• Aucun résultat trouvé

Prompt gamma imaging with a slit camera for real time range control in particle therapy

N/A
N/A
Protected

Academic year: 2021

Partager "Prompt gamma imaging with a slit camera for real time range control in particle therapy"

Copied!
224
0
0

Texte intégral

(1)

Prompt gamma imaging with a slit camera for real time range control in particle therapy

Julien SMEETS

Th` ese pr´ esent´ ee en vue de l’obtention du grade de Docteur en Sciences de l’Ing´ enieur

Universit´ e Libre de Bruxelles Ecole polytechnique de Bruxelles ´

Service de M´ etrologie Nucl´ eaire

Ann´ ee acad´ emique 2011-2012

(2)
(3)

Thesis directors :

P

r

Alain DUBUS (ULB) P

r

Jean-Claude DEHAES (ULB)

Collaborators :

Frauke ROELLINGHOFF (IBA) D

r

Arthur BENILOV (IBA)

Project leaders :

Damien PRIEELS (IBA)

D

r

Fr´ ed´ eric STICHELBAUT (IBA)

Members of the jury :

P

r

Denis DAUVERGNE (UCBL) P

r

Jean-Claude DEHAES (ULB)

P

r

Alain DUBUS (ULB) P

r

Pascal KOCKAERT (ULB) P

r

Jean-Marc SPARENBERG (ULB)

D

r

Edmond STERPIN (UCL) D

r

Fr´ ed´ eric STICHELBAUT (IBA)

P

r

Stefaan TAVERNIER (VUB)

(4)
(5)

Abstract

In a growing number of cutting edge centres around the world, radiotherapy treatments delivered by beams of protons and carbon ions offer the opportunity to target tumours with unprecedented conformality. But a sharper dose distribu- tion increases the need for efficient quality control. Treatments are still affected by uncertainties on the penetration depth of the beam within the patient, requi- ring medical physicists to add safety margins. To reduce these margins and deliver safer treatments, different projects investigate real time range control by imaging prompt gammas emitted along the proton or carbon ion tracks in the patient.

This thesis reports on the feasibility, development and test of a new type of prompt gamma camera for proton therapy. This concept uses a knife-edge slit colli- mator to obtain a 1-dimensional projection of the beam path on a gamma camera.

It was optimized, using the Monte Carlo code MCNPX version 2.5.0, to select high energy photons correlated with the beam range and detect them with both high counting statistics and sufficient spatial resolution for use in clinical routine. To va- lidate the Monte Carlo model, spectrometry measurements of secondary particles emitted by a PMMA target during proton irradiation at 160 MeV were realised.

An excellent agreement with the simulations was observed when using subtraction methods to isolate the gammas in direct incidence. A first prototype slit camera using the HiCam gamma detector was consequently prepared and tested success- fully at 100 and 160 MeV beam energies. If we neglect electronic dead times and rejection of detected events, the current solution with its collimator at 15 cm from beam axis can achieve a 1-2 mm standard deviation on range estimation in a homo- geneous PMMA target for numbers of protons that correspond to doses in water at Bragg peak as low as 15 cGy at 100 MeV and 25 cGy at 160 MeV assuming pencil beams with a Gaussian profile of 5 mm sigma at target entrance.

This thesis also investigates the applicability of the slit camera for carbon ion therapy. On the basis of Monte Carlo simulations with the code MCNPX ver- sion 2.7.E, this type of camera appears not to be able to identify the beam range with the required sensitivity. The feasibility of prompt gamma imaging itself seems questionable at high beam energies given the weak correlation of secondaries lea- ving the patient.

(6)

ABSTRACT

(7)

Remerciements

Je souhaite tout d’abord exprimer ma gratitude envers le FNRS pour mes man- dats d’aspirant, et M. Jongen pour la confiance qu’il m’a accord´ee en me donnant l’opportunit´e de travailler sur ce projet excitant et ambitieux.

Je tiens `a remercier Alain et Jean-Claude de m’avoir enseign´e les concepts mo- bilis´es dans ce travail et de l’avoir accompagn´e.

Je remercie tout particuli`erement Damien et Fr´ed´eric pour l’encadrement du projet, de mˆeme que Frauke et Arthur avec qui j’ai eu la chance de travailler. Ces r´esultats sont aussi les leurs.

Je dois malheureusement renoncer `a citer tous ceux qui, comme Fran¸cois, C´e- dric et S´ebastien, ont apport´e une contribution indispensable aux mesures exp´eri- mentales, mais je les remercie sinc`erement.

Je f´elicite les coll`egues de Milan, Paolo, Roberta, Irene, Andrea, Carlo, Tom- maso et Manuel pour la r´ealisation du d´etecteur qui a permis la r´eussite de ces mesures.

Je remercie ´egalement le Pr Tavernier pour ses conseils ´eclair´es.

Je salue encore les coll`egues de Bordet, Lyon et Maastricht pour les ´echanges tr`es int´eressants.

Un tout grand merci `a Frauke, Thibault et Jean-Claude pour leurs retours sur le manuscrit.

Je remercie enfin les membres du jury d’avoir accept´e d’examiner mon travail.

Je souhaite beaucoup de succ`es `a Thibault, Julien, Pierre, Farshid et Jonathan.

(8)

REMERCIEMENTS

(9)

Foreword

Particle therapy offers precision, it demands accuracy.

The present thesis investigates prompt gamma imaging with a slit camera in proton and carbon ion therapy. Our aim is to try and address the issue of range uncertainty by performing real time range monitoring during particle therapy treat- ment.

The thesis is organized in three parts.

In the first part, we introduce the issue of range uncertainty in particle therapy.

In chapter 1, we identify the sources of range uncertainty and their consequences.

Then we distinguish the various solutions under investigation to solve this prob- lem. In chapter 2, we focus on the research efforts in prompt gamma imaging. We list the design options and objectives, as well as the preliminary results, pros and cons of the different camera concepts that have been proposed.

In the second part, we develop the slit camera concept for proton therapy.

Chapter 3 reports on the feasibility, optimization and performance evaluation of such a camera by means of Monte Carlo simulations. Chapter 4 directly compares simulation results with spectrometry measurements of prompt gamma emissions during proton irradiation to assess the accuracy of the simulations and validate the conclusions of the simulation study. As a result, a first prototype was built to establish a first proof of concept. These measurements are analysed in chapter 5.

In view of the preparation of new improved prototypes, chapter 6 finally details the sensitivity of the camera to the values of the parameters that determine its performances.

In the third part, we study the feasibility of the slit camera concept for carbon ion therapy. Chapter 7 describes the difficulties faced with carbon ions by means of Monte Carlo simulations and chapter 8 evaluates the accuracy of these simulations by comparison with experimental results of the Lyon group.

I hope you’ll have a pleasant, critical and insightful reading!

(10)

FOREWORD

(11)

Contents

1 Range uncertainty in particle therapy 1

1.1 The nature of the problem . . . 2

1.2 Solutions under investigation . . . 6

1.3 Conclusion . . . 23

2 State of the art in prompt gamma imaging 25 2.1 Design options . . . 26

2.2 Design objectives . . . 29

2.3 Collimated cameras . . . 30

2.4 Non-collimated cameras . . . 37

3 Monte Carlo design study for proton therapy 41 3.1 Monte Carlo simulations . . . 42

3.2 Reference setup . . . 44

3.3 Secondary photons and neutrons . . . 45

3.4 Detection profile with a perfect system . . . 49

3.5 Optimizing the camera . . . 52

3.6 Influence of beam energy . . . 58

3.7 Influence of distance . . . 62

3.8 Performance evaluation . . . 64

3.9 Application to a treatment plan . . . 67

3.10 Time requirements . . . 68

4 Assessment of the simulations for protons 69 4.1 Experimental setup . . . 70

4.2 Measured spectra . . . 71

4.3 Comparison with the simulations . . . 74

4.4 Conclusion . . . 78

4.5 Further considerations . . . 79

5 Experimental evaluation for proton therapy 81 5.1 HiCam camera . . . 82

(12)

CONTENTS

5.5 Measured profiles . . . 88

5.6 Range control . . . 91

5.7 Further measurements . . . 95

6 Parameter sensitivity for proton therapy 97 6.1 Reference setup . . . 98

6.2 Monte Carlo simulations . . . 99

6.3 Detection profile characteristic values . . . 100

6.4 Collimator material . . . 102

6.5 Collimator thickness . . . 109

6.6 Collimator slit angle . . . 112

6.7 Collimator slit width . . . 115

6.8 Scintillator material . . . 118

6.9 Scintillator thickness . . . 125

6.10 Scintillator segment width . . . 129

6.11 Energy selection window . . . 131

6.12 Time-of-flight discrimination . . . 136

6.13 Current mode . . . 139

6.14 Conclusion . . . 141

7 Monte Carlo feasibility study for carbon ions 143 7.1 Convention . . . 144

7.2 Monte Carlo simulations . . . 144

7.3 Secondary emissions . . . 145

7.4 Detection profile with the proton solution . . . 149

7.5 Time-of-flight spectra . . . 154

7.6 Detection profile with a perfect system . . . 157

7.7 TOF selection with the proton solution . . . 161

7.8 Conclusion . . . 163

8 Assessment of the simulations for carbon ions 165 8.1 Experimental results of the Lyon group . . . 166

8.2 Comparison with the simulations . . . 169

8.3 Conclusion . . . 175

Conclusions and outlook 177

A Count rate requirements 181

B Influence of target composition 183

C Beam entrance in the target 191

D Mean ionization potential 197

(13)

Chapter 1

Range uncertainty in proton and carbon ion therapy

This first chapter is dedicated to the description of the motivation of our work.

We first examine the problem posed by range uncertainties in proton and carbon ion therapy. Then we list the different approaches that are investigated to solve this problem. Three categories are distinguished: pretreatment methods, monitoring methods and post treatment methods. We review their status and detail their fields of applications, strengths and weaknesses.

(14)

CHAPTER 1. RANGE UNCERTAINTY IN PARTICLE THERAPY

1.1 The nature of the problem

Nowadays, proton and carbon ion beams offer the possibility to accurately tar- get tumours thanks to their rising dose distribution culminating at the Bragg peak (cf. figure 1.1). The energy of the beam is selected to benefit from the finite range of the ions: concentrate high dose in the treatment volume while sparing distal healthy tissues. This physical advantage is also a challenge for medical physicists and radiation oncologists given the uncertainties affecting the beam range within the patient. The more conformal the dose distribution, the more sensitive to any error.

Figure 1.1. Simulated depth-dose distributions of monoenergetic pencils beams of protons (top) and carbon ions (bottom) in water, with Monte Carlo code MCNPX version 2.7.E.

(15)

1.1. THE NATURE OF THE PROBLEM Although proton therapy systems can select beam energy with high accuracy, different sources of range uncertainty have been identified. They can be ranked in two categories (Bortfeld et al. 2007):

1. Dose calculation errors (up to ∼5 mm):

• Artifacts in the planning X-ray computed tomography (CT) scan, in par- ticular in highly inhomogeneous tissues or in the presence of metallic im- plants.

• Ambiguity in the conversion of CT number (Hounsfield Unit) to proton and carbon ion stopping power (Schaffner and Pedroni 1998, J¨akel et al.

2001, Jiang et al. 2001).

• Treatment planning algorithm approximations.

2. Differences between treatment preparation and delivery (up to∼10 mm):

• Setup errors: patient mispositioning, misalignment between the patient and the range compensator (tailoring the distal edge of the beam to the posterior surface of the target volume in passive scattering mode).

• Internal organ motion: breathing.

• Anatomical changes between treatment fractions: weight change, tumour shrinkage, cavity filling.

All these sources can add up to a maximum uncertainty of 10 to 15 mm (Bortfeld et al. 2007).

The problem is illustrated in figure 1.2 by Lu (2007). An undershoot (a pene- tration depth shorter than prescribed) results in serious underdosage of the target volume, while anovershoot (a range deeper than required) could additionally dam- age an organ at risk beyond the tumour. The issue is much less critical for high energy photons whose dose profile is much less steep.

Figure 1.2. Potential dose difference resulting from a 1.0 cm water equivalent depth for a 10 MV photon beam and a proton SOBP (spread

(16)

CHAPTER 1. RANGE UNCERTAINTY IN PARTICLE THERAPY

Range uncertainty needs to be taken into account when designing the treat- ment plan with practical consequences that compromise part of the benefit of particle therapy. The beam range is prescribed with a safety margin to ensure target volume coverage. As a result, surrounding healthy tissues are part of the target volume as we can observe on the prostate treatment plan in figure 1.3 from Trofimov et al. (2007).

Figure 1.3. Two-field prostate treatment plan with safety margin.

From Trofimov et al. (2007).

Paganetti (2012) recently reviewed and quantified the different sources of un- certainties. Their potential reduction thanks to the use of Monte Carlo simula- tions was also considered. The resulting estimations of the total uncertainty are reported in figure 1.4 and compared to the margins applied in different proton therapy centres. At MGH in Boston for example, the standard range margin ap- plied is 3.5 % + 1 mm. This means an intentional overshoot of 8 mm for a 20 cm range.

Another typical consequence of range uncertainty is the use of the beam lat- eral edge instead of the distal edge to spare organs at risk. For example, lateral fields are used in prostate treatment to spare the rectum wall, despite the fact that an anterior field provides a better conformation to the target volume near the rectum as in figure 1.5 from Tang et al. (2012). The anterior field relies on a sharp distal penumbra (the dose level falls from 95 % to 50 % over a distance of

∼4 mm), while the lateral fields rely on a broader lateral penumbra (∼10 mm at 50-95 %) but benefit from a lower uncertainty (Lu 2008b). In the absence of range uncertainties, the anterior field would be an advantageous option. Furthermore, the femoral heads are less exposed by an anterior field. In a more general context, the combination of several lateral beams to bypass a critical structure is referred to as field patching. Range uncertainties thus complicate the treatment planning process, increase of the number of fields and so the treatment time.

(17)

1.1. THE NATURE OF THE PROBLEM

Figure 1.4. Dotted lines: typically applied range uncertainty mar- gins in proton therapy treatment planning as currently applied at the Massachusetts General Hospital in Boston (3.5 % + 1 mm), the MD An- derson Proton Therapy Center in Houston (3.5 % + 3 mm), the Loma Linda University Medical Center (3.5 % + 3 mm), the Roberts Proton Therapy Center at the University of Pennsylvania (3.5 % + 3 mm) and the University of Florida Proton Therapy Institute (2.5 % + 1.5 mm).

Note that these centres may apply bigger margins in specific treat- ment scenarios. Dashed line: estimated uncertainty without the use of Monte Carlo dose calculation. Solid line: estimated uncertainty for complex geometries without the use of Monte Carlo dose calcula- tion. Dashed-dotted line: estimated uncertainty with the use of Monte Carlo dose calculation. Estimations are based on 1.5 standard devia- tions. Edited from Paganetti (2012).

Figure 1.5. Comparison of dose distributions for a prostate treated by two lateral fields (left) or one anterior-posterior field (right). Prostate,

(18)

CHAPTER 1. RANGE UNCERTAINTY IN PARTICLE THERAPY

1.2 Solutions under investigation

In the present brief status review, we will classify different solutions under investigation to address the issue of range uncertainty in particle therapy in three categories: first the methods that aim at reducing uncertainties before treatment delivery, then the methods that aim at monitoring the treatment during delivery and finally post treatment verification methods. This classification is not rigid and we will see that some methods may change of category depending on the way they are implemented.

1.2.1 Pretreatment reduction of uncertainties

Reducing the different sources of range uncertainties to a minimum before treatment delivery is of course the best practice. In this section, one should not overlook that we omit to describe the wide range of techniques related to motion management, not only patient motion but also internal organ motion, through immobilization devices, respiration gating, tumour tracking, etc. that eventually reduce range uncertainties by accurately ascertaining the target position.

Robust planning

The robust planning approach aims at reducing uncertainties on the treatment outcome. It takes range uncertainties into account at the treatment planning stage to mitigate their potential consequences at the treatment delivery stage. Efforts are made so that range uncertainties would not lead to significant underdosage of the target volume or excessive exposition of a critical healthy structure. This is commonly integrated by medical physicists who apply such principles as field patching, safety margins, using the lateral penumbra rather than the distal one to avoid critical structures, avoid beam angles with heterogeneities in their paths (in particular metal implants), etc.

These principles also need to be integrated in treatment plan optimization soft- ware. The introduction of new proton therapy systems able to perform pencil beam scanning opens the door for IMPT (intensity modulated proton therapy) thanks to non-uniform beams. The development of efficient treatment planning strategies for IMPT requires a lot of effort given the higher number of degrees of freedom available. The different treatment plan solutions are evaluated by software to come up with a good solution and the robustness with respect to range uncertainties has to be included in the score functions. Different methods have been proposed. Un- kelbach et al. (2007) presented two approaches to reduce sensitivity to both range uncertainties and setup errors. The first approach is probabilistic: the range of a pencil beam is treated as a random variable and the average treatment plan quality is optimized in consequence (Unkelbach et al. 2009). The second approach is a ro- bust formulation that optimizes the worst case dose distribution resulting from the variation of the pencil beam ranges within specified domains. A different worst- case strategy was also implemented by Pflugfelder et al. (2008) considering a dose

(19)

1.2. SOLUTIONS UNDER INVESTIGATION distribution in which each voxel receives the worst dose it can be attributed from any scenario. More recently, Chen et al. (2012) included robustness in a multi- criteria optimization (MCO) framework. Rather than optimizing a score function for given weights of the different objectives, the treatment planner is offered the possibility to examine the trade-off between robustness and nominal plan quality by exploring a collection of different plans optimizing different objectives.

Proton radiography

Range uncertainties could be efficiently reduced by performing a proton ra- diography with the therapeutic beam before every treatment fraction is delivered.

Taking a radiograph with a proton beam of sufficient energy for the protons to exit the patient would give the possibility to measure the proton residual energy thanks to a position-sensitive range telescope and compare it to the predictions.

In this way, physicists could identify for example setup errors and evolutions of the tumour size causing under- or overshoot. If the high energy protons used for the radiography go through the whole patient thickness and exit with the expected residual range, one could be reasonably confident that the lower energy protons used for treatment would stop at the prescribed penetration depth. Ideally, the treatment would even be planned using proton computed tomography data. This would eliminate the present ambiguity in the conversion of CT numbers to proton stopping powers using classical X-ray CT scans.

Radiography and computed tomography with beams of protons and heavy ions have been investigated by different groups who confirmed the great potential of these approaches, but they are not yet used in clinical routine as both technical and financial difficulties are still faced. The current status was recently presented by Rinaldi (2011). Sufficiently high proton beam energies are necessary to escape all patients with financial consequences on the design of the beam production system and the corresponding shielding. Every proton exiting the patient contributes to the information collected for a proton radiograph, so that this pre-irradiation would cause a lower dose than classical X-ray shots. A radiography setup was implemented at Paul Scherrer Institute (PSI) in Villigen and Schneider et al. (2004) took proton images of the head of a dog patient showing that the dose during exposure was only 0.03 mGy, which is approximately a factor 50 to 100 smaller than for a comparable X-ray image. Proton radiography also offers a better contrast, i.e. a better density resolution, but a poorer spatial resolution compared to X-rays as recently confirmed by Depauw and Seco (2011). The reason is that multiple Coulomb scattering of protons results in non-purely straightforward trajectories.

Carbon ions are less affected by multiple Coulomb scattering, but not only primary ions leave the patient: a significant number of secondary fragmentation products

(20)

CHAPTER 1. RANGE UNCERTAINTY IN PARTICLE THERAPY Dual-energy X-ray computed tomography

As long as proton computed tomography is not available, dual-energy X-ray computed tomography is another option to reduce range uncertainties due to the ambiguity in the conversion of CT numbers to proton stopping powers. During the usual treatment planning process, images of the patient are acquired with a classi- cal single energy X-ray CT scanner. The CT numbers are then converted to proton stopping power ratios to deduce water-equivalent path lengths and estimate the range of the proton beam. This conversion from CT numbers to proton stopping power ratios is realised thanks to a calibration curve based on CT acquisitions of various tissue equivalent phantoms. The problem is that both CT numbers and stopping power ratios depend on both the electron density and the elemental com- position of the tissue. Tissues with different densities and compositions can yield the same CT numbers, while having different proton stopping power ratios, and vice versa. Calibration methods commonly rely on standard tissue compositions established by the ICRU (1989). In practice, of course, significant variations can be observed in the actual elemental tissue composition of a particular patient.

According to a recent study of Yang et al. (2012), the degeneracy of Hounsfield numbers in the presence of these tissue composition variations is the dominant contribution to uncertainties in stopping power ratios in soft tissues. As a result, Yang et al. (2012) emphasize the interest of dual-energy computed tomography. A dual-energy CT scanner uses two different X-ray energy spectra. The two sets of data and their energy dependence give the possibility to simultaneously retrieve both the density and the effective atomic number Z of the tissues. In a previous study, Yang et al. (2011) already concluded that the combination of a kV and a MV X-ray spectrum for the dual-energy CT could be more effective in reducing errors than the more conventional combination of two kV X-ray spectra due to random noise and beam hardening effects.

Radioactive probing beam

For carbon ion therapy, a very efficient concept is the use of a radioactive beam.

The stable 12C isotope can be replaced by the positron emitting isotopes 10C or

11C with similar dose deposition curves. Measuring the penetration depth of the beam is then quite easy: most radioactive carbon nuclei stop around their mean projected range and emit positrons that produce annihilation photons that escape the patient and are detected by a gamma camera. This was first realised at the Lawrence Berkeley Laboratory in the 80’s with a19Ne beam and more recently at the Heavy-Ion Medical Accelerator (HIMAC) in Chiba with beams of10C and11C (Kanazawa et al. 2001). These radioactive carbon ion beams are produced by frag- mentation reactions of a stable 12C beam in a 51 mm thick beryllium target and the required isotope is selected on the base of its magnetic rigidity with a bending magnet followed by a slit. The production yield is 1 % for 11C at full acceptance and 0.04 % for10C at full acceptance (Iseki et al. 2004). Depending on the selected acceptance, the radioactive beam intensity is therefore significantly smaller than

(21)

1.2. SOLUTIONS UNDER INVESTIGATION 1 % of the primary beam current. Treating the patient with a radioactive beam therefore requires a very high intensity12C beam and the corresponding shielding.

Iseki et al. (2004) investigated the use of such a low current radioactive beam as a probing beam. Just before the treatment is delivered with the stable 12C beam, range measurement is realised with a small dose of radioactive beam. Iseki et al. (2004) found11C not to be advantageous because its long half-life (20.4 min) results in low count rates and the measurement is blurred by other positron emit- ters produced along the incident beam tracks in the patient. Thanks to a shorter half-life (19.3 s), beams of10C were preferred despite a lower production yield and a deeper maximum range of the emitted positrons (up to 9.2 mm for positrons of

10C and up to 4.2 mm for11C). Measurements were conducted with a10C beam of 345.8 MeV/u average energy incident on a PMMA target (PolyMethyl MethAcry- late, (C5H8O2)n, 1.19 g/cm3) and imaged by a pair of Anger-type scintillation detectors operating in time coincidence. The data analysis revealed that the range could be measured with an uncertainty as low as 0.3 mm for 2.7×105 ions im- planted, corresponding to a biological dose of only 9.6 cGyE (gray-equivalent dose taking into account a relative biological effectiveness of 4.1).

1.2.2 Real time treatment monitoring methods

After sources of uncertainty have been reasonably reduced to a minimum, real time monitoring of the beam range inside the patient would offer a feedback during treatment delivery: the treatment could be stopped and corrected before the entire fraction is delivered if the measured range does not match the prescription.

Time resolved diode dosimetry

Time resolved diode dosimetry is an extremely promising approach for quality control with passive scattering delivery systems. It is principally meant for prostate treatments where diodes are introduced in the rectum with a rectal balloon as il- lustrated in figure 1.6. Measurements on an anthropomorphic phantom already demonstrated millimetre accuracy in range control for an anterior-posterior field (cf. figure 1.5) and the technique is further developed at Massachusetts General Hospital (MGH) in Boston. This method could also be applied for other treat- ments on condition that diodes can be inserted in a body cavity.

Intra-cavity dosimeters are already used for quality control in conventional photon and electron radiotherapy. One specific difficulty for proton therapy is the flatness of the dose distribution before and beyond the Bragg peak as the dosimeter can generally not be placed right at the dose distal fall-off. If the dosimeter is placed before the Bragg peak in the plateau of the SOBP (spread out Bragg peak), the dose distribution is flat and gives no indication on the residual range of the protons.

(22)

CHAPTER 1. RANGE UNCERTAINTY IN PARTICLE THERAPY

Figure 1.6. Schematic view of an in vivo range verification with diode detectors (yellow dots) for a patient whose prostate (red delineation) is treated by an anterior proton field (yellow arrows) with an endorectal balloon (white curve). From Tang et al. (2012).

Lu (2008a) solved this difficulty by taking advantage of the time-dependence of the signal measured by the diodes exposed to the plateau of a SOBP. Indeed, pas- sively scattered proton therapy systems deliver SOBP by intercepting the incident beam with a rotating wheel composed of segments of variable thickness: the range modulator. The different segments of the wheel cause the necessary degradation of the beam to produce the different Bragg peak depths that build the SOBP. As a consequence, the dose rate measured by a diode at a fixed depth varies periodically as a function of time. Even if the diode measures the same total dose at all depths of the SOBP dose plateau, the time-dependence is different and thus encodes the residual range.

This method could be applied before and during a prostate treatment. When the patient is positioned, a probing beam would first deliver a low dose (less than 1 cGy) with a deeper range than prescribed so that the diodes are exposed in the plateau of the SOBP. The millisecond dose-time dependences measured by the diodes are compared to a database of reference dose-time functions (obtained by calibration measurements or Monte Carlo simulations) to deduce the residual range at these points. If this residual range does not match the expected value within specified margins (less than 1 mm), the range is adjusted. After this pre- treatment verification, the diodes measure the rectal dose continuously during the actual treatment delivery with the adjusted prescription range: protons are sup- posed to stop before the diodes to spare the anterior rectal wall. If for any reason this rectum dose gets too high, the treatment could be interrupted immediately, and resumed after a new adjustment of the range.

(23)

1.2. SOLUTIONS UNDER INVESTIGATION Lu (2008b) also proposed an alternative solution to the time dependence to encode the depth along the proton track in passive scattering mode. The idea is to split the single spread-out Bragg peak field with the usual plateau into two separate fields with a sloped depth-dose profile as in figure 1.7 from Lu (2008b).

The combination of the two sloped fields would still result in the original single field with plateau. In this way, the intra-cavity dosimeter would measure the dose from the two sloped fields and the ratio of these two doses would correspond to a unique depth along the beam axis.

Figure 1.7. Depth-dose profileD(x) for an example SOBP field and its complementary field pairDa(x) andDb(x) so thatDa(x) +Db(x) = D(x). From Lu (2008b).

Prompt gamma imaging

After collisions with protons, many nuclei of the patient tissue are left excited and decay by emitting neutrons or prompt gammas very quickly (10−19 to 10−9s).

Prompt gammas have a wide energy spectrum, mainly between 0 and 7 MeV, with a few characteristic rays of the nuclei that are present or produced in the target.

Reproducing data from Kozlovsky et al. (2002), gamma ray lines from proton re- actions with 12C and 16O are presented in table 1.1 and their cross sections are plotted in figure 1.8. The most intense peaks we can expect are 4.44 MeV due to

16O and 12C, 6.129 MeV, 6.916 MeV and 7.115 MeV due to 16O, and 0.718 MeV due to 12C. The emission yield of prompt gammas depends of course on the tar- get and the incident beam energy, but a good order of magnitude to remember is one prompt gamma emitted per 10 protons incident on the patient (cf. section 3.3).

(24)

CHAPTER 1. RANGE UNCERTAINTY IN PARTICLE THERAPY

Energy [MeV] Transition Reaction Mean life [s]

0.718 10B∗0.718 → g.s. 12C(p,x)10B 1.0×10−9

12C(p,x)10C()10B 27.8

16O(p,x)10B 1.0×10−9 1.022 10B∗1.74010B∗0.718 12C(p,x)10B 7.5×10−15

16O(p,x)10B 7.5×10−15 1.635 14N∗3.94814N∗2.313 16O(p,x)14N 6.9×10−15 2.000 11C∗2.000 → g.s. 12C(p,x)11C 1.0×10−14 2.124 11B∗2.125 → g.s. 12C(p,x)11B 5.5×10−15 2.313 14N∗2.313 →g.s. 16O(p,x)14N 9.8×10−14 2.742 16O∗8.87216O∗6.130 16O(p,p’)16O 1.8×10−13 3.684 13C∗3.685 → g.s. 16O(p,x)13C 1.6×10−15 3.853 13C∗3.854 → g.s. 16O(p,x)13C 1.2×10−11 4.438 12C∗4.439 → g.s. 12C(p,p’)12C 6.1×10−14

16O(p,x)12C 6.1×10−14 4.444 11B∗4.445 → g.s. 12C(p,2p)11B 5.6×10−19 5.105 14N∗5.106 →g.s. 16O(p,x)14N 6.3×10−12 5.180 15O∗5.181 → g.s. 16O(p,x)15O <4.9×10−14 5.240 15O∗5.241 → g.s. 16O(p,x)15O 3.3×10−12 5.269 15N∗5.270 →g.s. 16O(p,x)15N 2.6×10−12 5.298 15N∗5.299 →g.s. 16O(p,x)15N 1.2×10−14 6.129 16O∗6.130 → g.s. 16O(p,p’)16O 2.7×10−11 6.175 15O∗6.176 → g.s. 16O(p,x)15O <2.3×10−14 6.322 15N∗6.324 →g.s. 16O(p,x)15N 1.0×10−15 6.337 11C∗6.339 → g.s. 12C(p,x)11C <1.1×10−13 6.476 11C∗6.478 → g.s. 12C(p,x)11C <8.7×10−15 6.741 11B∗6.743 → g.s. 12C(p,x)11B 4.3×10−20 6.790 11B∗6.792 → g.s. 12C(p,x)11B 5.6×10−19 6.916 16O∗6.917 → g.s. 16O(p,p’)16O 6.8×10−15 7.115 16O∗7.117 → g.s. 16O(p,p’)16O 1.2×10−14 7.299 15N∗7.301 →g.s. 16O(p,x)15N 1.4×10−16 15.10 12C∗15.11 → g.s. 12C(p,p’)12C 1.5×10−17

Table 1.1. Gamma ray lines from proton reactions with12C and16O.

From data of Kozlovsky et al. (2002).

(25)

1.2. SOLUTIONS UNDER INVESTIGATION

Figure 1.8. Gamma production cross sections from proton reactions with 12C (top) and 16O (bottom). Plotted from data of Kozlovsky et al. (2002). Cross sections are drawn in the same color when they use the same data multiplied by different factors. For 12C, the 4.438 and 4.444 MeV peaks are merged as the 4.44 MeV peak.

(26)

CHAPTER 1. RANGE UNCERTAINTY IN PARTICLE THERAPY

Stichelbaut and Jongen (2003) suggested that one could try and measure those gammas that leave the patient to deduce the range. As prompt gammas are pro- duced along the proton tracks, the path of a pencil beam within the patient could be imaged as a line source by an adequate gamma camera. Thanks to Monte Carlo simulations, they predicted that the prompt gamma emission profile should be correlated with the proton trajectory. They emphasized the significant impact of the various nuclear cascade and evaporation models of the different Monte Carlo codes on the emission yields. The correlation is illustrated in figure 1.9 for protons, as well as for carbon ions.

Figure 1.9. Depth-dose profiles (left) and photon production profiles (right) for pencil beams of protons (top) and carbon ions (bottom) of different energies (indicated in MeV/u) incident along the axis of a cylindrical water target (40 cm axis and 20 cm diameter) divided in 200 bins of 2 mm thickness. All data were simulated with Monte Carlo code PHITS using the QMD nuclear cascade model and the GEM nuclear evaporation model. Edited from Stichelbaut (2008).

Thanks to a first scanning system, Min et al. (2006) confirmed this correlation experimentally by measuring prompt gammas emitted at 90° with respect to the proton beam direction. Since then, different concepts of cameras have been inves- tigated to perform prompt gamma imaging (PGI). Classical gamma cameras used in nuclear medicine are not adapted for the detection of high energy gammas in the presence of an important neutron background, so dedicated cameras are needed.

(27)

1.2. SOLUTIONS UNDER INVESTIGATION Lee et al. (2011) studied possible correlations with Geant4 simulations and con- cluded that detection systems should focus on gammas between 3 and 10 MeV.

As first measurements were all accomplished with homogeneous targets, Min et al. (2010) studied the prompt gamma distribution emitted by a non-homogeneous human phantom model in MCNPX simulations. They concluded that the emission profile was still correlated with the range despite the heterogeneous composition and complicated shape of the phantom. Other research efforts will be developed in chapter 2 and this technique will be further investigated in the present thesis.

Proton vertex imaging

Interaction vertex imaging (IVI) is investigated by the group of Lyon (Testa et al. 2012, Henriquet et al. 2012) and aims at measuring another type of prompt radiations from nuclear reactions in carbon ion therapy: secondary protons. Frag- mentation reactions can indeed result in the emission of protons with sufficient energy to leave the patient. Those protons are detected to reconstruct the posi- tion of the nuclear reaction (the so-calledvertex).

Advantages with respect to prompt photons are:

• the high proton emission yield with 12C (cf. section 7.3);

• the 100 % interaction probability in the detector when the proton is inter- cepted;

• the possibility to track the proton in successive layers of a detection system to determine its trajectory.

Difficulties specific to protons are:

• the numbers of protons that escape the patient is strongly dependent on the tumour depth;

• the low-energy protons emitted at the end of incident ion tracks are attenu- ated in the patient;

• the multiple Coulomb scattering of the protons in the patient.

Two techniques illustrated in figure 1.10 by Testa et al. (2012) were examined:

1. Single-proton imaging (SP-IVI): A single secondary proton is detected by a tracking detetector and the resulting trajectory is intersected with the incident ion trajectory that is measured thanks to a beam hodoscope.

2. Double-proton imaging (DP-IVI): Two protons resulting from the same nu- clear fragmentation reaction are detected simultaneously by two different tracking detectors and the resulting trajectories are intersected to deduce the emission point.

A feasibility study was conducted with Geant4 Monte Carlo simulations and

(28)

CHAPTER 1. RANGE UNCERTAINTY IN PARTICLE THERAPY

position of the tracking detectors at 20°with respect to the beam axis is the result of a compromise. Proton emissions are highly forward-peaked, so the statistic is better for small angles. Furthermore, the protons emitted at these small angles have a higher average energy and undergo less scattering. But the smaller the angle, the higher the uncertainty on the geometric determination of the vertex.

Figure 1.10. Single proton (SP) and double proton (DP) interaction vertex imaging (IVI) techniques. From Testa et al. (2012).

Secondary electron bremsstrahlung imaging

For carbon ion beams, an alternative to the measurement of prompt gammas, secondary protons or annihilation gammas has been investigated recently by Ya- maguchi et al. (2012): the measurement of bremsstrahlung. The generation of bremsstrahlung by secondary electrons is higher than gamma generation through nuclear processes and their emission distribution is correlated with the beam range.

Unlike gammas, the production yield of bremsstrahlung decreases along the carbon ion tracks. As bremsstrahlung photons have a continuous low energy spectrum, they cannot be easily distinguished from other background events.

During the irradiation of a cylindrical water-filled target (10 cm diameter) with 290 MeV/u carbon ions, Yamaguchi et al. (2012) obtained a correlated detection profile by scanning photons between 63 and 68 keV through the slit of a lead col- limator at 90° with a CdTe semiconductor detector. The experimental setup is reproduced in figure 1.11 and the detection profile in figure 1.12. The energy win- dow was selected to exclude a characteristic X-ray line of the lead collimator at 75 keV. The 0.5 mm thin semiconductor detector could detect these low energy photons with excellent energy resolution while being quite transparent to high en- ergy gammas. The number of photon counts in the 63 and 68 keV window linearly decreased along the ion track and the negative slope of this linear interpolation abruptly changed to a less negative value beyond the ion penetration depth, so that the penetration depth could be retrieved from the detection profile of figure 1.12.

(29)

1.2. SOLUTIONS UNDER INVESTIGATION These interesting preliminary results demand further investigation. One limita- tion of this method is that it would most probably not be adapted for deep-seated tumours as low energy bremsstrahlung photons from secondary electrons would be absorbed in the patient. During these first measurements, the transmission rate of the 60 keV photons through the 5 cm radius of the water target was 37 %.

Figure 1.11. Experimental setup for the validation of the range esti- mation method with bremsstrahlung. From Yamaguchi et al. (2012).

Figure 1.12. Detection profile of events between 63 and 68 keV along

(30)

CHAPTER 1. RANGE UNCERTAINTY IN PARTICLE THERAPY

1.2.3 Post treatment verification methods

Post treatment verification methods only estimate the beam penetration depth after the treatment was delivered. This information is still very valuable for three reasons. First, if an under- or overshoot is detected after a treatment fraction, the reason can be investigated before the next fraction is delivered. In particular, pos- itive range verification after the first fraction is enough to discard the risk related to uncertainties in the conversion of CT numbers to proton stopping powers during treatment planning. Second, statistics on treatment outcomes are more reliable if we can distinguish treatments that were delivered as planned from treatments that were affected by unacceptable range errors. Third, the information obtained improves our knowledge of range uncertainties, which is capital to describe relevant scenarios in the robust planning approach.

Positron emission tomography

Since its proposal (Maccabee et al. 1969), the use of positron emission tomog- raphy (PET) for treatment verification is developed by different groups for proton (Parodi et al. 2007) and carbon ion therapy (Fiedler et al. 2010). β+ emitting iso- topes including 11C, 13N and 15O (T1/2 = 20 min, 10 min and 2 min, respectively) are produced along the beam path, so that positron emitters can be imaged along the ion tracks even without a radioactive probing beam (cf. section 1.2.1). Their distribution can be measured with a PET/CT scanner that can be either in beam, in room or offline. This technique already benefited from a lot of research effort and is probably the one for which the most accurate conclusions can already be drawn. It was already used successfully with patients, but suffers some limitations.

Cross sections for the reactions of proton and carbon ion beams with abundant tissue nuclei are reproduced in table 1.2 from Fiedler et al. (2011a).

Table 1.2. Partial cross sections σ of the reactions of proton and carbon ion beams with the most abundant nuclei of human tissue.

Edited from Fiedler et al. (2011a).

(31)

1.2. SOLUTIONS UNDER INVESTIGATION The relation between induced activity and absorbed dose is complex. It ob- viously depends on the irradiation time, the time between irradiation and PET scanning and the PET scanning time itself. But it is also dependent on tissue composition and proton energy (Vynckier et al. 1993). The low induced activities require acquisition for several minutes after a usual treatment fraction. As a result, patient motion and biological wash-out of the radioactive nuclei in the organism blur the signal. We should also note that protons do not activate tissue along the last millimetres of their path as their energy is then smaller than the thresholds of the corresponding nuclear reactions.

Interpretation of the results is not straightforward as differences between mea- sured and simulated activity distributions can be explained by uncertainties in the PET verification as well as by errors in the treatment delivery (Knopf et al. 2009).

A practical method to estimate the beam range in the patient is to simulate the treatment plan to obtain the Monte Carlo calculated dose distribution and the Monte Carlo expected PET signal. These distributions are compared with the treatment plan dose distribution and the measured PET respectively and the ob- served shifts are added to deduce an upper bound on the range deviation, as described in figure 1.13 from Knopf et al. (2008). Unfortunately, it is very difficult to check the range when two opposed fields are applied.

Figure 1.13. Use of PET acquisition and Monte Carlo methods to check the range. From Knopf et al. (2008).

Different configurations are possible. The acquisition can be realised in-room with a PET scanner in the treatment room, or even online during treatment de- livery with a specific PET system, or simply offline, in which case the patient is quickly transported to a PET facility close to the treatment facility after his treat-

(32)

CHAPTER 1. RANGE UNCERTAINTY IN PARTICLE THERAPY

motion (Knopf et al. 2008), but they reduce patient throughput in the treatment room. Online imaging systems are also exposed to prompt gammas, neutrons and scattered radiations.

Studies conducted with offline PET/CT after proton therapy at MGH in Boston found that the spatial reproducibility of the measured activity distributions can be as good as 1 mm (Knopf et al. 2009). But unfortunately, range verification with an accuracy of 1 to 2 mm is only feasible for a few particular positions and tumour sites, namely low-perfused bony structures of the head and the neck pa- tients for which we can accurately co-register the planning CT and the PET/CT.

In table 1.3, Knopf et al. (2011) listed and valuated the challenges of the method for different tumour sites to emphasize the dependence of the achievable accuracy on the tumour location. They concluded accordingly that the offline approach is valuable for intracranial and cervical spine tumours, but faces severe limitations for abdominopelvic tumours.

Table 1.3. Tumour site-specific weighting factors for different chal- lenges of the PET/CT range verification method. A value of 1 indi- catesthis site is only slightly influenced by this factor; 2 means this site is influenced by this factor, but it is possible to minimize the influ- ence, and 3 fears this site is highly influenced by this factor and the success of future methodological improvements is questionable. Edited from Knopf et al. (2011).

At GSI in Darmstadt, a dedicated in-beam PET system was integrated in the treatment unit to perform in-beam acquisitions during carbon ion therapy (Fiedler et al. 2011a). The scanner is composed of two heads rather than a full ring not to limit patient positioning and handling. For more than 440 patients, acquisitions were started with the irradiation and stopped 40 s after the ion beam. This gave the opportunity to verify the field position in 3D after delivery of each fraction.

For treatments composed of different fields, only the first delivered field could be reliably imaged online, as images of the next fields are corrupted by the remaining activity of the first field. The order of the fields was thus changed on a day to day basis to acquire all the fields during the treatment.

(33)

1.2. SOLUTIONS UNDER INVESTIGATION The development of time-of-flight (TOF) PET scanners is expected to further improve reconstructed data in the future (Crespo et al. 2007). This technique aims at measuring the time difference between the two co-linear gamma rays detected to compute the location of the annihilation event along the line-of-response. New high density inorganic scintillators with very fast decay times have been developed and improve the time resolution of coincidence detectors. This opens perspectives for a TOF PET with a few hundreds of picoseconds time resolution. Using the TOF information would help limit the region of interest and improve the signal to noise ratio so that data treatment would be drastically reduced and real-time images might even be available during irradiation.

Magnetic resonance imaging of the vertebral column

Krejcarek et al. (2007) proposed a very specific post treatment range verifi- cation technique for craniospinal irradiation (CSI, a whole brain and spinal cord radiation therapy). After irradiation, the vertebral bone marrow is replaced by fatty marrow and this replacement is visible on magnetic resonance imaging (MRI).

Certain tumours of the central nervous system (CNS) present an important risk of subarachnoid dissemination and metastases along the neuraxis. Craniospinal ir- radiation is prescribed to reduce this risk and proton beams give the opportunity to limit the radiation dose to the bowel, lungs and heart. Two different approaches are adopted depending on the age of the patient. For adult and adolescent patients who have reached maximal growth, full dose is only prescribed to the thecal sac (surrounding the spinal cord and filled with cerebral spinal fluid) and exiting nerve roots. But for younger patients, the entire vertebral body has to be exposed to the prescription dose to avoid a future asymmetric growth of the vertebrae that would result in an abnormal curvature of the vertebral column.

Krejcarek et al. (2007) demonstrated that the fatty replacement of bone mar- row resulting from proton therapy can be seen on T1-weighted MRIs after the conclusion of treatment. Gensheimer et al. (2010) studied 10 patients who had completed vertebral growth and thus received a dose distribution planned to fall off in the vertebrae. Two or three posteroanterior fields were delivered in 1.8 to 2.0 Gy fractions for a total spine dose ranging from 23.4 to 50.4 Gy(RBE) (weighted by the relative biological effectiveness of 1.1). Using spine MRIs, they investigated proton range errors and found an average overpenetration in 8 patients and an average underpenetration in 2 patients. Range errors were on the order of a few millimetres. Images of three patients are reproduced in figure 1.14.

For the specific case of craniospinal irradiation, this method gives the possibility to verify that the spinal cord receives the planned dose and that surrounding

(34)

CHAPTER 1. RANGE UNCERTAINTY IN PARTICLE THERAPY

treatment is delivered. But a potential solution is investigated by Gensheimer et al. (2010). It aims at verifying the proton range from bone marrow oedema as it was already detected by other researchers using short tau inversion recovery (STIR, a fat suppression technique) MRI only a few days after initiation of photon radiation therapy.

Figure 1.14. Comparison of planned (blue) and observed (red) 50 % isodose lines in lumbar spine MR images after craniospinal irradiation.

Patient 5 (left) has minimal range errors, patient 6 (centre) seems to have a generalized beam overpenetration, and patient 10 (right) may have underpenetration in some areas. Edited from Gensheimer et al.

(2010).

(35)

1.3. CONCLUSION

1.3 Conclusion

As introduced in this chapter, range control with millimetre accuracy remains a challenge to get full benefit of proton and carbon ion therapies. A lot of effort is currently being invested to come up with a satisfactory solution and many ideas emerge. Very different techniques are investigated with different domains of ap- plications depending on the nature of the ion beam (protons, stable or radioactive carbon ions), the delivery mode (passive scattering or active scanning), the beam time structure (continuous or pulsed beam), the tumour depth, the type of tissue, the presence of a natural cavity, etc. Certain methods already prove useful, but exhibit limitations. Others have just been proposed. A ranking is thus not yet possible. Considering advantages and disadvantages of the different options, we should probably not expect one single method to set up for all cases. Different types of solutions will likely prove valuable for different ion beam therapy systems and different treatment sites.

In the present thesis, we further investigate the promising prompt gamma imag- ing technique. Major expected advantages of this technique with respect to other options are:

• The opportunity to perform real-time monitoring. With respect to the es- tablished PET technique, we would not have to wait for positron emitters to decay.

• The compatibility with pencil beam scanning mode. Actively scanned beams are considered as the future of particle therapy as they allow better confor- mality, reduced patient specific hardware and reduced neutron dose from the nozzle. The prompt gamma imaging technique opens perspectives to image pencil beams on an individual basis. Prompt gammas are emitted so quickly that their fluence is not corrupted by the activity produced by the first fields.

• The compatibility with all treatment sites. Prompt gamma imaging is thought to be a solution for complicated sites. The high energy gammas can escape the patient, even from deap-seated target volumes. As prompt gammas are emitted within less than 1 ns, their fluence is not affected by washout effects in well-perfused soft tissues.

(36)

CHAPTER 1. RANGE UNCERTAINTY IN PARTICLE THERAPY

(37)

Chapter 2

State of the art in prompt gamma imaging for proton and carbon

ion therapy

In this second chapter, we discuss the research efforts in prompt gamma imag- ing. We only focus on projects that aim at determining the range of proton and carbon ion beams and we therefore omit such techniques as the well-established prompt-gamma neutron activation analysis. We first list the design options and discuss the design objectives. We then review the different concepts that have been reported in the literature. They will be divided in two categories: collimated and non-collimated cameras. These two categories are also referred to as passively and electronically collimated cameras in the literature. Prompt gamma imaging has emerged as a hot topic in particle therapy and most of the results reported hereafter were published in the course of the present thesis.

(38)

CHAPTER 2. STATE OF THE ART IN PROMPT GAMMA IMAGING

2.1 Design options

Our objective is here to detect prompt gammas emitted along the proton (or carbon ion) tracks to measure the penetration depth of the beam inside the patient.

A prompt gamma camera is placed next to the patient, detects the prompt gammas that escape the patient and associates them with their emission points along the proton beam tracks. As the incident proton beam axis is a priori well known, the camera consequently needs two essential pieces of information to reconstruct the emission point when a prompt gamma is detected:

1. The interaction coordinates: The camera needs to know where the prompt gamma is detected, i.e. the position of the interaction that results in the detection of the prompt gamma. Gamma cameras usually use high density scintillation crystals to interact with the gammas. Continuous as well as pixellated crystals can be used in combination with pixellated photodetec- tors to collect the scintillation light resulting from the energy deposition of the incident gamma. With a continuous crystal, the position of interaction is estimated from the signal of the different photodetectors by a centre of gravity, also called centroid, method. With a pixellated crystal, events de- tected in a particular pixel are simply attributed to this pixel and the spatial resolution is directly related to the size of the pixels.

2. The incident direction: The camera needs to know the direction of the prompt gamma before it was detected. When detecting charged particles, this is not complicated: the charged particle is tracked through various lay- ers of position sensitive detectors where it continuously deposits energy and the interaction coordinates in the various layers reproduce the incident di- rection if the particle was not significantly scattered. But neutral particles as prompt gammas only undergo discrete interactions that result in absorp- tion or scattering. Two solutions are consequently available. First, we can use a collimator. A thick absorber with a small aperture is placed between the patient and the camera and we assume that prompt gammas that are detected in the camera could only pass through the aperture to reach the interaction coordinate in the camera. Second, we can detect prompt gammas in a specific detection system by causing a Compton scattering in a first layer of the detector, and then by detecting the scattered photon and/or the elec- tron set in motion in additional layers of the detection system. We can then use balance equations to deduce potential incident directions of the prompt gamma.

Thanks to these two pieces of information, the camera starts from the interaction coordinates, projects the photon flight backward and intersects it with the proton beam axis to deduce the emission point. The image of these emission points is then associated with the proton tracks, taking into account that prompt gamma emis- sions increase at the end of the proton tracks and stop below the energy threshold of associated nuclear reactions, between 5 and 30 MeV for 16O and 12C as we can observe from data of Kozlovsky et al. (2002) (cf. figure 1.8).

(39)

2.1. DESIGN OPTIONS One major difficulty was passed over in silence so far: not only photons corre- lated with the range escape the patient and reach the camera. The photon fluence is also composed of uncorrelated photons that were scattered in the patient, or emitted by secondary particles inside the patient or elsewhere in the treatment room. An important background of uncorrelated neutrons is also incident on the camera. Discrimination methods are therefore required to select correlated events and discard the uninteresting ones. Five different strategies can be combined:

1. Filter material: We can cause a differential absorption of photons and neu- trons by using a filter material with specific physical properties. At the scintillator level, the objective is to maximize photon detection with respect to neutron detection. We will thus favour high-density inorganic scintilla- tors, and avoid elements like H, 6Li, 10B and 157Gd that have high neutron cross sections. At the collimator level, we also favour high-density and high-Z materials to absorb photons, but two strategies are possible with respect to neutrons. We can either choose a material transparent or sensitive to neu- trons. A collimator that interacts with neutrons reduces the neutron fluence on the detector, but usually results in the emissions of secondary particles that have a higher probability of being detected if they reach the detector thanks to a lower energy.

2. Energy discrimination: Neutrons leaving the patient have a wide continuous energy spectrum, from the incident proton beam energy down to thermal en- ergy. Photons have a more reduced spectrum. In particular, prompt gammas correlated with the range escape the patient with energies mostly between 1 and 8 MeV with a few characteristic rays of the excited nuclei produced in the target. Energy discrimination has been applied for all camera prototypes proposed so far and is essential to isolate the prompt gamma contribution from the dominant background of uncorrelated events. Different energy win- dows have been selected for different camera prototypes in the literature, with lower thresholds as low as 2 MeV and upper limits up to 10 MeV. Discard- ing events below 2 MeV eliminates low energy neutron and photon events, including prompt gammas that lost their correlation after being scattered, and cutting above 10 MeV suppresses high energy neutron detections.

3. Angular selection: Nuclear collisions along the proton tracks result in the isotropic emission of prompt gammas. But the neutron distribution is forward- peaked along the beam axis. One can thus improve the photon to neutron fluence ratio on the camera by choosing backward emission angles. For a same thickness of tissue to escape the patient after emission along the pro- ton tracks, and for a same solid angle covered by the camera, more neutrons are detected at a 60° angle with respect to beam axis, than at 90° and at 120°, while photon detection is the same. But focusing on photons emitted at 120° causes other difficulties and all camera prototypes developed so far

(40)

CHAPTER 2. STATE OF THE ART IN PROMPT GAMMA IMAGING

the ones emitted just before the end of the proton tracks, are usually not the ones seen by the camera with the largest solid angle, nor the ones who reach the camera after passing through the least tissue. Furthermore, the error on the determination of the gamma emission point by backward projection of its trajectory is reduced when considering photons around 90°.

4. Time-of-flight discrimination: Photons travel from their emission point to the camera at the speed of light: they cover a 30 cm distance in only 1.0 ns.

Neutrons are slower: they cover the same 30 cm distance in 2.3 ns at 100 MeV, 6.9 ns at 10 MeV and 21.7 ns at 1 MeV. Neutrons are thus detected after the prompt gammas. Using a scintillation crystal with short rise and decay times and adequate electronics, we can select events that are detected within a spe- cific time window corresponding to photons only. This requires the camera to be placed far enough from the proton beam axis so that the photons can take sufficient advance on the neutrons with respect to camera time resolution.

Also, information on the time structure of the proton beam is necessary to determine when the incident protons hit the patient, and so when the prompt gammas should reach the camera. One can either use the RF signal from the cyclotron, directly related to the proton bunches of a pulsed beam, or a specific detector crossed by the protons just before they hit the patient. In practice, this method is efficient but not perfect: the proton beam delivered by an IBA C230 isochronous cyclotron, for example, is pulsed with a period of 10 ns, so neutrons emitted by a given proton bunch can reach the camera within the selection window associated with later proton bunches.

5. Pulse shape discrimination: The shape of the pulse produced by a given scintillator-photodetector chain depends on the nature of the particle re- sponsible for the energy deposition event. When they interact in the crystal, different types of particles have different ionization densities and excite the different types of luminescence centres with different decay times in differ- ent proportions. The fraction of light in the slow component consequently depends on the nature of the detected particle. By using a crystal with fast and slow decaying components, one can thus distinguish the events due to photons from the ones due to neutrons by performing pulse shape analysis.

Experimental results applying these methods for carbon ion beams were reported by Testa et al. (2010b) and Le Foulher et al. (2010b).

(41)

2.2. DESIGN OBJECTIVES

2.2 Design objectives

The design of a prompt gamma camera is a multi-criteria optimization problem for which no unique optimal solution is expected to be found. In the present section, we list the different objectives and indicate our own priorities. Basically, a trade-off has to be found between:

1. Accuracy.

2. Cost.

So far, no system has experimentally demonstrated sufficient accuracy for a patient in clinical treatment conditions, so the cost objective was not yet a discriminating factor. In fact, no system was ready to be tested in such conditions yet.

For a given number of incident protons of a spot of a treatment plan, the accuracy of the camera can be characterized as the standard deviation of the range estimation with respect to the exact range. Two standard deviations can be considered for a 95 % confidence interval. The accuracy is directly related to the quantity of data collected and the quality of these data. A compromise is necessary to reach the optimal accuracy. Three contradictory qualities are desired:

1. High counting statistics.

2. High correlated to uncorrelated events ratio.

3. High spatial resolution.

Mandatory specifications need to be satisfied:

1. At a 95 % confidence level, the accuracy must be better than 5 mm for most distal spots in pencil beam scanning mode. In practice, a camera would prove extremely valuable for an accuracy better than 2 mm at this confidence level, corresponding to a 1 mm standard deviation.

2. The camera’s footprint must be compatible with the treatment room and the patient positioning system.

3. The camera should be fast enough not to significantly delay treatments de- livered in pencil beam scanning mode.

Additional nice to have options include:

1. Measurement of the beam entrance point in the patient.

2. Reconstruction of a 2 or 3-dimensional image of the beam.

3. Absolute dose verification in 3 dimensions.

The latter still appears extremely challenging as dose deposition by the proton beam is related to electronic processes, not nuclear collisions.

Références

Documents relatifs

100 ps rms was obtained on the PG-ToF with 65 MeV proton beams using such a system. The simulations undertaken, accounting for such a resolution with the same detection setup, led to

High temporal resolution, of the order of 100 ps, improves the precision of ion range deter- mination using a Compton camera and it is envisaged that the CLaRyS Compton camera will

Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, 227(1), 91–96. Comparison of n-gamma discrimination

The source position should ideally be reconstructed from events (called true events thereafter) corresponding to one Compton scattering in each scatter detector without energy

A Low Noise and High Dynamic Range CMOS Integrated Electronics associated with Double Sided Silicon Strip Detectors for a Compton Camera gamma-ray Detecting System... A Low Noise

The effect of temporal resolution on ion range measurement based on the line cone reconstruction method was investigated via Monte Carlo simulation and a method to increase its

OE (right). Both reconstruction methods allowed for a cor- rect identification of zero shifts with a standard deviation σ smaller than 0 .9 mm for almost all scenarios, excepting

Les formes, les conditions et les délais dans lesquels doivent être faites ces déclarations sont fixés par arrêté du Ministre de la Santé Publique.