underestimated which is in accordance with findings in paper I. More-over, it was shown that the ‘V effect’ considerably modifies the lineal energyspectra of charged fragments measured by TEPCs in phantoms and should becorrected for when using such measurements for radiobiological models.

In conclusion, FLUKA was shown to simulate the energy imparted in mi-crodosimetric volumes adequately, including features of the measured datawhich are caused by energy loss straggling, the balance of delta electronswhich enter and exit the cavity and energy depositions by delta electrons inthe cavity due to ions passing outside of the TEPC cavity in the plastic wall ofthe detector.

40

4. Modelling of Comptonscattering and two-quantapositron annihilation in medium

4.1 Motivation

The very conformal treatment which can be delivered with ion beam therapymakes methods which allow to monitor independently the correct treatmentdelivery in situ highly desirable for routine clinical quality assurance, as dis-cussed in section 1.1.5. Current strategies aim at exploiting patient-exitingsecondary radiation to deduce particle ranges and dosimetric information invivo and include imaging of radioactivity distributions with PET and prompt-γimaging. MC techniques are employed for feasibility studies (Parodi et al.,2005; Polf et al., 2009b) and full ion beam treatment simulations (Parodi et al.,2008a,b) for PET and prompt-γ imaging. In addition to electromagnetic andnuclear physics processes which are relevant for dosimetric and radiobiologi-cal aspects in ion beam therapy, both monitoring techniques require an accuratemodelling of very specific physics processes. These include the production ofprompt-γ and β+-emitters in nuclear reactions, the description of the positronannihilation process, as well as the faithful transport of signal photons andparticles causing background signals. Also physiological aspects, such as thewash-out effect before β+-decay which is mainly due to perfusion (Parodiet al., 2007b), has to be taken into account for a full modelling.

Two original physics models which provide an improved description ofCompton scattering and the angular correlation (and Doppler broadening) oftwo-quanta positron annihilation in medium were developed, integrated inFLUKA and validated in the frame of this thesis project. Details can be foundin paper IV. The development of these models complements a larger effort inthe community to improve and adapt the MC codes FLUKA and GEANT4 forthe modelling of ion beam imaging techniques for treatment verification in theframe of ENVISION1. At the same time, the models describe basic physicsprocesses and are hence more generally applicable and potentially of benefit

1http://envision.web.cern.ch/ENVISION

41

4.2. MODEL APPLICATIONS

for a wider range of applications as discussed in the following section.

4.2 Model applications

Applications are of importance for models since they specify the desirableaccuracy of the developed models for different regions of their predictions.Hence, the possible application areas influence the choice of the modelling ap-proaches and their evaluation criteria. Compton scattering is a fundamentalinteraction process being relevant for a wide spectra of simulation applica-tions including notably: dosimetry, source shielding, photon detectors, medi-cal linacs, PET and dose monitoring with prompt photons. These applicationsgenerally benefit from a more accurate description of the Compton scatteringprocess. The advantage of the proposed model compared with less sophisti-cated modelling approaches describing Compton scattering, such as the Klein-Nishina formula and the form-factor approach, are apparent specifically forlow photon energies and heavy target elements, as pointed out in paper IV.The Doppler broadening due to the motion of shell electrons is visible fora wide range of energies and target elements. The correct prediction of theangular photon distribution is in particular important for directional measure-ment devices, such as Compton cameras (Richard et al., 2011, 2009; Seo et al.,2011).

While the Compton scattering process has a fairly large application area,angular correlation effects and Doppler broadening of two-quanta annihilationradiation are relevant mostly only for applications where the focus is set on thetwo-quanta annihilation process. Most prominently this is the case for PET.The two-quanta positron annihilation is the underlying physical process whichis made use of by PET to obtain information about the location of positronemitters. Besides its use for treatment monitoring in ion beam therapy, PET iswidely used in the field of molecular imaging to determine in vivo distributionsof biologically active molecules, such as proteins labelled with a positron emit-ter (Gambhir, 2002; Rudin and Weissleder, 2003). For PET image reconstruc-tion, the e+e− annihilation is assumed to occur on the line of response (LOR)between two detectors activated by the annihilation photons. Figure 4.1 illus-trates the annihilation position and a resulting LOR. Assuming an acollinearityof θ = 5 mrad (the FWHM of the angular correlation distribution of annihila-tion photon pairs for water is about 10 mrad) and a distance between the an-nihilation reaction and the photon detection of d = 50 cm, the misestimationof the position is ∆x = d sin(θ/2) = 1.25 mm. In comparison, today’s clinicalPET systems are able to reach resolutions of 3 mm and smaller (Kennedy et al.,2006; Schlegel, 2010). PET systems used for the monitoring of cancer therapytreatments with protons and ions can reconstruct particle ranges with a resolu-

42

CHAPTER 4. COMPTON SCATTERING AND E+E− ANNIHILATION

tion of about 3 mm (Jäkel et al., 2001a) and aim at even higher resolutions ofless than 1 mm. Consequently, the acollinearity of these photons in the labo-ratory frame is an intrinsic limiting factor for the spatial resolution achievablewith the PET technique which is considered by many PET image reconstruc-tion algorithms. This uncertainty adds to various other factors degrading theimage resolution of PET, such as the range the positron travels before anni-hilation, photon scattering, the spatial detector resolution, false coincidenceevents, wash-out effects of the positron-emitting isotopes, patient motion, etc.All these factors lead to a blurring of the image, and are generally taken intoaccount in PET image reconstruction algorithms (Shibuya et al., 2007).

MC particle transport codes are used to describe positron emission, positronannihilation, and other physical interactions relevant for PET (Jan et al., 2004;Sommerer et al., 2009). Hence, they rely on an accurate description of theunderlying physical processes. Predicting the correct angular distribution oftwo-photon annihilation radiation helps thus to simulate the spatial resolutionof PET more realistically.

θ

LOR

d

θ2

e+e−

∆x

Figure 4.1: Acollinearity of the annihilation photons is a factor, intrinsic to thetwo-quanta positron annihilation process in medium, which limits the spatial res-olution theoretically achievable with the PET technique. An e+e− pair annihilat-ing into two photons with an angle θ is depicted. If the annihilation photons aredetected at a distance d, the maximum misestimation of the annihilation positionwith respect to the LOR is ∆x.

43

4.2. MODEL APPLICATIONS

44

5. Prototype of a MC-basedtreatment planning tool forprotons and ions

5.1 Introduction

Certified TPSs which use fast analytical dose engines are employed at iontherapy facilities for the optimization of RBE-weighted dose distributions inthe patient in order to satisfy therapeutic dose prescriptions. Though com-putationally more demanding, MC computations of patient doses can achievemore accurate dose predictions in complex patient geometries (as discussed insection 1.2) and are increasingly used to support treatment planning and dosi-metric quality assurance for ion beam therapy. In this chapter, the prototype ofa newly-developed MC-based treatment planning (MCTP) tool for protons andions with Z ≤ 8 is presented. The versatile tool is aimed to be used for researchapplications and to support treatment planning studies at ion beam therapy fa-cilities. This thesis project has contributed to the tool by developing the codeand implementing the algorithms which are used for inverse treatment plan op-timization (see paper VI), by extending the tool to ions other than protons andby evaluating its performances. After a general introduction to the tool, thefollowing sections describe briefly the CT-conversion to MC geometry as wellas the dose-to-medium conversion scheme used. An evaluation of the MCTPtool for protons can be found in paper VI. Treatment planning for carbon ionsis described in section 5.3. Section 5.4 introduces a study performed with theMCTP tool which investigates the biological robustness of ion beam therapytreatment plans.

5.2 Workflow and components of the MCTP tool

The MCTP tool is composed of various program components which requireseveral treatment- and facility-specific input information as illustrated in fig-ure 5.1. The required input includes: (1) beam parameters for the generationof the initial pencil beam particle phase spaces for different lateral positions,

45

5.2. WORKFLOW AND COMPONENTS OF THE MCTP TOOL

Figure 5.1: Treatment- and facility-specific input for the MCTP tool.

beam energies and focus sizes, also referred to as a beam library, (2) the pa-tient planning CT or phantom geometry together with CT-specific parametersfor the conversion and calibration procedure to materials (described in sec-tion 5.2.1) and (3) regions of interest for the plan optimization and evaluation,i.e. PTVs and OARs, and the corresponding prescribed doses. For treatmentplanning performed using a variable RBE, as necessary for carbon ions, (4)radiobiological input tables derived from a radiobiological model (see sec-tion 1.1.1 and 3.4) are needed for the calculation of RBE-weighted dose DRBE.Starting from a pre-optimized set of pencil beams the MC code FLUKA is usedfor the computation of MC dose kernels. These dose kernels are then used byan iterative optimizer to obtain the ‘MC-optimized treatment plan’. A descrip-tion of the workflow and details of the MCTP tool can be found in paper VI,including in particular information about the pre-optimization, the FLUKAconfiguration and set-up, the generation of radiobiological input tables using aHIT re-implementation of the LEM and the optimization procedure and algo-rithms.

46

CHAPTER 5. MONTE CARLO TREATMENT PLANNING

5.2.1 Conversion of CT data for MC calculations and range calibra-tion

Unlike pencil beam algorithms, which generally use a water-equivalent ap-proach, MC codes compute particle transport and dose deposition naturallyin media and model the human body as being composed of materials withdiffering atomic compositions. A stoichiometric conversion scheme is usedto convert CT numbers into materials with a certain composition and massdensity. This is inevitably connected with approximations since there is noinjective correspondence between CT numbers and tissue compositions. TheCT segmentation procedure according to Schneider et al. (2000) and Parodiet al. (2007a) is used by the MCTP tool. It divides materials into 27 materialgroups for different HU intervals: 1 group for air, 1 group for lung tissue, 7groups for soft tissues, 15 groups for skeletal tissues and 3 groups for high-Zmaterials. Each group has a fixed material composition and a nominal massdensity with corresponding characteristics for physics processes. Instead, acontinuous mass density is assigned to each voxel based on a conversion fromthe corresponding HU (Schneider et al., 2000). Such a scheme has the advan-tage to reduce significantly memory usage and initialization time, compared toan approach using thousands of material definitions which would be requiredto cover accurately the whole HU range. Nuclear processes in a given voxelare assumed in the scheme to rescale with the ratio of the actual mass densitydivided by the nominal mass density of the material group. A HU-dependenttuning of the stopping power (and other electromagnetic processes) calculatedby the MC codes is used to assure a matching of the MC-calculated particleranges with ranges predicted by the TPS using WEPL/range tables (see sec-tion 1.1.4). This allows to evaluate differences in MC- and TPS-calculatedtreatment plans which are primarily due to approximations in the analyticaldose engine for heterogeneous patient geometries, opposed to differences re-sulting from different particle range calibrations1.

5.2.2 Conversion of dose-to-medium to dose-to-water

MC computations on the patient-CT grid use generally media with differingelemental compositions, as described in the previous section. Thus, MC codescompute a priori dose-to-medium (also dose-to-tissue). Since clinical expe-rience and dosimetry protocols, are based on dose-to-water as dose enginesused for treatment planning traditionally use a water-equivalent approach, it

1To be precise, it should be noted that the range uncertainties originating from theCT conversion scheme for MC are smaller than the ones for the CT conversion schemeused for analytical dose calculation according to evaluations from Paganetti (2012).

47

5.3. TREATMENT PLANNING FOR CARBON IONS

is also recommended to compute dose-to-water for MC implementations tomake the MC predictions directly comparable with TPS predictions (Chettyet al., 2007).1 The methodology described by Sommerer et al. (2012) is usedto compute dose-to-water for the MCTP tool. In essence, it uses the track-length fluence of all crossing particles multiplied with the corresponding massstopping powers for water. Dose conversion methods performed during MCparticle transport, such as the one mentioned here, are commonly referred tobe ‘on-the-fly’. This is in contrast to retroactive conversion methods which areapplied after MC dose calculations (Paganetti, 2009).

5.3 Treatment planning for carbon ions

Capabilities and performance of the MCTP tool for the computation of treat-ment plans in phantoms and patient CT geometries are presented in paper VIfor CNAO treatment conditions for proton therapy. In collaboration with HIT,the tool was adapted to be able to perform also treatment planning for carbonions with HIT treatment conditions for carbon ion therapy. This was achievedby merging the MCTP tool with parts of the FLUKA-based MC frameworkFICTION (FLUKA Integrated Framework for CT-based calculations in IonTherapy), developed at HIT (Sommerer et al., 2012) and using the correspond-ing visualisation tool (Unholtz et al., 2011). The FICTION tool was validatedand fine-tuned carefully to reproduce dosimetric measurements for HIT treat-ment conditions.

5.3.1 Examples of carbon ion treatment plans

Two test plans obtained using the MCTP tool are presented. The MCTP toolwas used to compute and optimize a treatment field for a cube-shaped PTV inwater. The pencil beams were optimized to yield a constant absorbed dose of1 Gy in the PTV. Figure 5.2 shows the contour plots of the absorbed dose.

As a further test case, a test patient plan for a carbon ion treatment of abrain tumour for HIT treatment conditions was created using the MCTP tool.A biological optimization was performed to obtain an RBE-weighted dose of3 GyE in the PTV using RBE values predicted by the LEM-I for chordomacell survival. Figures 5.3 and 5.4 show a comparison of the example treatmentplan created by the certified analytical TPS used at HIT (SIEMENS SyngoPT Planning) and a MC re-calculation of the same treatment plan using theFLUKA-based FICTION tool (Sommerer et al., 2012) together with a treat-ment plan computed with the MCTP tool and displayed with FICTION.

1A further argument is that cells, which are the main target of radiation therapy,contain a large fraction of water and can generally be well approximated by water.

48

CHAPTER 5. MONTE CARLO TREATMENT PLANNING

Figure 5.2: Carbon ion field computed and optimized by the MCTP tool for acube-shaped PTV in water. The pencil beams were optimized to yield a constantabsorbed dose of 1 Gy in the PTV.

0

20

40

60

80

100

0 0.5 1 1.5 2 2.5 3 3.5

Vo

lum

e [

%]

RBE-weighted dose [GyE]

TPS CTVMCRECALC CTV

MCTP CTV

0

20

40

60

80

100

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35

Vo

lum

e [

%]

RBE-weighted dose [GyE]

TPS BrainStemMCRECALC BrainStem

MCTP BrainStem

Figure 5.3: Test treatment plan of a brain tumour treated with carbon ions. DVHsof the CTV and the brain stem of a treatment plan created with an analyticaltreatment planning tool, the MC re-calculation of the same plan and a treatmentplan created with the MCTP tool.

49

5.3. TREATMENT PLANNING FOR CARBON IONS

a) TPS

b) MC re-calculation

c) MCTP

Figure 5.4: Test treatment plan of a brain tumour treated with carbon ions.Sagittal (left), coronal (middle) and axial (right) view are shown for (a) a treat-ment plan created with a TPS using an analytical dose engine, (b) the MC re-calculation of the same plan and (c) a treatment plan created with the MCTPtool. RBE-weighted doses are shown in colourwash-scale. HUs of the planningCT are shown in grey-scale. Contours show the CTV and the brain stem.

50

CHAPTER 5. MONTE CARLO TREATMENT PLANNING

For the used test patient plan only minor differences between the treatmentplan created using the analytical TPS, the MC re-calculation, and the one cre-ated with the MCTP tool are observable. The MC re-calculation of the plancreated by the analytical TPS predicts a slight over-dosage in the CTV. Thisover-dosage is corrected using the MCTP tool. This test patient plan servesas a show case. Larger discrepancies between plans obtained using a MC re-calculation of an analytical TPS and MCTP can be expected in the presence oftissue regions in the beam with large heterogeneities and patient cases involv-ing metallic implants (Paganetti et al., 2008; Parodi et al., 2007b).

5.4 Investigating the biological robustness of ion beamtherapy treatment plans

RBE values clinically used for the treatment of patients with carbon ions aresubject to uncertainties originating from several sources (see paper V for moredetails). In terms of their clinical importance one can discriminate betweenabsolute misestimations and relative misestimations. Absolute misestimationsof RBE lead to a systematic but constant over- or under-dosage of tumoursand healthy tissues for a given treatment protocol. Such misestimations canbe detected and corrected for by clinical dose escalation trials. Dose esca-lation trials are common practice in carbon ion therapy and have the aim tofind an adequate compromise between tumour control probability and normaltissue complication probability (Linz et al., 2012, Ch. 30, p. 517). Instead,systematic relative misestimations of RBE and survival are more difficult todetect clinically and lead generally to a worse treatment outcome and a flat-tened dose-response curve. Figure 5.5 illustrates schematically different typesof misestimations of RBE-weighted doses in the tumour volume for a singlecarbon ion field. Possible relative misestimations include (but are not limitedto) unwanted RBE-weighted dose gradients in the tumour volume, over- orunder-dosage as a function of size of the SOBP and over- or under-dosage asa function of depth of the SOBP.

In paper V the sensitivity of carbon ion treatment fields to uncertainties inbiological treatment parameters is investigated. The study evaluates the riskfor different types of RBE misestimations occurring according to LEM predic-tions (see section 1.1.1) for single carbon ion fields and differently optimizedopposed carbon ion fields. The study is performed using the MCTP tool de-scribed in this chapter.

51

5.4. BIOLOGICAL ROBUSTNESS OF TREATMENT PLANS

Figure 5.5: Schematical illustration of absolute and relative false predictions ofRBE-weighted doses in the tumour volume for a single carbon ion field.

52

6. Conclusions and perspectives

Dose and fluence computation tools using the MC technique serve since manyyears as reference tools for accurate dose computations for radiotherapy. Ionbeam therapy, as an emerging radiation therapy modality, requires a continu-ous effort to develop and improve MC particle transport codes for applicationsin dosimetry and imaging in ion beam therapy. In the framework of this the-sis project, physics models of the MC codes FLUKA and GEANT4 have beenvalidated and developed for applications in ion beam therapy. Furthermore, theprototype of a novel MCTP tool for ion beam therapy was developed, success-fully tested using realistic treatment conditions and applied in a research studyinvestigating the robustness of ion therapy treatment plans to uncertainties inbiological treatment parameters. Major conclusions of the present thesis are:

F FLUKA and GEANT4 are able to predict integral thick target yieldsand cross sections of secondary charged fragments for carbon ions withtherapeutic energies generally within 10-30% of experimental values.Yields differential in angle predicted by FLUKA are in good agreementwith experimental data. Large angles (≈ 10◦) tend to be slightly under-estimated. For nuclear models of GEANT4, angular charged fragmentyields tend be overestimated for larger angles (≈ 10◦) and underesti-mated for forward angles (≈ 0◦) in some cases. The energy distribu-tions at angles contributing most to the integral yields are generally wellpredicted by both codes. Observed differences in the production of sec-ondary charged fragments are expected to be relevant for imaging ap-plications using these fragments and for studies examining cancer riskdue to out-of-field doses. Their impact on the RBE in the high dose re-gion of carbon ion treatment fields is estimated to be on the level of afew percent only. Existing experimental data are scarce and often lim-ited in accuracy. This impedes a more detailed benchmarking of nuclearmodels implemented in MC codes. New measurements of nuclear inter-actions will allow for more sensitive benchmarks in near future. For in-stance, detailed measurements of differential cross sections of 12C ionsat 400 MeV per nucleon on a carbon target were recently acquired bythe FIRST experiment. Once evaluated, these measurements can help toimprove the accuracy of nuclear models for ion beam therapy.

53

F FLUKA is able to simulate microdosimetric quantities, such as linealenergy, for radiation fields generated by ion beams. It is capable of re-producing the momenta of measured lineal energy spectra yF and yDwithin ±10% for single ions with charges from 1 to 26 and energiesfrom 80 up to 1000 MeV per nucleon for all tested cases. The respectiveRMS deviation is 4.7% for yF and 4.4% for yD. Lineal energy in mixedradiation fields produced by therapeutic carbon ion beams in phantomscan be also predicted by FLUKA. The agreement with measured valueswas found to be within ±21% (RMS 11%) and ±11% (RMS 5%) foryF and yD, respectively. The ‘V effect’ (see section 3.3 for definition)was found to be of importance for microdosimetric measurements ofion beams in phantoms with walled measurement devices and should betaken into account and corrected for when interpreting the data or usingit for the estimation of biological effects. The obtained results allow toenvisage the use of FLUKA for the generation of microdosimetric quan-tities as input for radiobiological models which describe the biologicaleffectiveness of ion beams based on microdosimetric quantities, suchas the MKM. Moreover, it was shown that the predictive capabilities ofFLUKA physics models make FLUKA a reliable tool for design andoptimization studies of novel microdosimetric measurement devices.

F Two newly-developed physics models allow for the detailed simulationof Compton scattering and of angular correlation (and Doppler broad-ening) of two-quanta positron annihilation in medium. Both modellingapproaches take into account the shell electron motion and binding ef-fects. Thereby, they allow for an accurate simulation of physics pro-cesses which are relevant for PET and prompt-γ imaging for ion beamtherapy treatment monitoring in vivo. When compared with measure-ments the models show a satisfactory agreement. Both models were de-veloped as native FLUKA models and are included in FLUKA releases.

F This thesis project contributed to the development of the prototype ofa novel MCTP tool, based on FLUKA as dose and fluence engine. Itallows to perform single- and simultaneous multiple-field optimizationfor realistic patient treatment conditions and for dosimetric quality as-surance for irradiation conditions at CNAO and HIT. It uses iterativeoptimization procedures which allow for the optimization of absorbeddose and RBE-weighted dose using radiobiological input tables gener-ated by radiobiological models. Using a re-implementation of the LEM,the MCTP tool can be employed to perform treatment planning studiesusing ions with atomic numbers Z ≤ 8. The tool is aimed to be usedin future for research applications (e.g. next point) and to support treat-

54

CHAPTER 6. CONCLUSIONS AND PERSPECTIVES

ment planning at ion beam facilities. It can be used to re-calculate agiven treatment plan computed by certified TPSs which use analyticaldose engines and provide a ‘second opinion’ based on MC. PerformingMC treatment planning, in a next step, a new MC-optimized plan can beobtained. Future developments of the MCTP tool could consist in fur-ther optimizing its computation speed, simplifying its workflow as wellas integrating it in a more automated and flexible environment.

F The impact of uncertainties in input parameters to the biophysical mod-elling of RBE values for treatment plans was investigated by a sensi-tivity analysis. The study was performed using the newly-developedMCTP tool. Moderate variations (up to ±50%) of model input param-eters, such as α and β , were found to result mostly in absolute misesti-mations of RBE-weighted doses in the PTV for single fields. However,larger variations of model parameters or variations of RBE predictionsby different biophysical modelling approaches may also lead to signifi-cant RBE-weighted dose gradients in the PTV. By optimizing for a moreuniform radiation quality in the PTV, RBE-weighted dose gradients canbe substantially reduced compared to other optimization strategies. Forthis work, optimization for a more homogeneous radiation quality in thePTV was performed using RBE as a measure for radiation quality. Forthe investigated scenarios, this measure worked reliably. Approachesusing physical quantities, such as LET or lineal energy, could be used asa measure of radiation quality for treatment plan optimization in futureand might be preferable since they are independent of any radiobiologi-cal modelling. Improvements of biological robustness by optimizing fora constant RBE in the PTV occur at the expense of an increased entrancedose compared to optimization strategies which are aimed to minimizethe integral dose to the patient. Treatment fields using SFUD optimiza-tion represent an intermediate solution in terms of integral dose and bi-ological robustness. In general, the benefits of sparing normal tissuesand critical structures should be balanced against physical and biologi-cal robustness for ion therapy treatment plans. A detailed considerationof all these aspects is desirable for treatment planning. Several strategieshave been proposed in the literature to optimize IMPT treatment planssystematically with respect to uncertainties in physical parameters. Anextension to take systematically into account also biological uncertain-ties in ion therapy treatment plans could help to improve the outcome ofion therapy treatments.

55

56

Acknowledgements

My sincere gratitude goes to ...

Manjit Dosanjh, who reached out to me, unexpectedly, and invited meto apply for this Ph.D. project in the first place, for all the time, guid-ance and great support she gave to me and many other of us ‘scientificyoungsters’.

Alfredo Ferrari, for his willingness to take me as a ‘foster Ph.D.’ in hisgroup, for his guidance, support and interest in my Ph.D. activities, formaking all the physics look so easy and clear - which they are not, andfor his enthusiasm speaking about ‘non-politics’-related business.

Irena Gudowska, for her meticulous and accurate corrections and forintroducing me to and translating to me the world of Medical RadiationPhysics in Stockholm.

Andres Brahme, for the amount of interest he took in my Ph.D. studies.

Vincenzo Patera, for giving me the opportunity to work for the FIRSTexperiment and the MCTP project, for his incredible capacity of paintinga smile on other people’s faces and always looking on the bright side oflife – even though thunderstorms might be all around you.

Andrea Mairani, my ‘direct wire to the clinics’ and unofficial supervisor,for receiving endless emails with ‘some other questions’ (and actuallyanswering them), for his easy and uncomplicated way during and afterworking hours.

Alessio Sarti, for his help and endurance with decoding of existing andcoding of the FIRST reconstruction code, for happily being always therefor all other kinds of nasty problems. QAD is a lifestyle!

Katia Parodi and Julia Bauer, for all their support and help with theextension of the MCTP tool to HIT conditions and for supporting themerging and use of parts of the FICTION code and tools. ... and yes, I’malso willing to pay in chocolate bars from treacherous CERN vendingmachines instead of coffees!

Martha, Ida, Jörgen and Björn, for their kind help with the Sammanfat-tning på svenska.

The many office mates, colleagues and co-Ph.D. students at CERN andKI which accompanied me during my four Ph.D. years . . . while chang-ing at least five times office. In particular, Björn, Thiansin, Martha andMarta, Katarzyna, Daniel, Vasso, Mary, Pablo, Silvia, Alberto, Juan,Adriano and the EN/STI group.

The PARTNER researchers, for a fantastic atmosphere and many funnyevenings out. You are a great bunch!

My friends here in Switzerland, in particular Bébé, Fra e Fra, Matthias,Pierre and Harald.

My dear parents, for all their interest, care and support over the last yearsand during my studies. I wouldn’t want to exchange you!

The rest of my beloved and wonderful patchwork family!

Ilaria, for being exactly as she is and for making me so happy whilebeing it. Thank you for your love, patience . . . and understanding whensomebody had his mind, again, in a totally different place.

This research project was supported by a Marie Curie Initial Training NetworkFellowship of the European Community’s Seventh Framework Programme un-der contract number PITNGA-2008-215840-PARTNER and by ENVISION,which is co-funded by the European Commission under FP7 Grant AgreementN. 241851.

II

References

Abou-Haidar, Z., et al., 2012 see paper III∗. 30

Agodi, C., et al., 2008. The INFN TPS project. Il Nuovo Cimento 31 (1), 99–108. 18, 19

Agosteo, S., Pola, A., 2011. Silicon microdosimetry. Radiat. Prot. Dosim. 143 (2-4), 409–415. 35, 36, 39

Agostinelli, S., Allison, J., Amako, K., Apostolakis, J., Araujo, H., 2003. Geant4 – A simulation toolkit.Nucl. Instrum. Meth. A 506, 250–303. 17

Aichelin, J., 1991. “Quantum” molecular dynamics–a dynamical microscopic n-body approach to investi-gate fragment formation and the nuclear equation of state in heavy ion collisions. Phys. Rep. 202 (5-6),233–360. 23

Allison, J., et al., 2006. Geant4 developments and applications. IEEE T. Nucl. Sci. 53 (1), 270–278. 17

Andersen, V., et al., 2004. The FLUKA code for space applications: Recent developments. Adv. Space Res.34, 1302–1310. 23, 24, 27, 28

Ballarini, F., Battistoni, G., Cerutti, F., A. Empl, A. F., Ferrari, A., Gadioli, E., Garzelli, M., Ottolenghi, A.,Pinsky, L. S., Ranft, J., Roesler, S., Sala, P. R., Smirnov, G., 2004. Nuclear models in FLUKA: Presentcapabilities, open problems and future improvements. presented at ND2004, Santa Fe, 26 Sep - 1 Oct(2004), SLAC-PUB-10813. 22, 24

Baron, M. H., Pommier, P., Favrel, V., Truc, G., Balosso, J., Rochat, J., 2004. A “one-day survey”: As areliable estimation of the potential recruitment for proton- and carbon-ion therapy in France. Radiother.Oncol. 73 (Suppl. 2), S15–S17. 6

Battistoni, G., Cerutti, F., Engel, R., Fassò, A., Ferrari, A., Gadioli, E., Garzelli, M. V., Ranft, J., Roesler,S., Sala, P. R., 2006. Recent developments in the FLUKA nuclear reaction models. Proceed. 11th Int.Conf. on Nucl. React. Mech., Varenna, Italy, 12-16th June, 483–495. 25

Battistoni, G., Muraro, S., Sala, P. R., Cerutti, F., Ferrari, A., Roesler, S., Fassò, A., Ranft, J., 2007. TheFLUKA code: Description and benchmarking. Proceed. of the Hadronic Shower Simulation Workshop2006, AIP Conference Proceed. 896 (2007) 31–49. 17, 25, 26, 28

Bert, C., Durante, M., 2011. Motion in radiotherapy: Particle therapy. Phys. Med. Biol. 56 (16), R113. 7

Bert, C., Engenhart-Cabillic, R., Durante, M., 2012. Particle therapy for noncancer diseases. Med. Phys.39 (4), 1716–1727. 5

Bertsch, G., Gupta, S. D., 1988. A guide to microscopic models for intermediate energy heavy ion colli-sions. Phys. Rep. 160 (4), 189–233. 23

Bhat, M. R., 1987. Procedures manual for the evaluated nuclear structure data file. Tech. Rep. BNL-NCS-40503 INFORMAL REPORT (Brookhaven: The National Nuclear Data Center Brookhaven NationalLaboratory). 27

REFERENCES

Bohr, N., Wheeler, J. A., Sep 1939. The mechanism of nuclear fission. Phys. Rev. 56, 426–450. 26

Brenner, D. J., Hall, E. J., 2008. Secondary neutrons in clinical proton radiotherapy: A charged issue.Radiother. Oncol. 86 (2), 165–170. 7

Cerutti, F., Battistoni, G., Capezzali, G., Colleoni, P., Ferrari, A., Gadioli, E., Mairani, A., Pepe, A., 2006.Low energy nucleus-nucleus reactions: the BME approach and its interface with FLUKA. Proceed.11th Int. Conf. on Nucl. Reac. Mechan., Varenna, Italy, 12-16th June, 507–511, book of proceedings,available at http://www.mi.infn.it/∼gadioli/Varenna2006/Proceedings/Cerutti_F.pdf. 23, 28

Cesari, V., Colautti, P., Magrin, G., De Nardo, L., Y. Baek, W., Grosswendt, B., Alkaa, A., Khamphan,C., Ségur, P., Tornielli, G., 2002. Nanodosimetric measurements with an avalanche confinement TEPC.Radiation Protection Dosimetry 99 (1-4), 337–342. 35

Chauvie, S., Francis, Z., Guatelli, S., Incerti, S., Mascialino, B., Montarou, G., Moretto, P., Nieminen, P.,Pia, M. G., 2006. Monte Carlo simulation of interactions of radiation with biological systems at thecellular and DNA levels: The Geant4-DNA Project. Radiat. Res. 166 (4), 652–689. 33

Chen, G. T., Singh, R. P., Castro, J. R., Lyman, J. T., Quivey, J. M., 1979. Treatment planning for heavy ionradiotherapy. Int. J. Radiat. Oncol. 5 (10), 1809–1819. 11

Chetty, I. J., Curran, B., Cygler, J. E., DeMarco, J. J., Ezzell, G., Faddegon, B. A., Kawrakow, I., Keall,P. J., Liu, H., Ma, C.-M. C., Rogers, D. W. O., Seuntjens, J., Sheikh-Bagheri, D., Siebers, J. V., 2007.Report of the AAPM Task Group No. 105: Issues associated with clinical implementation of MonteCarlo-based photon and electron external beam treatment planning. Med. Phys. 34 (12), 4818–4853. 15,48

Combs, S. E., Jäkel, O., Haberer, T., Debus, J., 2010. Particle therapy at the Heidelberg Ion Therapy Center(HIT) – Integrated research-driven university-hospital-based radiation oncology service in Heidelberg,Germany. Radiother. Oncol. 95 (1), 41–44. 5

Cucinotta, F. A., Nikjoo, H., Goodhead, D. T., 1999. Applications of amorphous track models in radiationbiology. Radiat. Environ. Biophys. 38, 81–92. 34

Dauvergne, D., Battaglia, M., Montarou, G., Testa, E., 2009. New methods of real-time control imaging forion therapy. Hyper Articles en Ligne available at: http://hal.archives-ouvertes.fr/in2p3-00363382/. 14

Dementyev, A. V., Sobolevsky, N. M., 1999. SHIELD – Universal Monte Carlo hadron transport code:Scope and applications. Radiat. Meas. 30, 553–557. 17

Elsässer, T., Weyrather, W. K., Friedrich, T., Durante, M., Iancu, G., Krämer, M., Kragl, G., Brons, S.,Winter, M., Weber, K.-J., Scholz, M., 2010. Quantification of the relative biological effectiveness forion beam radiotherapy: Direct experimental comparison of proton and carbon ion beams and a novelapproach for treatment planning. Int. J. Radiat. Oncol. 78 (4), 1177–1183. 4, 34

Endo, M., Koyama-ito, H., Minohara, S., Miyahara, N., Tomura, H., Kanai, T., Kawachi, K., Tsuji, H.,Morita, K., 1996. HIPLAN–a heavy ion treatment planning system at HIMAC. J. Jpn. Soc. Ther. Radiol.Oncol. 8, 231–238. 11

Endo, S., Takada, M., Onizuka, Y., Tanaka, K., Maeda, N., Ishikawa, M., Miyahara, N., Hayabuchi, N.,Shizuma, K., Hoshi, M., 2007. Microdosimetric evaluation of secondary particles in a phantom producedby carbon 290 MeV/nucleon ions at HIMAC. Radiat. Res. 48 (5), 397–406. 39

Endo, S., Takada, M., Tanaka, H., Onizuka, Y., Tanaka, K., Miyahara, N., Baba, H., Oishi, A., Ishikawa, M.,Hoshi, M., Kimura, S., Minematsu, M., Morimune, Y., Kojima, Y., Shizuma, K., 2010. Measurementof microdosimetric spectra produced from a 290 MeV/n spread out Bragg peak carbon beam. Radiat.Environ. Biophys. 49, 469–475. 39

IV

REFERENCES

Enghardt, W., Crespo, P., Fiedler, F., Hinz, R., Parodi, K., Pawelke, J., Pönisch, F., 2004a. Charged hadrontumour therapy monitoring by means of PET. Nucl. Instrum. Meth. A 525 (1–2), 284–288, Proceedingsof the International Conference on Imaging Techniques in Subatomic Physics, Astrophysics, Medicine,Biology and Industry. 14

Enghardt, W., Parodi, K., Crespo, P., Fiedler, F., Pawelke, J., Pönisch, F., 2004b. Dose quantification fromin-beam positron emission tomography. Radiother. Oncol. 73, Supplement 2, S96–S98. 13, 14

Epherre, M., Gradsztajn, E., Klapisch, R., Reeves, H., 1969. Comparison between the evaporation and thebreak-up models of nuclear de-excitation. Nucl. Phys. A 139 (3), 545–553. 26

Fassò, A., Ferrari, A., Ranft, J., Sala, P. R., 1995a. FLUKA: Performances and applications in the interme-diate energy range. Proceed. 1st AEN/NEA Specialists’ Meeting on Shielding Aspects of Accelerators,Targets and Irradiation Facilities (SATIF 1), Arlington (Texas), United States, Apr. 28-29 (1994), OECDDocuments (1995) 287–304. 25, 26

Fassò, A., Ferrari, A., Ranft, J., Sala, P. R., 1995b. An update about FLUKA. Proceed. 2nd Workshopon Simulating Accelerator Radiation Environments (SARE 2), CERN, Geneva, Switzerland, 9-11 Oct,CERN Report TIS-RP/97-05 (1997), 158–170. 25

Fermi, E., 1950. High energy nuclear events. Progress of Theoretical Physics 5 (4), 570–583. 26

Ferrari, A., Sala, P. R., 1994. A new model for hadronic interactions at intermediate energies for the FLUKAcode. Proceed. Int. Conference on Monte Carlo Simulation in High Energy and Nuclear Physics 1993,277–288. 25

Ferrari, A., Sala, P. R., 1998. The physics of high energy reactions. Proceed. Workshop on Nuclear ReactionData and Nuclear Reactors Physics, Design and Safety, World Scientific, p. 424 , Miramare-Trieste,Italy, 15 April-17 May 1996, available at http://www.fluka.org/content/publications/1996_trieste.pdf.22, 25, 27

Ferrari, A., Sala, P. R., 2002. Nuclear reactions in Monte Carlo codes. Radiat. Prot. Dosim. 99 (1-4), 29–38.22, 23, 26, 27

Ferrari, A., Sala, P. R., Fassò, A., Ranft, J., 2005. FLUKA: A multi-particle transport code. Tech. Rep.CERN-2005-10, INFN/TC_05/11, SLAC-R-773. 17, 22

Fippel, M., Soukup, M., 2004. A Monte Carlo dose calculation algorithm for proton therapy. Med. Phys.31 (8), 2263–2273. 16

FLUKA Collaboration, 2012. FLUKA Manual. Available at http://www.fluka.org/. 28

Furusawa, Y., Fukutsu, K., Aoki, M., Itsukaichi, H., Eguchi-Kasai, K., Ohara, H., Yatagai, F., Kanai, T.,Ando, K., 2000. Inactivation of aerobic and hypoxic cells from three different cell lines by accelerated3He-, 12C- and 20Ne-ion beams. Radiat. Res. 154 (5), 485–496. 34

Gambhir, S. S., 2002. Molecular imaging of cancer with positron emission tomography. Nat. Rev. Cancer2, 683–693. 42

Geant4, 2012. Geant4 Physics Reference Manual. Available at:http://geant4.web.cern.ch/geant4/support/userdocuments.shtml. 22

Gerlach, R., Roos, H., M. Kellerer, A., 2002. Heavy ion RBE and microdosimetric spectra. Radiat. Prot.Dosim. 99 (1-4), 413–418. 37

Golosio, B., et al., 2011 see paper I∗. 30

Goodwin, E. H., Bailey, S. M., Chen, D. J., Cornforth, M. N., 1996. The effect of track structure on cellinactivation and chromosome damage at a constant LET of 120 keV/µm. Adv. Space Res. 18 (1-2),93–98. 34

V

REFERENCES

Guetersloh, S. B., Borak, T. B., Taddei, P. J., Zeitlin, C., Heilbronn, L., Miller, J., Murakami, T., Iwata, Y.,2004. The response of a spherical tissue-equivalent proportional counter to different ions having similarlinear energy transfer. Radiat. Res. 161, 64–71. 34

Haberer, T., Becher, W., Schardt, D., Kraft, G., 1993. Magnetic scanning system for heavy ion therapy.Nucl. Instrum. Meth. A 330 (1-2), 296–305. 7

Hall, E. J., 2006. Intensity-modulated radiation therapy, protons, and the risk of second cancers. Int. J.Radiat. Oncol. 65 (1), 1–7. 5, 7

Hansen, D. C., Lühr, A., Sobolevsky, N., Bassler, N., 2012. Optimizing SHIELD-HIT for carbon ion treat-ment. Phys. Med. Biol. 57 (8), 2393. 30

Hawkins, R. B., 1994. A statistical theory of cell killing by radiation of varying linear energy transfer.Radiat. Res. 140, 366–347. 4, 37

Hawkins, R. B., 2004. A microdosimetric-kinetic model for the effect of non-Poisson distribution of lethallesions on the variation of RBE with LET. Radiat. Res. 160, 61–69. 4, 34, 37

Henriquet, P., Testa, E., Chevallier, M., Dauvergne, D., Dedes, G., Freud, N., Krimmer, J., Létang, J. M.,Ray, C., Richard, M.-H., Sauli, F., 2012. Interaction vertex imaging (IVI) for carbon ion therapy moni-toring: A feasibility study. Phys. Med. Biol. 57 (14), 4655. 14

Hill, J. C., Wohn, F. K., Winger, J. A., Smith, A. R., Mar 1988. Electromagnetic dissociation for high-Zprojectiles and at ultrarelativistic energies. Phys. Rev. Lett. 60, 999–1001. 24

Hotta, K., Kohno, R., Takada, Y., Hara, Y., Tansho, R., Himukai, T., Kameoka, S., Matsuura, T., Nishio,T., Ogino, T., 2010. Improved dose-calculation accuracy in proton treatment planning using a simplifiedMonte Carlo method verified with three-dimensional measurements in an anthropomorphic phantom.Phys. Med. Biol. 55 (12), 3545. 16

Hughes, G., Adams, K. J., Chadwick, M. B., Comly, J. C., Frankle, S. C., Hendricks, J. S., Little,R. C., Prael, R. E., Waters, L. S., Young Jr, P. G., 1997. MCNPXTM – The LAHETTM/MCNPTMcode merger. Proceed. of 3rd Workshop on Simulating Accelerator Radiation Environments (SARE 3),Tsukuba, Japan, 7–9 May, KEK Proceed. (June 1997) H/R/D 44–51. 17

Hultqvist, M., Lillhök, J.-E., Lindborg, L., Gudowska, I., Nikjoo, H., 2010. Nanodosimetry in a 12C ionbeam using Monte Carlo simulations. Radiat. Meas. 45 (10), 1238–1241. 3, 18, 37

IAEA, 2008. Relative Biological Effectiveness in ion beam therapy. Tech. Rep. IAEA Technical ReportSeries No 461 (Vienna: International Atomic Energy Agency). 1, 3, 8, 38

ICRP, 2007. The 2007 recommendations of the International Commission on Radiological Protection. Tech.Rep. ICRP Publication No 103 (Oxford: Pergamon). 34

ICRU 16, 1970. Linear Energy Transfer. Tech. Rep. ICRU Report No 16 (Bethesda, MD: ICRU). 34

ICRU 36, 1983. Microdosimetry. Tech. Rep. ICRU Report No 36 (Bethesda, MD: ICRU). 34, 35, 36, 38

ICRU 78, 2007. Prescribing, recording, and reporting proton-beam therapy. Tech. Rep. ICRU Report No 78(Bethesda, MD: ICRU). 3

Inaniwa, T., Kanematsu, N., Furukawa, T., Noda, K., 2011. Optimization algorithm for overlapping-fieldplans of scanned ion beam therapy with reduced sensitivity to range and setup uncertainties. Phys. Med.Biol. 56 (6), 1653. 12

Jahnke, L., Fleckenstein, J., Wenz, F., Hesser, J., 2012. GMC: a GPU implementation of a Monte Carlodose calculation based on Geant4. Phys. Med. Biol. 57 (5), 1217. 18

VI

REFERENCES

Jäkel, O., 2006. Ranges of ions in metals for use in particle treatment planning. Phys. Med. Biol. 51 (9),N173. 10

Jäkel, O., Jacob, C., Schardt, D., Karger, C. P., Hartmann, G. H., 2001b. Relation between carbon ion rangesand x-ray CT numbers. Med. Phys. 28 (4), 701–703. 9, 19

Jäkel, O., Krämer, M., Karger, C. P., Debus, J., 2001a. Treatment planning for heavy ion radiotherapy:Clinical implementation and application. Phys. Med. Biol. 46 (4), 1101. 9, 43

Jan, S., Santin, G., Strul, D., Staelens, S., Assié, K., Autret, D., Avner, S., Barbier, R., Bardiès, M., Bloom-field, P. M., Brasse, D., Breton, V., Bruyndonckx, P., Buvat, I., Chatziioannou, A. F., Choi, Y., Chung,Y. H., Comtat, C., Donnarieix, D., Ferrer, L., Glick, S. J., Groiselle, C. J., Guez, D., Honore, P.-F.,Kerhoas-Cavata, S., Kirov, A. S., Kohli, V., Koole, M., Krieguer, M., van der Laan, D. J., Lamare, F.,Largeron, G., Lartizien, C., Lazaro, D., Maas, M. C., Maigne, L., Mayet, F., Melot, F., Merheb, C.,Pennacchio, E., Perez, J., Pietrzyk, U., Rannou, F. R., Rey, M., Schaart, D. R., Schmidtlein, C. R., Si-mon, L., Song, T. Y., Vieira, J.-M., Visvikis, D., de Walle, R. V., Wieërs, E., Morel, C., 2004. GATE: asimulation toolkit for PET and SPECT. Phys. Med. Biol. 49 (19), 4543. 43

Jarlskog, C., Paganetti, H., 2008a. Physics settings for using the Geant4 toolkit in proton therapy. IEEE T.Nucl. Sci. 55 (3), 1018–1025. 19

Jarlskog, C. Z., Paganetti, H., 2008b. Risk of developing second cancer from neutron dose in proton therapyas function of field characteristics, organ, and patient age. Int. J. Radiat. Oncol. 72 (1), 228–235. 3, 18

Jemal, A., Siegel, R., Ward, E., Hao, Y., Xu, J., Murray, T., Thun, M. J., 2008. Cancer statistics, 2008.CA-Cancer J. Clin. 58 (2), 71–96. 1

Kase, Y., Kanai, T., Matsufuji, N., Furusawa, Y., Elsässer, T., Scholz, M., 2008. Biophysical calculation ofcell survival probabilities using amorphous track structure models for heavy-ion irradiation. Phys. Med.Biol. 53 (1), 37. 38

Kase, Y., Kanai, T., Matsumoto, Y., Furusawa, Y., Okamoto, H., Asaba, T., Sakama, M., Shinoda, H., 2006.Microdosimetric measurements and estimation of human cell survival for heavy-ion beams. Radiat. Res.166 (4), 629–638. 38

Kellerer, A. M., Chmelevsky, D., 1975. Criteria for the applicability of LET. Radiat. Res. 63, 226–234. 34

Kellerer, A. M., Rossi, H. H., 1984. On the determination of microdosimetric parameters in time-varyingradiation fields: The variance-covariance method. Radiat. Res. 97 (2), 237–245. 35

Kempe, J., Gudowska, I., Brahme, A., 2007. Depth absorbed dose and LET distributions of therapeutic 1H,4He, 7Li, and 12C beams. Med. Phys. 34, 183–192. 21

Kennedy, J. A., Israel, O., Frenkel, A., Bar-Shalom, R., Azhari, H., 2006. Super-resolution in PET imaging.IEEE T. Med. Imaging 25 (2), 137–147. 42

Krämer, M., Jäkel, O., Haberer, T., Kraft, G., Schardt, D., Weber, U., 2000. Treatment planning for heavy-ion radiotherapy: Physical beam model and dose optimization. Phys. Med. Biol. 45 (11), 3299. 11,12

Krämer, M., Scholz, M., 2006. Rapid calculation of biological effects in ion radiotherapy. Phys. Med. Biol.51, 1959–1970. 12

Levy, R. P., 2007. PET-CT: Evolving role in hadron therapy. Nucl. Inst. Meth. B 261 (1-2), 782–785. 8

Li, J. S., Shahine, B., Fourkal, E., Ma, C.-M., 2005. A particle track-repeating algorithm for proton beamdose calculation. Phys. Med. Biol. 50 (5), 1001. 16

Lillhök, J. E., Grindborg, J.-E., Lindborg, L., Gudowska, I., Carlsson, G. A., Söderberg, J., Kopec, M.,Medin, J., 2007. Nanodosimetry in a clinical neutron therapy beam using the variance-covariancemethod and Monte Carlo simulations. Phys. Med. Biol. 52 (16), 4953. 35, 37

VII

REFERENCES

Lindborg, L., Grindborg, J., 1997. Nanodosimetric results and radiotherapy beams: A clinical application?Radiat. Prot. Dosim. 70 (1-4), 541–546. 37

Linz, U., et al., 2012. Ion Beam Therapy–Fundamentals, Technology, Clinical Applications. Springer VerlagBerlin Heidelberg. 4, 5, 7, 9, 10, 11, 13, 17, 19, 38, 51

Lomax, A., 1999. Intensity modulation methods for proton radiotherapy. Phys. Med. Biol. 44 (1), 185. 13

Loncol, T., Cosgrove, V., Denis, J., Gueulette, J., Mazal, A., Menzel, H., Pihet, P., Sabattier, R., 1994.Radiobiological effectiveness of radiation beams with broad LET spectra: Microdosimetric analysisusing biological weighting functions. Radiat. Prot. Dosim. 52 (1-4), 347–352. 37

Lühr, A., Hansen, D. C., Teiwes, R., Sobolevsky, N., Jäkel, O., Bassler, N., 2012. The impact of modelingnuclear fragmentation on delivered dose and radiobiology in ion therapy. Phys. Med. Biol. 57 (16), 5169.21, 30

Mairani, A., Brons, S., Cerutti, F., Fassò, A., Ferrari, A., Krämer, M., Parodi, K., Scholz, M., Sommerer,F., 2010. The FLUKA Monte Carlo code coupled with the local effect model for biological calculationsin carbon ion therapy. Phys. Med. Biol. 55 (15), 4273–4289. 29

Martino, G., Durante, M., Schardt, D., 2010. Microdosimetry measurements characterizing the radiationfields of 300 MeV/u 12C and 185 MeV/u 7Li pencil beams stopping in water. Phys. Med. Biol. 55 (12),3441. 39

Matsufuji, N., Fukumura, A., Komori, M., Kanai, T., Kohno, T., 2003. Influence of fragment reaction ofrelativistic heavy charged particles on heavy-ion radiotherapy. Phys. Med. Biol. 48, 1605–1623. 21, 39

Mayer, R., Mock, U., Jäger, R., Pötter, R., Vutuc, C., Eiter, H., Krugmann, K., Hammer, J., Hirn, B.,Hawliczek, R., Knocke-Abulesz, T.-H., Lukas, P., Nechville, E., Pakisch, B., Papauschek, M., Raunik,W., Rhomberg, W., Sabitzer, H., Schratter-Sehn, A., Sedlmayer, F., Wedrich, I., Auberger, T., 2004.Epidemiological aspects of hadron therapy: A prospective nationwide study of the Austrian projectMedAustron and the Austrian Society of Radiooncology (OEGRO). Radiother. Oncol. 73 (Suppl. 2),S24–S28. 6

Menzel, H., Pihet, P., Wambersie, A., 1990. Microdosimetric specification of radiation quality in neutronradiation therapy. Int. J. Radiat. Biol. 57 (4), 865–883. 37

Merchant, T. E., 2009. Proton beam therapy in pediatric oncology. Cancer J. 4 (15), 782–785. 5

Min, C.-H., Kim, C. H., Youn, M.-Y., Kim, J.-W., 2006. Prompt gamma measurements for locating the dosefalloff region in the proton therapy. Appl. Phys. Lett. 89 (18), 183517. 14

Niita, K., Sato, T., Iwase, H., Nose, H., Nakashima, H., Sihver, L., 2006. PHITS – A particle and heavy iontransport code system. Radiat. Meas. 41, 1080–1090. 17

Nikjoo, H., Emfietzoglou, D., Watanabe, R., Uehara, S., 2008. Can Monte Carlo track structure codes revealreaction mechanism in DNA damage and improve radiation therapy? Radiat. Phys. Chem. 77 (10-12),1270–1279, The International Symposium on Charged Particle and Photon Interaction with Matter -ASR 2007. 33

Nikjoo, H., Uehara, S., Emfietzoglou, D., Cucinotta, F., 2006. Track-structure codes in radiation research.Radiat. Meas. 41 (9-10), 1052–1074. 33

Okada, T., Kamada, T., Tsuji, H., etsu Mizoe, J., Baba, M., Kato, S., Yamada, S., Sugahara, S., Yasuda,S., Yamamoto, N., Imai, R., Hasegawa, A., Imada, H., Kiyohara, H., Jingu, K., Shinoto, M., Tsujii,H., 2010. Carbon ion radiotherapy: Clinical experiences at National Institute of Radiological Science(NIRS). J. Radiat. Res. 51 (4), 355–364. 5

Orecchia, R., Krengli, M., Jereczek-Fossa, B. A., Franzetti, S., Gerard, J. P., 2004. Clinical and researchvalidity of hadrontherapy with ion beams. CRC CR Rev. Oncol.-Hem. 51 (2), 81–90. 6

VIII

REFERENCES

Paganetti, H., 2009. Dose to water versus dose to medium in proton beam therapy. Phys. Med. Biol. 54 (14),4399. 48

Paganetti, H., 2012. Range uncertainties in proton therapy and the role of Monte Carlo simulations. Phys.Med. Biol. 57 (11), R99. 15, 47

Paganetti, H., Jiang, H., Parodi, K., Slopsema, R., Engelsman, M., 2008. Clinical implementation of fullMonte Carlo dose calculation in proton beam therapy. Phys. Med. Biol. 53, 4825–4853. 17, 18, 19, 51

Paganetti, H., Olko, P., Kobus, H., Becker, R., Schmitz, T., Waligorski, M. P., Filges, D., Müller-Gärtner,H.-W., 1997. Calculation of relative biological effectiveness for proton beams using biological weightingfunctions. Int. J. Radiat. Oncol. 37 (3), 719–729. 34, 37

Parodi, K., Bortfeld, T., Enghardt, W., Fiedler, F., Knopf, A., Paganetti, H., Pawelke, J., Shakirin, G., Shih,H., 2008a. PET imaging for treatment verification of ion therapy: Implementation and experience at GSIDarmstadt and MGH Boston. Nucl. Instrum. Meth. A 591 (1), 282–286. 13, 41

Parodi, K., Bortfeld, T., Haberer, T., 2008b. Comparison between in-beam and offline positron emissiontomography imaging of proton and carbon ion therapeutic irradiation at synchrotron- and cyclotron-based facilities. Int. J. Radiat. Oncol. 71 (3), 945 – 956. 13, 41

Parodi, K., Brons, S., Cerutti, F., Ferrari, A., Mairani, A., Paganetti, H., Sommerer, F., 2009a. The FLUKAcode for application of Monte Carlo methods to promote high precision ion beam therapy. Proceed. 12thInt. Conf. on Nucl. Reac. Mechan., Varenna, Italy, 15th-19th June, book of proceedings, available athttp://cdsweb.cern.ch/record/1238366/files/p509.pdf. 17, 19

Parodi, K., Ferrari, A., Sommerer, F., Paganetti, H., 2007a. Clinical CT-based calculations of dose andpositron emitter distributions in proton therapy using the FLUKA Monte Carlo code. Phys. Med. Biol.52 (12), 3369. 18, 47

Parodi, K., Mairani, A., Brons, S., Hasch, B. G., Sommerer, F., Naumann, J., Jäkel, O., Haberer, T., Debus,J., 2012. Monte Carlo simulations to support start-up and treatment planning of scanned proton andcarbon ion therapy at a synchrotron-based facility. Phys. Med. Biol. 57 (12), 3759. 17, 18, 19, 30

Parodi, K., Mairani, A., Brons, S., Naumann, J., Krämer, M., Sommerer, F., Haberer, T., 2010. The influenceof lateral beam profile modifications in scanned proton and carbon ion therapy: A Monte Carlo study.Phys. Med. Biol. 55 (17), 5169. 11

Parodi, K., Paganetti, H., Shih, H. A., Michaud, S., Loeffler, J. S., DeLaney, T. F., Liebsch, N. J., Mun-zenrider, J. E., Fischman, A. J., Knopf, A., Bortfeld, T., 2007b. Patient study of in vivo verification ofbeam delivery and range, using positron emission tomography and computed tomography imaging afterproton therapy. Int. J. Radiat. Oncol. 68 (3), 920–934. 15, 41, 51

Parodi, K., Ponisch, F., Enghardt, W., 2005. Experimental study on the feasibility of in-beam PET foraccurate monitoring of proton therapy. IEEE T. Nucl. Sci. 52 (3), 778–786. 41

Pedroni, E., Bacher, R., Blattmann, H., Bohringer, T., Coray, A., Lomax, A., Lin, S., Munkel, G., Scheib,S., Schneider, U., 1995. The 200-MeV proton therapy project at the Paul Scherrer Institute: Conceptualdesign and practical realization. Med. Phys. 22 (1), 37–53. 7

Peterson, S., Polf, J. C., Bues, M., Ciangaru, G., Archambalt, L., Beddar, S., Smith, A. R., 2009. Clinicalimplementation of full Monte Carlo dose calculation in proton beam therapy. Phys. Med. Biol. 54,3217–3229. 17

Petti, P. L., Lennox, A. J., 1994. HIPLAN – A heavy ion treatment planning system at HIMAC. Annu. Rev.Nucl. Part. Sci. 44, 155–197. 30

Pflugfelder, D., Wilkens, J. J., Oelfke, U., 2008. Worst case optimization: a method to account for un-certainties in the optimization of intensity modulated proton therapy. Phys. Med. Biol. 53 (6), 1689.12

IX

REFERENCES

Pleskac, R., et al., 2012 see paper II∗. 30, 31

Polf, J. C., Peterson, S., McCleskey, M., Roeder, B. T., Spiridon, A., Beddar, S., Trache, L., 2009b. Mea-surement and calculation of characteristic prompt gamma ray spectra emitted during proton irradiation.Phys. Med. Biol. 54 (22), N519. 41

PTCOG, 2012. Particle therapy co-operative group. Webpage: http://ptcog.web.psi.ch/. 1

Quesada, J. M., 2010. Private communication. 29

Richard, M.-H., Chevallier, M., Dauvergne, D., Freud, N., Henriquet, P., Le Foulher, F., Létang, J., Mon-tarou, G., Ray, C., Roellinghoff, F., Testa, E., Testa, M., Walenta, A., 2011. Design guidelines fora double scattering compton camera for prompt-γ imaging during ion beam therapy: A Monte Carlosimulation study. IEEE T. Nucl. Sci. 58 (1), 87–94. 42

Richard, M.-H., Chevallier, M., Dauvergne, D., Freud, N., Henriquet, P., Le Foulher, F., Letang, J. M.,Montarou, G., Ray, C., Roellinghoff, F., Testa, E., Testa, M., Walenta, A. H., 2009. Design study ofa Compton camera for prompt-γ imaging during ion beam therapy. In: Nuclear Science SymposiumConference Record (NSS/MIC), 2009 IEEE. pp. 4172–4175. 42

Rietzel, E., Schardt, D., Haberer, T., 2007. Range accuracy in carbon ion treatment planning based onCT-calibration with real tissue samples. Radiat. Oncol. 2, 14. 9, 19

Rinaldi, I., 2011. Investigation of novel imaging methods using therapeutic ion beams. Ph.D. thesis, RupertoCarola University of Heidelberg, Germany. 13

Roesler, S., Engel, R., Ranft, J., 2001. The Monte Carlo event generator DPMJET III. Proceed. of the Mon-teCarlo 2000 Conference, Lisbon, 23-26 October 2000, book of proceedings Springer-Verlag, Berlin,1033–1038. 22

Rogers, D. W. O., 2006. Fifty years of Monte Carlo simulations for medical physics. Phys. Med. Biol.51 (13), R287. 15

Rollet, S., Angelone, M., Magrin, G., Marinelli, M., Milani, E., Pillon, M., Prestopino, G., Verona, C.,Verona-Rinati, G., 2012. A novel microdosimeter based upon artificial single crystal diamond. IEEE T.Nucl. Sci. 59 (5), 2409–2415. 39

Rollet, S., Colautti, P., Grosswendt, B., Herault, J., Wind, M., Gargioni, E., Beck, P., Latocha, M., Moro, D.,2010. Microdosimetric assessment of the radiation quality of a therapeutic proton beam: Comparisonbetween numerical simulation and experimental measurements. Radiat. Prot. Dosim. 45, 1330–1333.34, 37, 39

Rossi, H. H., Rosenzweig, W., 1955. A device for the measurement of dose as a function of specific ioniza-tion. Radiology 64, 404–410. 34

Rudin, M., Weissleder, R., 2003. Molecular imaging in drug discovery and development. Nat. Rev. DrugDiscov. 2, 123–131. 42

Sato, T., Kase, Y., Watanabe, R., Niita, K., Sihver, L., 2009. Biological dose estimation for charged-particletherapy using an improved PHITS code coupled with a microdosimetric kinetic model. Radiat. Res.171 (1), 107–117. 38, 39

Sato, T., Watanabe, R., Kase, Y., Tsuruoka, C., Suzuki, M., Furusawa, Y., Niita, K., 2011. Analysis ofcell-survival fractions for heavy-ion irradiations based on microdosimetric kinetic model implementedin the particle and heavy ion transport code system. Radiat. Prot. Dosim. 143 (2-4), 491–496. 38

Schaffner, B., Pedroni, E., 1998. The precision of proton range calculations in proton radiotherapy treatmentplanning: experimental verification of the relation between CT-HU and proton stopping power. Phys.Med. Biol. 43 (6), 1579. 9, 10, 19

X

REFERENCES

Schaffner, B., Pedroni, E., Lomax, A., 1999. Dose calculation models for proton treatment planning usinga dynamic beam delivery system: An attempt to include density heterogeneity effects in the analyticaldose calculation. Phys. Med. Biol. 44 (1), 27. 11

Schardt, D., Elsässer, T., Schulz-Ertner, D., 2010. Heavy-ion tumor therapy: Physical and radiobiologicalbenefits. Rev. Mod. Phys. 82, 383–425. 6, 7

Schlegel, W., 2010. If you can’t see it, you can miss it: The role of biomedical imaging in radiation oncol-ogy. Radiat. Prot. Dosim. 139 (1-3), 321–326. 42

Schlegel, W., Bortfeld, T., Grosu, A. L., 2006. 3D Imaging for Radiotherapy. New Technologies in Radia-tion Oncology. Springer Heidelberg. 8

Schneider, W., Bortfeld, T., Schlegel, W., 2000. Correlation between CT numbers and tissue parametersneeded for Monte Carlo simulations of clinical dose distributions. Phys. Med. Biol. 45 (2), 459. 47

Scholz, M., Kellerer, A. M., Kraft-Weyrather, W., Kraft, G., 1997. Computation of cell survival in heavyion beams for therapy. Radiat. Environ. Biophys. 36, 59–66. 4

Schulz-Ertner, D., 2009. The clinical experience with particle therapy in adults. Cancer J. 15 (4), 306–311.5

Seo, H., Park, J. H., Ushakov, A., Kim, C. H., Kim, J. K., Lee, J. H., Lee, C. S., Lee, J. S., 2011. Experi-mental performance of double-scattering Compton camera with anthropomorphic phantom. J. Instrum.6 (01), C01024. 42

Shibuya, K., Yoshida, E., Nishikido, F., Suzuki, T., Tsuda, T., Inadama, N., Yamaya, T., Murayama, H.,2007. Annihilation photon acollinearity in PET: Volunteer and phantom FDG studies. Phys. Med. Biol.52 (17), 5249. 43

Sommerer, F., Cerutti, F., Parodi, K., Ferrari, A., Enghardt, W., Aiginger, H., 2009. In-beam PET monitoringof mono-energetic 16O and 12C beams: Experiments and FLUKA simulations for homogeneous targets.Phys. Med. Biol. 54 (13), 3979. 18, 43

Sommerer, F., Unholtz, D., Mairani, A., Bauer, J., Parodi, K., 2012. On the integration of CT-based FLUKAcalculations of absorbed and RBE-weighted dose and dose-averaged LET in the clinical environment.manuscript. 17, 48

Sorge, H., Stöcker, H., Greiner, W., 1989a. Poincaré invariant Hamiltonian dynamics: Modelling multi-hadronic interactions in a phase space approach. Ann. Phys. 192 (2), 266–306. 23, 27

Sorge, H., Stöcker, H., Greiner, W., 1989b. Relativistic quantum molecular dynamics approach to nuclearcollisions at ultrarelativistic energies. Nucl. Phys. A 498, 567–576. 23, 27

Soukup, M., Fippel, M., Alber, M., 2005. A pencil beam algorithm for intensity modulated proton therapyderived from Monte Carlo simulations. Phys. Med. Biol. 50 (21), 5089. 11

St. Clair, W. H., Adams, J. A., Bues, M., Fullerton, B. C., La Shell, S., Kooy, H. M., Loeffler, J. S.,Tarbell, N. J., 2004. Advantage of protons compared to conventional x-ray or IMRT in the treatment ofa pediatric patient with medulloblastoma. Int. J. Radiat. Oncol. 58 (3), 727–734. 2, 5

Suit, H., DeLaney, T., Goldberg, S., Paganetti, H., Clasie, B., Gerweck, L., Niemierko, A., Hall, E., Flanz,J., Hallman, J., Trofimov, A., 2010. Proton vs carbon ion beams in the definitive radiation treatment ofcancer patients. Radiother. Oncol. 95 (1), 3–22. 5

Testa, E., Bajard, M., Chevallier, M., Dauvergne, D., Foulher, F. L., Freud, N., Létang, J., Poizat, J., Ray,C., Testa, M., 2008. Monitoring the Bragg peak location of 73 MeV/u carbon ions by means of promptgamma-ray measurements. Appl. Phys. Lett. 93 (9), 093506. 14

XI

REFERENCES

Testa, E., Bajard, M., Chevallier, M., Dauvergne, D., Foulher, F. L., Freud, N., Létang, J., Poizat, J., Ray, C.,Testa, M., 2009. Dose profile monitoring with carbon ions by means of prompt-gamma measurements.Nucl. Instrum. Meth. B 267 (6), 993–996. 14

Testa, M., Bajard, M., Chevallier, M., Dauvergne, D., Freud, N., Henriquet, P., Karkar, S., Le Foulher, F.,Létang, J., Plescak, R., Ray, C., Richard, M.-H., Schardt, D., Testa, E., 2010. Real-time monitoring ofthe Bragg-peak position in ion therapy by means of single photon detection. Radiat. Environ. Bioph. 49,337–343. 14

Tourovsky, A., Lomax, A. J., Schneider, U., Pedroni, E., 2005. Monte Carlo dose calculations for spotscanned proton therapy. Phys. Med. Biol. 50 (5), 971. 16, 17

Tripathi, R. K., Cucinotta, F. A., Wilson, J. W., 1999. Accurate universal parameterization of absorptioncross-sections III – Light systems. Nucl. Instr. Meth. B 155, 349–356. 24

Unholtz, D., Sommerer, F., Bauer, J., van Straaten, D., Haberer, T., Debus, J., Parodi, K., 2011. Post-therapeutical β+-activity measurements in comparison to simulations towards in-vivo verification ofion beam therapy. In: Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC), 2011IEEE. pp. 2273–2276. 48

Unkelbach, J., Soukup, M., Alber, M., Bortfeld, T., 2009. Range, setup and dose calculation errors inIMPT and their interrelation. In: Dössel, O., Schlegel, W. C., Magjarevic, R. (Eds.), World Congress onMedical Physics and Biomedical Engineering, September 7 - 12, 2009, Munich, Germany. Vol. 25/1 ofIFMBE Proceedings. Springer Berlin Heidelberg, pp. 900–903. 12

Wang, C.-K. C., Zhang, X., Gifford, I., Burgett, E., Adams, V., Al-Sheikhly, M., 2007. Experimental valida-tion of the new nanodosimetry-based cell survival model for mixed neutron and gamma-ray irradiation.Phys. Med. Biol. 52 (17), N367. 37

Wedenberg, M., Lind, B. K., Toma-Dasu, I., Rehbinder, H., Brahme, A., 2010. Analytical description of theLET dependence of cell survival using the repairable-conditionally repairable damage model. Radiat.Res. 174 (4), 517–525. 34

Weisskopf, V., Aug 1937. Statistics and nuclear reactions. Phys. Rev. 52, 295–303. 26

Weisskopf, V. F., Ewing, D. H., 1940. On the yield of nuclear reactions with heavy elements. Phys. Rev. 57,472–485. 26

Widesott, L., Pierelli, A., Fiorino, C., Lomax, A. J., Amichetti, M., Cozzarini, C., Soukup, M., Schnei-der, R., Hug, E., Muzio, N. D., Calandrino, R., Schwarz, M., 2011. Helical tomotherapy vs. intensity-modulated proton therapy for whole pelvis irradiation in high-risk prostate cancer patients: Dosimetric,normal tissue complication probability, and generalized equivalent uniform dose analysis. Int. J. Radiat.Oncol. 80 (5), 1589–1600. 2

Yepes, P. P., Mirkovic, D., Taddei, P. J., 2010. A GPU implementation of a track-repeating algorithm forproton radiotherapy dose calculations. Phys. Med. Biol. 55 (23), 7107. 16

Yonai, S., Matsufuji, N., Namba, M., 2012. Calculation of out-of-field dose distribution in carbon-ionradiotherapy by Monte Carlo simulation. Med. Phys. 39 (8), 5028–5039. 3, 18

XII