Technical Note: Experimental carbon ion range verification in inhomogeneous phantoms using prompt gammas M. Pinto IPNL, Université de Lyon, Lyon F-69003, France; Université Lyon 1, Villeurbanne F-69622, France; and CNRS/IN2P3, UMR 5822, Villeurbanne F-69622, France
M. De Rydt IPNL, Université de Lyon, Lyon F-69003, France; Université Lyon 1, Villeurbanne F-69622, France; CNRS/IN2P3, UMR 5822, Villeurbanne F-69622, France; and Instituut voor Kern- en Stralingsfysica, KU Leuven, Celestijnenlaan 200D, Leuven B-3001, Belgium
D. Dauvergne and G. Dedes IPNL, Université de Lyon, Lyon F-69003, France; Université Lyon 1, Villeurbanne F-69622, France; and CNRS/IN2P3, UMR 5822, Villeurbanne F-69622, France
N. Freud CREATIS, Université de Lyon, Lyon F-69003, France; Université Lyon 1, Villeurbanne F-69622, France; and CNRS UMR 5220, INSERM U1044, INSA-Lyon, Centre Léon Bérard, 69008 Lyon, France
J. Krimmer IPNL, Université de Lyon, Lyon F-69003, France; Université Lyon 1, Villeurbanne F-69622, France; and CNRS/IN2P3, UMR 5822, Villeurbanne F-69622, France
J. M. Létang CREATIS, Université de Lyon, Lyon F-69003, France; Université Lyon 1, Villeurbanne F-69622, France; and CNRS UMR 5220, INSERM U1044, INSA-Lyon, Centre Léon Bérard, 69008 Lyon, France
C. Ray, E. Testa,a) and M. Testa IPNL, Université de Lyon, Lyon F-69003, France; Université Lyon 1, Villeurbanne F-69622, France; and CNRS/IN2P3, UMR 5822, Villeurbanne F-69622, France
(Received 16 June 2014; revised 23 March 2015; accepted for publication 26 March 2015; published 15 April 2015) Purpose: The purpose of this study was to experimentally assess the possibility to monitor carbon ion range variations—due to tumor shift and/or elongation or shrinking—using prompt-gamma (PG) emission with inhomogeneous phantoms. Such a study is related to the development of PG monitoring techniques to be used in a carbon ion therapy context. Methods: A 95 MeV/u carbon ion beam was used to irradiate phantoms with a variable density along the ion path to mimic the presence of bone and lung in homogeneous humanlike tissue. PG profiles were obtained after a longitudinal scan of the phantoms. A setup comprising a narrow single-slit collimator and two detectors placed at 90◦ with respect to the beam axis was used. The time of flight technique was applied to allow the selection between PG and background events. Results: Using the positions at 50% entrance and 50% falloff of the PG profiles, a quantity called prompt-gamma profile length (PGPL) is defined. It is possible to observe shifts in the PGPL when there are absolute ion range shifts as small as 1–2 mm. Quantitatively, for an ion range shift of −1.33 ± 0.46 mm (insertion of a Teflon slab), a PGPL difference of −1.93 ± 0.58 mm and −1.84 ± 1.27 mm is obtained using a BaF2 and a NaI(Tl) detector, respectively. In turn, when an ion range shift of 4.59 ± 0.42 mm (insertion of a lung-equivalent material slab) is considered, the difference is of 4.10 ± 0.54 and 4.39 ± 0.80 mm for the same detectors. Conclusions: Herein, experimental evidence of the usefulness of employing PG to monitor carbon ion range using inhomogeneous phantoms is presented. Considering the homogeneous phantom as reference, the results show that the information provided by the PG emission allows for detecting ion range shifts as small as 1–2 mm. When considering the expected PG emission from an energy slice in a carbon ion therapy scenario, the experimental setup would allow to retrieve the same PGPL as the high statistics of the full experimental dataset in 58% of the times. However, this success rate increases to 93% when using a better optimized setup by means of Monte Carlo simulations. C 2015 American Association of Physicists in Medicine. [http://dx.doi.org/10.1118/1.4917225] Key words: prompt gammas, particle therapy monitoring, inhomogeneous targets, carbon ion therapy, 4
2342
Med. Phys. 42 (5), May 2015
0094-2405/2015/42(5)/2342/5/$30.00
© 2015 Am. Assoc. Phys. Med.
2342
2343
Pinto et al.: Experimental carbon ion range verification in inhomogeneous targets
1. INTRODUCTION Particle therapy is an innovative and challenging technique in the field of radiation therapy. The use of protons and other ions, namely, 12C nuclei, allows for a precise irradiation of the tumor volume and a better sparing of healthy tissues. However, this ion property makes the control of the Bragg-peak position and its conformation to the tumor volume desirable in order to improve the quality assurance of clinical particle therapy. This work presents an experimental study of ion range shifts generated by target-density variations close to the Bragg peak on the prompt-gamma (PG) depth profile. The study was performed for carbon ions using strongly collimated detectors and density variations mimicking the presence of bone and lung in homogeneous humanlike tissue. A proof of principle of this method was published by Min et al.1 and Testa et al.2 for protons and carbon ions, respectively. In both cases, a homogeneous phantom, either water or polymethylmethacrylate (PMMA), was used. Therefore, one of the remaining questions addresses the validity of verification of ion range shifts using the PG detection technique for phantoms with a variable density and composition, for which the present work is focused on, showing its feasibility. 2. MATERIAL AND METHODS The experiment was performed at the GANIL facility (Caen, France) using a 12C beam of 95 MeV/u. Three phantoms were irradiated: a homogeneous PMMA phantom (ρ = 1.168 ± 0.018 g/cm3) consisting of 27 adjacent slices of 50 × 50 × 2.1 mm3 each, a PMMA phantom in which the sixth slice was replaced by a Teflon piece with a thickness of 2.0 mm (ρ = 2.150 ± 0.153 g/cm3), and a PMMA phantom with a 5.6-mm-thick section of lung-equivalent tissue (ρ = 0.207 ± 0.005 g/cm3, CIRS, Inc., Norfolk, VA—lung inhale) at the position of the fourth PMMA slice. The average density values were determined by measuring the dimensions and the mass of the slices, while the uncertainties were derived from the uncertainties associated to the measuring devices.
F. 1. Schematic representation of the experimental setup (top view) in the case of a phantom with an insert (red slice). Note the placement of several lead bricks in order to shield scintillators from unwanted events. Medical Physics, Vol. 42, No. 5, May 2015
2343
The setup comprised a single-slit collimator and two detectors positioned at 90◦ with respect to the beam direction and at 60.5 cm from the beam axis (see a schematic overview in Fig. 1). The lead collimator, consisting of two parts with an opening of 2 mm in between, ensures a proper selection of the secondary radiation, strongly limiting the accepted solid angle and allowing only events that originate from interaction points within a limited geometrical field of view (FOV) in front of the opening (5.2 mm). In order to obtain a PG profile as a function of depth, the phantoms are translated along the beam direction using a moving table and the secondary radiation is recorded in discrete steps. It should be noted that the step length was variable. The detection of events is performed with a hexagonal BaF2 detector (50 mm edge, 140 mm long) and a cylindrical NaI(Tl) detector (76 mm diameter, 76 mm long), placed on top of each other behind the collimator slit and providing two independent registrations of the PG measurements. The interface between the lower BaF2 and the upper NaI(Tl) detector is positioned at the same height as the center of the phantom. A smaller NaI(Tl) detector is installed 165 cm below the phantom center to register secondary radiation induced in the whole phantom (not shown in Fig. 1). Once calibrated by means of a Faraday cup, this scintillator allows to quantify the number of primary particles, assuming a linear relationship between the latter and the detected radiation yield. The calibration took also into consideration the effect of different phantom positions. The circular beam FWHM spot size was found to be approximately 5 mm at the phantom position and the beam was centered with the phantom cross section, thus ensuring that the entire beam was impinging on the phantoms. The detectors are plugged to NIM modules and conventional electronics with a VME-based acquisition system, which was triggered by the OR logical signal of the detectors. To distinguish PG events from the extensive background originated mainly from neutron-associated events, time of flight (TOF) discrimination is used. This implies an event-byevent registration of the time difference between the detected secondary radiation and the delayed signal of the incoming ion given by the high-frequency pulse of the cyclotron (∼12.5 MHz). In the TOF domain, events are accepted when they appear in a time window of 2.67 and 3.44 ns centered at the PG peak, respectively, in the BaF2 and NaI(Tl) spectra. These time windows were chosen in order to include all distinguishable events related to PG emission and they correspond to, at least, approximately ±2σ of the PG peak. In terms of energy selection, lower limits of 1000 keV and 600 keV are imposed on the BaF2 and NaI(Tl) spectra, respectively, a compromise that includes as many valid data as possible and avoids interference with the hardware threshold and the 511keV annihilation photons. The energy resolution ∆E/E with a Cs-137 source (662 keV) was estimated to be around 22.2% and 9.2% for the BaF2 and NaI(Tl) detectors, respectively. Nonetheless, even considering these conditions, the background events in the PG profiles are still significant.2 Therefore, a background subtraction method was developed, which is based on the time information and makes use of a reference TOF spectrum. Such a spectrum is obtained from measurement
2344
Pinto et al.: Experimental carbon ion range verification in inhomogeneous targets
positions assumed to not register PG and a subtraction is made between the intended spectrum and the reference one. The registered background events (mainly neutron-associated events) increase almost linearly along the beam axis since neutron emission is not isotropic and favors forward angles.2,3 Being so, a scaling factor is applied to the reference spectrum before the subtraction. This scaling factor corresponds to the ratio of the integrals outside the time window between the two TOF spectra involved in the subtraction. Therefore, it is assumed that the background in the PG peak region increases by the same factor as outside this region and that the shape of the several TOF spectra used to build a profile remains the same. The reference TOF spectrum was always the one corresponding to the first measurement point in the profile. In order to subtract any remaining events after the TOF-based background subtraction, a linear fit using the points before the target and after the PG profile is applied and the function retrieved is used to subtract the points in the profile. As the experimental ion range cannot be directly extracted from the PG profile, the prompt-gamma profile length (PGPL), defined as the distance between the positions marking 50% of the PG profile rise and 50% of the falloff, is used instead. The goal of this quantity is to have an absolute measure of the ion range. If only the PG profile falloff is used, it is not possible to attribute a detected deviation to an ion range shift since the patient may have been mispositioned. The determination of the 50% positions relies on the fit of the four-parameter sigmoid function first proposed by our collaboration for ion range retrieval in interaction vertex imaging4 and then adapted by Janssen et al.5 to find correlations in simulated data between proton range and the PG profile falloff. The position of the inflection point corresponds thereby to the 50% position. In order to ensure a consistent fitting procedure for the three profiles, it is assumed that no background remains after subtraction. The difference between the 50% positions
2344
obtained by the fit to the PG profiles entrance and falloff is defined as PGPL. The ion range was estimated based on a procedure with 4 (Ref. 6) simulations. The density value for each material and the associated uncertainty were used to randomly assign density values to the simulated materials following a Gaussian distribution. Several hundreds of simulations were performed, and for each simulation, the mean projected ion range was retrieved and used to fill a histogram. From this histogram, the mean and the standard deviation were calculated and used as ion range and its uncertainty. Ideally, the different ion ranges should be experimentally estimated using a plane-parallel ionization chamber in water. However, this was not possible due to the nature of the facility where this experiment was conducted.
3. RESULTS The zero position corresponds to a first online estimate of the phantom entrance centered in front of the collimator opening. During the off-line analysis, a shift of 1.5-mm was found in the PMMA-lung phantom (based on the position of the profile entrance rise) and corrected. Note that only statistical errors are shown to preserve the figure quality. The systematic errors on each data point are of the order of 20% and originate from uncertainties on the calibration of the small NaI(Tl) detector, used for the normalization to the number of primary ions. The results are shown in Fig. 2. The different PGPLs for the three phantoms are shown in Table I along with the variation of both ion range and PGPL in respect to the homogeneous phantom. Taking the PG profile obtained with the homogeneous PMMA phantom as a reference, it is possible to study the impact of the density changes on the profile lengths.
F. 2. Experimental PG profiles obtained for the PMMA, PMMA-Teflon, and PMMA-lung phantoms using the BaF2 detector with the PMMA-lung phantom position corrected with an offset of 1.5 mm. The NaI(Tl) data are not shown since they are similar to these ones. The thicker solid lines depict the sigmoid function fits to the data to estimate the PGPL. The rectangles represent the phantoms with or without inserts (shaded areas) and the dashed lines show the mean ion range for each phantom estimated with 4. Medical Physics, Vol. 42, No. 5, May 2015
2345
Pinto et al.: Experimental carbon ion range verification in inhomogeneous targets
2345
T I. Absolute and relative measurements of the PGPL of a 95 MeV/u 12C ion beam in the phantoms considered. The relative measurement is the variation with respect to the homogeneous PMMA phantom. Ion ranges are calculated (see text). Only statistical uncertainties are given. Absolute measurement
Relative measurement
Profile length (mm) Phantom PMMA PMMA-Teflon PMMA-lung
Ion range (mm) 20.67 ± 0.30 19.34 ± 0.35 25.26 ± 0.29
NaI(Tl) 20.77 ± 0.63 18.93 ± 1.10 25.16 ± 0.50
— −1.33 ± 0.46 4.59 ± 0.42
BaF2 21.16 ± 0.39 19.23 ± 0.43 25.26 ± 0.37
4. DISCUSSION When evaluating the PG profiles in Fig. 2, a good agreement between the three curves is obtained for the points at the phantom entrance. Furthermore, at positions that are not affected by the inserts, e.g., in the plateau and just before the falloff, similar PG yields are recorded. At the position of the inserts, clear variations are observed between the different longitudinal PG profiles. An increase in the PG yield is visible at the position of the Teflon while an explicit dip is present at the position of the lung-equivalent tissue. However, the quantification of the correlation of the PG emission with density and material composition is outside the scope of the present study and requires a different approach such as the one followed by Polf et al.7 Table I demonstrates a strong correlation between the ion range and the PGPL since their variations are the same within statistical uncertainties, turning the PGPL into an excellent probe for ion range studies. It should also be stressed the importance of doing this study with a relatively low-energy carbon ion beam since the ion range at this energy is around 2 cm in PMMA; thus in a clinical situation, such a distance is at least close to the tumor, probably even included in the planning target volume. Considering the possible systematic errors of these results, the PG monitoring with the collimated-camera technique may allow for detecting ion range variations as small as 1–2 mm situated at a distance less than 2 cm from the Bragg peak. Further supporting such a claim, for the conditions of the present study, was the observation off-line of a 1.5 mm shift on the PMMA-lung profile in respect to the other two profiles. Moreover, a 1 mm ion range shift was clearly visible on the registered PG profiles (see PMMA vs PMMA-Teflon phantoms in Fig. 2). However, these results are not the only outcome to consider from this study since the methods applied herein may constitute promising approaches for a clinical device. In this regard, both the background subtraction method and the PGPL determination may be good candidates for clinical application. The former is a relatively simple method that can be easily implemented in, e.g., a field-programmable gate array (FPGA) for online control. It requires only TOF information and a reference detector placed proximal to the patient (outside the detector FOV). A blind detector (i.e., a detector behind shielding) placed in the multislit collimator device may be suited as well, although further experimental work is needed to validate such an assumption. On the other hand, the PGPL quantity Medical Physics, Vol. 42, No. 5, May 2015
Profile length (mm) Ion range (mm)
BaF2 — −1.93 ± 0.58 4.10 ± 0.54
NaI(Tl) — −1.84 ± 1.27 4.39 ± 0.80
as described in this study provides a good correlation with ion range, thus being a good possible quantity for absolute ion range monitoring based on the information provided by a single device. One may also consider the use of optical and beam delivery systems to pinpoint the position where the beam is entering the patient but this requires the integration of at least three separate devices: optical, beam delivery, and PG camera systems. It should also be noted that better functions to retrieve the PGPL may be considered for the actual monitoring. This choice will inevitably depend on the use of background subtraction, the sampling of the collected data, and robustness of the correlation with the ion range. As an example, nonuniform rational basis spline (NURBS) functions are good candidates as well.8 These results and conclusions are obtained for a relatively simple scenario while a clinical one is much more complex. Moreover, the typical number of carbon ions delivered to the phantom for each measurement position is 1010 with a beam intensity of 2 × 108 ions s−1. As comparison, Krämer et al.9 state a total of 7 × 108 carbon ions in a target volume (around 120 cm3) to deliver an absorbed dose of 1 Gy with ∼10 000 raster positions and 39 energy slices. Based on these numbers, the monitoring on a raster-scanning basis seems unfeasible since less than one event per data point per raster spot is expected. Regarding monitoring at the energy slice scale, a procedure to estimate the PGPL retrieval at this statistics level was developed. Since the total number of ions per longitudinal position in the experiment is known, the experimental data are first rescaled to the statistics of an energy slice (i.e., 1.8 × 107 ions) and the data points are then randomly sampled with a Poisson distribution. Finally, the sigmoid function is used to fit the new profiles. This process is repeated hundreds of times with different seed numbers to gather enough statistical evidence. It was found that for 58% of the cases, the same PGPL (within ±2σ) was retrieved for the three phantoms with respect to the full experimental statistics. However, these conclusions are for the present configuration with a very challenging design (i.e., very fine collimation). It is therefore expected that an optimized device would yield a significantly improved performance. In fact, by using a design optimized to achieve a reasonable spatial resolution during PG monitoring with proton beams (configuration “case 3” in Ref. 10), this success rate increases to 93% using 4 simulations. Please note that one cannot use directly the results from simulations since they overestimate the experimental data. Therefore, the
2346
Pinto et al.: Experimental carbon ion range verification in inhomogeneous targets
experimental setup was simulated and the scaling factor that minimizes the root-mean-square error between experimental and simulated data was determined (0.42). The outcomes of the simulations were then scaled by this factor, and the method to compute the PGPL retrieval success rate at the energy slice scale was applied to them. One can argue that some of the profile features associated with the inhomogeneities may be lost with cameras having higher slit FOV; however, such cameras are more efficient. As already shown for PG emitted during proton irradiation,8 the retrieval precision of the PG profile falloff is inversely proportional to the signal-to-noise ratio of the PG profile; thus, a higher collection efficiency ultimately improves the ion range shift estimation. The only restriction to this improvement is the blurring of the profile, which prevents the retrieval of the profile characteristics. The estimation of such a limit for the slit FOV is outside the scope of the present work, but at least the optimized design used does not have a significant effect on the PGPL estimation, as it can be understood from the 93% success rate in having the same PGPL as with the full statistics of the experimental data (in this case, slit FOV is 13.1 mm, cf. Table 6 in Ref. 10). In addition, the uncertainty in the PGPL shifts with the statistics of an energy slice is of the same order of the typical clinical margins [i.e., 2–3 mm (Ref. 11)]. It should also be stressed that the configuration considered was optimized for proton therapy; therefore, it is not necessarily the most adequate to be used with carbon ions. Since it was a design already available in the literature, it was used to discuss the potential of the technique. Nevertheless, there is a remark concerning monitoring at the scale of an energy slice. In scenarios where the waterequivalent path length is similar for all the ions in an energy slice (e.g., treatment of a brain tumor), a single PG profile falloff is expected. On the other hand, for the cases where this condition is not met, the ions will have different ranges; thus, several falloffs will be present in the detected PG profile. Although the best clinical practice for such a case is not as yet clear, one can envisage three approaches: (1) the treatment planning system triggers the monitoring device to only detect PG events on a same-ion range basis; (2) more complex functions are used for the fitting, which may allow to separate the different PGPL; and (3) the monitoring system only considers the comparison between an expected distribution (e.g., Monte Carlo simulations) and the detected one, thus it does not estimate the PGPL for each case with different ranges (although one can always retrieve them from the simulations). Interesting enough is the fact that one can compare independently the different expected ion ranges of the treatment plan; hence, the detection of a potential ion range shift can be cross-checked in several beam sets (i.e., energy slice or ion range based). Finally, it should be noted that the methods for background subtraction and PGPL determination should also be applicable to other cases, notably to proton therapy.
Medical Physics, Vol. 42, No. 5, May 2015
2346
5. CONCLUSIONS Three PG profiles were measured using a 95 MeV/u 12C primary beam with a homogeneous PMMA phantom and two others with heterogeneities along the ion path resulting in ion range shifts. The longitudinal PG profiles were reconstructed using TOF to discriminate between PG and background, and a single-slit collimator was used to ensure a proper longitudinal position and angular selection. Measurements were performed with two independent detectors, resulting in a good mutual agreement. A correlation was observed between ion range shifts induced by the heterogeneities and the detected PG profiles. Ion range shifts as small as one to two millimeters can be detected in the last two centimeters before the Bragg peak as was shown for the different phantoms.
ACKNOWLEDGMENTS This work has been supported by the Rhône Alpes Program for Research in Hadrontherapy (PRRH, CPER 2007-13), the ENVISION European project (G.A. #241851), the ENTERVISION network (G.A. #264552), Labex PRIMES (ANR11-LABX-0063), and France Hadron (ANR-11-INBS-0007). M.D.R. acknowledges support from the FWO Flanders (Fonds voor Wetenschappelijk Onderzoek).
a)Author
to whom correspondence should be addressed. Electronic mail: [email protected] 1C.-H. Min et al., “Prompt gamma measurements for locating the dose falloff region in the proton therapy,” Appl. Phys. Lett. 89(18), 183517 (2006). 2E. Testa et al., “Monitoring the Bragg peak location of 73 MeV/u carbon ions by means of prompt γ-ray measurements,” Appl. Phys. Lett. 93(9), 093506 (2008). 3K. Gunzert-Marx et al., “Secondary beam fragments produced by 200 MeV 12 u−1 C ions in water and their dose contributions in carbon ion radiotherapy,” New J. Phys. 10(7), 075003 (2008). 4P. Henriquet et al., “Interaction vertex imaging (IVI) for carbon ion therapy monitoring: A feasibility study,” Phys. Med. Biol. 57(14), 4655–4669 (2012). 5F. M. F. C. Janssen et al., “Factors influencing the accuracy of beam range estimation in proton therapy using prompt gamma emission,” Phys. Med. Biol. 59(15), 4427–4441 (2014). 6S. Agostinelli et al., “ 4 - a simulation toolkit,” Nucl. Instrum. Methods Phys. Res., Sect. A 506(3), 250–303 (2003). 7J. C. Polf et al., “Measurement of characteristic prompt gamma rays emitted from oxygen and carbon in tissue-equivalent samples during proton beam irradiation,” Phys. Med. Biol. 58(17), 5821–5831 (2013). 8F. Roellinghoff et al., “Real-time proton beam range monitoring by means of prompt-gamma detection with a collimated camera,” Phys. Med. Biol. 59(5), 1327–1338 (2014). 9M. Krämer et al., “Treatment planning for heavy-ion radiotherapy: Physical beam model and dose optimization,” Phys. Med. Biol. 45(11), 3299–3317 (2000). 10M. Pinto et al., “Design optimisation of a TOF-based collimated camera prototype for online hadrontherapy monitoring,” Phys. Med. Biol. 59(24), 7653–7674 (2014). 11O. Jäkel et al., “Treatment planning for heavy ion radiotherapy: Clinical implementation and application,” Phys. Med. Biol. 46(4), 1101–1116 (2001).
M. De Rydt IPNL, Université de Lyon, Lyon F-69003, France; Université Lyon 1, Villeurbanne F-69622, France; CNRS/IN2P3, UMR 5822, Villeurbanne F-69622, France; and Instituut voor Kern- en Stralingsfysica, KU Leuven, Celestijnenlaan 200D, Leuven B-3001, Belgium
D. Dauvergne and G. Dedes IPNL, Université de Lyon, Lyon F-69003, France; Université Lyon 1, Villeurbanne F-69622, France; and CNRS/IN2P3, UMR 5822, Villeurbanne F-69622, France
N. Freud CREATIS, Université de Lyon, Lyon F-69003, France; Université Lyon 1, Villeurbanne F-69622, France; and CNRS UMR 5220, INSERM U1044, INSA-Lyon, Centre Léon Bérard, 69008 Lyon, France
J. Krimmer IPNL, Université de Lyon, Lyon F-69003, France; Université Lyon 1, Villeurbanne F-69622, France; and CNRS/IN2P3, UMR 5822, Villeurbanne F-69622, France
J. M. Létang CREATIS, Université de Lyon, Lyon F-69003, France; Université Lyon 1, Villeurbanne F-69622, France; and CNRS UMR 5220, INSERM U1044, INSA-Lyon, Centre Léon Bérard, 69008 Lyon, France
C. Ray, E. Testa,a) and M. Testa IPNL, Université de Lyon, Lyon F-69003, France; Université Lyon 1, Villeurbanne F-69622, France; and CNRS/IN2P3, UMR 5822, Villeurbanne F-69622, France
(Received 16 June 2014; revised 23 March 2015; accepted for publication 26 March 2015; published 15 April 2015) Purpose: The purpose of this study was to experimentally assess the possibility to monitor carbon ion range variations—due to tumor shift and/or elongation or shrinking—using prompt-gamma (PG) emission with inhomogeneous phantoms. Such a study is related to the development of PG monitoring techniques to be used in a carbon ion therapy context. Methods: A 95 MeV/u carbon ion beam was used to irradiate phantoms with a variable density along the ion path to mimic the presence of bone and lung in homogeneous humanlike tissue. PG profiles were obtained after a longitudinal scan of the phantoms. A setup comprising a narrow single-slit collimator and two detectors placed at 90◦ with respect to the beam axis was used. The time of flight technique was applied to allow the selection between PG and background events. Results: Using the positions at 50% entrance and 50% falloff of the PG profiles, a quantity called prompt-gamma profile length (PGPL) is defined. It is possible to observe shifts in the PGPL when there are absolute ion range shifts as small as 1–2 mm. Quantitatively, for an ion range shift of −1.33 ± 0.46 mm (insertion of a Teflon slab), a PGPL difference of −1.93 ± 0.58 mm and −1.84 ± 1.27 mm is obtained using a BaF2 and a NaI(Tl) detector, respectively. In turn, when an ion range shift of 4.59 ± 0.42 mm (insertion of a lung-equivalent material slab) is considered, the difference is of 4.10 ± 0.54 and 4.39 ± 0.80 mm for the same detectors. Conclusions: Herein, experimental evidence of the usefulness of employing PG to monitor carbon ion range using inhomogeneous phantoms is presented. Considering the homogeneous phantom as reference, the results show that the information provided by the PG emission allows for detecting ion range shifts as small as 1–2 mm. When considering the expected PG emission from an energy slice in a carbon ion therapy scenario, the experimental setup would allow to retrieve the same PGPL as the high statistics of the full experimental dataset in 58% of the times. However, this success rate increases to 93% when using a better optimized setup by means of Monte Carlo simulations. C 2015 American Association of Physicists in Medicine. [http://dx.doi.org/10.1118/1.4917225] Key words: prompt gammas, particle therapy monitoring, inhomogeneous targets, carbon ion therapy, 4
2342
Med. Phys. 42 (5), May 2015
0094-2405/2015/42(5)/2342/5/$30.00
© 2015 Am. Assoc. Phys. Med.
2342
2343
Pinto et al.: Experimental carbon ion range verification in inhomogeneous targets
1. INTRODUCTION Particle therapy is an innovative and challenging technique in the field of radiation therapy. The use of protons and other ions, namely, 12C nuclei, allows for a precise irradiation of the tumor volume and a better sparing of healthy tissues. However, this ion property makes the control of the Bragg-peak position and its conformation to the tumor volume desirable in order to improve the quality assurance of clinical particle therapy. This work presents an experimental study of ion range shifts generated by target-density variations close to the Bragg peak on the prompt-gamma (PG) depth profile. The study was performed for carbon ions using strongly collimated detectors and density variations mimicking the presence of bone and lung in homogeneous humanlike tissue. A proof of principle of this method was published by Min et al.1 and Testa et al.2 for protons and carbon ions, respectively. In both cases, a homogeneous phantom, either water or polymethylmethacrylate (PMMA), was used. Therefore, one of the remaining questions addresses the validity of verification of ion range shifts using the PG detection technique for phantoms with a variable density and composition, for which the present work is focused on, showing its feasibility. 2. MATERIAL AND METHODS The experiment was performed at the GANIL facility (Caen, France) using a 12C beam of 95 MeV/u. Three phantoms were irradiated: a homogeneous PMMA phantom (ρ = 1.168 ± 0.018 g/cm3) consisting of 27 adjacent slices of 50 × 50 × 2.1 mm3 each, a PMMA phantom in which the sixth slice was replaced by a Teflon piece with a thickness of 2.0 mm (ρ = 2.150 ± 0.153 g/cm3), and a PMMA phantom with a 5.6-mm-thick section of lung-equivalent tissue (ρ = 0.207 ± 0.005 g/cm3, CIRS, Inc., Norfolk, VA—lung inhale) at the position of the fourth PMMA slice. The average density values were determined by measuring the dimensions and the mass of the slices, while the uncertainties were derived from the uncertainties associated to the measuring devices.
F. 1. Schematic representation of the experimental setup (top view) in the case of a phantom with an insert (red slice). Note the placement of several lead bricks in order to shield scintillators from unwanted events. Medical Physics, Vol. 42, No. 5, May 2015
2343
The setup comprised a single-slit collimator and two detectors positioned at 90◦ with respect to the beam direction and at 60.5 cm from the beam axis (see a schematic overview in Fig. 1). The lead collimator, consisting of two parts with an opening of 2 mm in between, ensures a proper selection of the secondary radiation, strongly limiting the accepted solid angle and allowing only events that originate from interaction points within a limited geometrical field of view (FOV) in front of the opening (5.2 mm). In order to obtain a PG profile as a function of depth, the phantoms are translated along the beam direction using a moving table and the secondary radiation is recorded in discrete steps. It should be noted that the step length was variable. The detection of events is performed with a hexagonal BaF2 detector (50 mm edge, 140 mm long) and a cylindrical NaI(Tl) detector (76 mm diameter, 76 mm long), placed on top of each other behind the collimator slit and providing two independent registrations of the PG measurements. The interface between the lower BaF2 and the upper NaI(Tl) detector is positioned at the same height as the center of the phantom. A smaller NaI(Tl) detector is installed 165 cm below the phantom center to register secondary radiation induced in the whole phantom (not shown in Fig. 1). Once calibrated by means of a Faraday cup, this scintillator allows to quantify the number of primary particles, assuming a linear relationship between the latter and the detected radiation yield. The calibration took also into consideration the effect of different phantom positions. The circular beam FWHM spot size was found to be approximately 5 mm at the phantom position and the beam was centered with the phantom cross section, thus ensuring that the entire beam was impinging on the phantoms. The detectors are plugged to NIM modules and conventional electronics with a VME-based acquisition system, which was triggered by the OR logical signal of the detectors. To distinguish PG events from the extensive background originated mainly from neutron-associated events, time of flight (TOF) discrimination is used. This implies an event-byevent registration of the time difference between the detected secondary radiation and the delayed signal of the incoming ion given by the high-frequency pulse of the cyclotron (∼12.5 MHz). In the TOF domain, events are accepted when they appear in a time window of 2.67 and 3.44 ns centered at the PG peak, respectively, in the BaF2 and NaI(Tl) spectra. These time windows were chosen in order to include all distinguishable events related to PG emission and they correspond to, at least, approximately ±2σ of the PG peak. In terms of energy selection, lower limits of 1000 keV and 600 keV are imposed on the BaF2 and NaI(Tl) spectra, respectively, a compromise that includes as many valid data as possible and avoids interference with the hardware threshold and the 511keV annihilation photons. The energy resolution ∆E/E with a Cs-137 source (662 keV) was estimated to be around 22.2% and 9.2% for the BaF2 and NaI(Tl) detectors, respectively. Nonetheless, even considering these conditions, the background events in the PG profiles are still significant.2 Therefore, a background subtraction method was developed, which is based on the time information and makes use of a reference TOF spectrum. Such a spectrum is obtained from measurement
2344
Pinto et al.: Experimental carbon ion range verification in inhomogeneous targets
positions assumed to not register PG and a subtraction is made between the intended spectrum and the reference one. The registered background events (mainly neutron-associated events) increase almost linearly along the beam axis since neutron emission is not isotropic and favors forward angles.2,3 Being so, a scaling factor is applied to the reference spectrum before the subtraction. This scaling factor corresponds to the ratio of the integrals outside the time window between the two TOF spectra involved in the subtraction. Therefore, it is assumed that the background in the PG peak region increases by the same factor as outside this region and that the shape of the several TOF spectra used to build a profile remains the same. The reference TOF spectrum was always the one corresponding to the first measurement point in the profile. In order to subtract any remaining events after the TOF-based background subtraction, a linear fit using the points before the target and after the PG profile is applied and the function retrieved is used to subtract the points in the profile. As the experimental ion range cannot be directly extracted from the PG profile, the prompt-gamma profile length (PGPL), defined as the distance between the positions marking 50% of the PG profile rise and 50% of the falloff, is used instead. The goal of this quantity is to have an absolute measure of the ion range. If only the PG profile falloff is used, it is not possible to attribute a detected deviation to an ion range shift since the patient may have been mispositioned. The determination of the 50% positions relies on the fit of the four-parameter sigmoid function first proposed by our collaboration for ion range retrieval in interaction vertex imaging4 and then adapted by Janssen et al.5 to find correlations in simulated data between proton range and the PG profile falloff. The position of the inflection point corresponds thereby to the 50% position. In order to ensure a consistent fitting procedure for the three profiles, it is assumed that no background remains after subtraction. The difference between the 50% positions
2344
obtained by the fit to the PG profiles entrance and falloff is defined as PGPL. The ion range was estimated based on a procedure with 4 (Ref. 6) simulations. The density value for each material and the associated uncertainty were used to randomly assign density values to the simulated materials following a Gaussian distribution. Several hundreds of simulations were performed, and for each simulation, the mean projected ion range was retrieved and used to fill a histogram. From this histogram, the mean and the standard deviation were calculated and used as ion range and its uncertainty. Ideally, the different ion ranges should be experimentally estimated using a plane-parallel ionization chamber in water. However, this was not possible due to the nature of the facility where this experiment was conducted.
3. RESULTS The zero position corresponds to a first online estimate of the phantom entrance centered in front of the collimator opening. During the off-line analysis, a shift of 1.5-mm was found in the PMMA-lung phantom (based on the position of the profile entrance rise) and corrected. Note that only statistical errors are shown to preserve the figure quality. The systematic errors on each data point are of the order of 20% and originate from uncertainties on the calibration of the small NaI(Tl) detector, used for the normalization to the number of primary ions. The results are shown in Fig. 2. The different PGPLs for the three phantoms are shown in Table I along with the variation of both ion range and PGPL in respect to the homogeneous phantom. Taking the PG profile obtained with the homogeneous PMMA phantom as a reference, it is possible to study the impact of the density changes on the profile lengths.
F. 2. Experimental PG profiles obtained for the PMMA, PMMA-Teflon, and PMMA-lung phantoms using the BaF2 detector with the PMMA-lung phantom position corrected with an offset of 1.5 mm. The NaI(Tl) data are not shown since they are similar to these ones. The thicker solid lines depict the sigmoid function fits to the data to estimate the PGPL. The rectangles represent the phantoms with or without inserts (shaded areas) and the dashed lines show the mean ion range for each phantom estimated with 4. Medical Physics, Vol. 42, No. 5, May 2015
2345
Pinto et al.: Experimental carbon ion range verification in inhomogeneous targets
2345
T I. Absolute and relative measurements of the PGPL of a 95 MeV/u 12C ion beam in the phantoms considered. The relative measurement is the variation with respect to the homogeneous PMMA phantom. Ion ranges are calculated (see text). Only statistical uncertainties are given. Absolute measurement
Relative measurement
Profile length (mm) Phantom PMMA PMMA-Teflon PMMA-lung
Ion range (mm) 20.67 ± 0.30 19.34 ± 0.35 25.26 ± 0.29
NaI(Tl) 20.77 ± 0.63 18.93 ± 1.10 25.16 ± 0.50
— −1.33 ± 0.46 4.59 ± 0.42
BaF2 21.16 ± 0.39 19.23 ± 0.43 25.26 ± 0.37
4. DISCUSSION When evaluating the PG profiles in Fig. 2, a good agreement between the three curves is obtained for the points at the phantom entrance. Furthermore, at positions that are not affected by the inserts, e.g., in the plateau and just before the falloff, similar PG yields are recorded. At the position of the inserts, clear variations are observed between the different longitudinal PG profiles. An increase in the PG yield is visible at the position of the Teflon while an explicit dip is present at the position of the lung-equivalent tissue. However, the quantification of the correlation of the PG emission with density and material composition is outside the scope of the present study and requires a different approach such as the one followed by Polf et al.7 Table I demonstrates a strong correlation between the ion range and the PGPL since their variations are the same within statistical uncertainties, turning the PGPL into an excellent probe for ion range studies. It should also be stressed the importance of doing this study with a relatively low-energy carbon ion beam since the ion range at this energy is around 2 cm in PMMA; thus in a clinical situation, such a distance is at least close to the tumor, probably even included in the planning target volume. Considering the possible systematic errors of these results, the PG monitoring with the collimated-camera technique may allow for detecting ion range variations as small as 1–2 mm situated at a distance less than 2 cm from the Bragg peak. Further supporting such a claim, for the conditions of the present study, was the observation off-line of a 1.5 mm shift on the PMMA-lung profile in respect to the other two profiles. Moreover, a 1 mm ion range shift was clearly visible on the registered PG profiles (see PMMA vs PMMA-Teflon phantoms in Fig. 2). However, these results are not the only outcome to consider from this study since the methods applied herein may constitute promising approaches for a clinical device. In this regard, both the background subtraction method and the PGPL determination may be good candidates for clinical application. The former is a relatively simple method that can be easily implemented in, e.g., a field-programmable gate array (FPGA) for online control. It requires only TOF information and a reference detector placed proximal to the patient (outside the detector FOV). A blind detector (i.e., a detector behind shielding) placed in the multislit collimator device may be suited as well, although further experimental work is needed to validate such an assumption. On the other hand, the PGPL quantity Medical Physics, Vol. 42, No. 5, May 2015
Profile length (mm) Ion range (mm)
BaF2 — −1.93 ± 0.58 4.10 ± 0.54
NaI(Tl) — −1.84 ± 1.27 4.39 ± 0.80
as described in this study provides a good correlation with ion range, thus being a good possible quantity for absolute ion range monitoring based on the information provided by a single device. One may also consider the use of optical and beam delivery systems to pinpoint the position where the beam is entering the patient but this requires the integration of at least three separate devices: optical, beam delivery, and PG camera systems. It should also be noted that better functions to retrieve the PGPL may be considered for the actual monitoring. This choice will inevitably depend on the use of background subtraction, the sampling of the collected data, and robustness of the correlation with the ion range. As an example, nonuniform rational basis spline (NURBS) functions are good candidates as well.8 These results and conclusions are obtained for a relatively simple scenario while a clinical one is much more complex. Moreover, the typical number of carbon ions delivered to the phantom for each measurement position is 1010 with a beam intensity of 2 × 108 ions s−1. As comparison, Krämer et al.9 state a total of 7 × 108 carbon ions in a target volume (around 120 cm3) to deliver an absorbed dose of 1 Gy with ∼10 000 raster positions and 39 energy slices. Based on these numbers, the monitoring on a raster-scanning basis seems unfeasible since less than one event per data point per raster spot is expected. Regarding monitoring at the energy slice scale, a procedure to estimate the PGPL retrieval at this statistics level was developed. Since the total number of ions per longitudinal position in the experiment is known, the experimental data are first rescaled to the statistics of an energy slice (i.e., 1.8 × 107 ions) and the data points are then randomly sampled with a Poisson distribution. Finally, the sigmoid function is used to fit the new profiles. This process is repeated hundreds of times with different seed numbers to gather enough statistical evidence. It was found that for 58% of the cases, the same PGPL (within ±2σ) was retrieved for the three phantoms with respect to the full experimental statistics. However, these conclusions are for the present configuration with a very challenging design (i.e., very fine collimation). It is therefore expected that an optimized device would yield a significantly improved performance. In fact, by using a design optimized to achieve a reasonable spatial resolution during PG monitoring with proton beams (configuration “case 3” in Ref. 10), this success rate increases to 93% using 4 simulations. Please note that one cannot use directly the results from simulations since they overestimate the experimental data. Therefore, the
2346
Pinto et al.: Experimental carbon ion range verification in inhomogeneous targets
experimental setup was simulated and the scaling factor that minimizes the root-mean-square error between experimental and simulated data was determined (0.42). The outcomes of the simulations were then scaled by this factor, and the method to compute the PGPL retrieval success rate at the energy slice scale was applied to them. One can argue that some of the profile features associated with the inhomogeneities may be lost with cameras having higher slit FOV; however, such cameras are more efficient. As already shown for PG emitted during proton irradiation,8 the retrieval precision of the PG profile falloff is inversely proportional to the signal-to-noise ratio of the PG profile; thus, a higher collection efficiency ultimately improves the ion range shift estimation. The only restriction to this improvement is the blurring of the profile, which prevents the retrieval of the profile characteristics. The estimation of such a limit for the slit FOV is outside the scope of the present work, but at least the optimized design used does not have a significant effect on the PGPL estimation, as it can be understood from the 93% success rate in having the same PGPL as with the full statistics of the experimental data (in this case, slit FOV is 13.1 mm, cf. Table 6 in Ref. 10). In addition, the uncertainty in the PGPL shifts with the statistics of an energy slice is of the same order of the typical clinical margins [i.e., 2–3 mm (Ref. 11)]. It should also be stressed that the configuration considered was optimized for proton therapy; therefore, it is not necessarily the most adequate to be used with carbon ions. Since it was a design already available in the literature, it was used to discuss the potential of the technique. Nevertheless, there is a remark concerning monitoring at the scale of an energy slice. In scenarios where the waterequivalent path length is similar for all the ions in an energy slice (e.g., treatment of a brain tumor), a single PG profile falloff is expected. On the other hand, for the cases where this condition is not met, the ions will have different ranges; thus, several falloffs will be present in the detected PG profile. Although the best clinical practice for such a case is not as yet clear, one can envisage three approaches: (1) the treatment planning system triggers the monitoring device to only detect PG events on a same-ion range basis; (2) more complex functions are used for the fitting, which may allow to separate the different PGPL; and (3) the monitoring system only considers the comparison between an expected distribution (e.g., Monte Carlo simulations) and the detected one, thus it does not estimate the PGPL for each case with different ranges (although one can always retrieve them from the simulations). Interesting enough is the fact that one can compare independently the different expected ion ranges of the treatment plan; hence, the detection of a potential ion range shift can be cross-checked in several beam sets (i.e., energy slice or ion range based). Finally, it should be noted that the methods for background subtraction and PGPL determination should also be applicable to other cases, notably to proton therapy.
Medical Physics, Vol. 42, No. 5, May 2015
2346
5. CONCLUSIONS Three PG profiles were measured using a 95 MeV/u 12C primary beam with a homogeneous PMMA phantom and two others with heterogeneities along the ion path resulting in ion range shifts. The longitudinal PG profiles were reconstructed using TOF to discriminate between PG and background, and a single-slit collimator was used to ensure a proper longitudinal position and angular selection. Measurements were performed with two independent detectors, resulting in a good mutual agreement. A correlation was observed between ion range shifts induced by the heterogeneities and the detected PG profiles. Ion range shifts as small as one to two millimeters can be detected in the last two centimeters before the Bragg peak as was shown for the different phantoms.
ACKNOWLEDGMENTS This work has been supported by the Rhône Alpes Program for Research in Hadrontherapy (PRRH, CPER 2007-13), the ENVISION European project (G.A. #241851), the ENTERVISION network (G.A. #264552), Labex PRIMES (ANR11-LABX-0063), and France Hadron (ANR-11-INBS-0007). M.D.R. acknowledges support from the FWO Flanders (Fonds voor Wetenschappelijk Onderzoek).
a)Author
to whom correspondence should be addressed. Electronic mail: [email protected] 1C.-H. Min et al., “Prompt gamma measurements for locating the dose falloff region in the proton therapy,” Appl. Phys. Lett. 89(18), 183517 (2006). 2E. Testa et al., “Monitoring the Bragg peak location of 73 MeV/u carbon ions by means of prompt γ-ray measurements,” Appl. Phys. Lett. 93(9), 093506 (2008). 3K. Gunzert-Marx et al., “Secondary beam fragments produced by 200 MeV 12 u−1 C ions in water and their dose contributions in carbon ion radiotherapy,” New J. Phys. 10(7), 075003 (2008). 4P. Henriquet et al., “Interaction vertex imaging (IVI) for carbon ion therapy monitoring: A feasibility study,” Phys. Med. Biol. 57(14), 4655–4669 (2012). 5F. M. F. C. Janssen et al., “Factors influencing the accuracy of beam range estimation in proton therapy using prompt gamma emission,” Phys. Med. Biol. 59(15), 4427–4441 (2014). 6S. Agostinelli et al., “ 4 - a simulation toolkit,” Nucl. Instrum. Methods Phys. Res., Sect. A 506(3), 250–303 (2003). 7J. C. Polf et al., “Measurement of characteristic prompt gamma rays emitted from oxygen and carbon in tissue-equivalent samples during proton beam irradiation,” Phys. Med. Biol. 58(17), 5821–5831 (2013). 8F. Roellinghoff et al., “Real-time proton beam range monitoring by means of prompt-gamma detection with a collimated camera,” Phys. Med. Biol. 59(5), 1327–1338 (2014). 9M. Krämer et al., “Treatment planning for heavy-ion radiotherapy: Physical beam model and dose optimization,” Phys. Med. Biol. 45(11), 3299–3317 (2000). 10M. Pinto et al., “Design optimisation of a TOF-based collimated camera prototype for online hadrontherapy monitoring,” Phys. Med. Biol. 59(24), 7653–7674 (2014). 11O. Jäkel et al., “Treatment planning for heavy ion radiotherapy: Clinical implementation and application,” Phys. Med. Biol. 46(4), 1101–1116 (2001).
