Advertisement

Sensitivity study of prompt gamma imaging of scanned beam proton therapy in heterogeneous anatomies

Published:November 25, 2015DOI:https://doi.org/10.1016/j.radonc.2015.11.002

      Abstract

      Background and purpose

      To investigate the use of a fast analytical prediction algorithm in the evaluation of the accuracy in Bragg peak position estimation using prompt gamma imaging in realistic anatomies.

      Material and methods

      Brain, nasal cavity and lung spot scanning treatments were planned on an anthropomorphic phantom. Plan delivery in a clinical proton therapy facility was monitored using a prompt gamma camera. A pencil-beam algorithm was developed to simulate prompt gamma acquisition. For each spot, the sensitivity to setup and CT conversion errors was evaluated based on error scenarios.

      Results

      Good agreement was found between simulations and measurements (average shift of 0.4 mm on whole-layer profiles). The spots with greatest sensitivity to setup or CT conversion errors could be identified. The comparison between expected and estimated shifts showed that the errors in shift estimation due to heterogeneities were in average lower than 1 mm in all cases except the lung. In the lung case, only 40% of the spots showed accuracy better than 2 mm.

      Conclusions

      The analytical prediction algorithm was successfully used to simulate prompt gamma acquisitions of scanned treatment plans. The accuracy in Bragg peak position estimation was generally sub-millimeter in heterogeneous anatomies, except in lung tissues.

      Keywords

      Proton therapy is sensitive to range uncertainties [
      • Paganetti H.
      Range uncertainties in proton therapy and the role of Monte Carlo simulations.
      ]. These uncertainties can be reduced by on-line in vivo monitoring of the Bragg peak position using prompt gamma (PG) imaging [

      Stichelbaut F, Jongen Y. Verification of the proton beam position in the patient by the detection of prompt gamma-rays emission. In: 39th Meet. of the Particle Therapy Co-Operative Group; 2003.

      ,
      • Smeets J.
      • Roellinghoff F.
      • Prieels D.
      • et al.
      Prompt gamma imaging with a slit camera for real-time range control in proton therapy.
      ,
      • Bom V.
      • Joulaeizadeh L.
      • Beekman F.
      Real-time prompt gamma monitoring in spot-scanning proton therapy using imaging through a knife-edge shaped slit.
      ,
      • Peterson S.W.
      • Robertson D.
      • Polf J.
      Optimizing a three-stage Compton camera for measuring prompt gamma rays emitted during proton radiotherapy.
      ,
      • Min C.H.
      • Kim C.H.
      • Youn M.Y.
      • Kim J.W.
      Prompt gamma measurements for locating the dose falloff region in the proton therapy.
      ,
      • Min C.H.
      • Lee H.R.
      • Kim C.H.
      • Lee S.B.
      Development of array-type prompt gamma measurement system for in vivo range verification in proton therapy.
      ,
      • Testa M.
      • Bajard M.
      • Chevallier M.
      • et al.
      Real-time monitoring of the Bragg-peak position in ion therapy by means of single photon detection.
      ,
      • Verburg J.M.
      • Seco J.
      Proton range verification through prompt gamma-ray spectroscopy.
      ,
      • Golnik C.
      • Hueso-González F.
      • Müller A.
      • et al.
      Range assessment in particle therapy based on prompt γ-ray timing measurements.
      ,
      • Pinto M.
      • Dauvergne D.
      • Freud N.
      • et al.
      Design optimisation of a TOF-based collimated camera prototype for online hadrontherapy monitoring.
      ]. In pencil beam scanning (PBS), PG emission can be measured for each pencil beam of the treatment and compared to the treatment planning in order to detect potential discrepancies in Bragg peak position due to errors in geometry and dose computation. Previous studies have shown that a knife-edge slit PG camera prototype allowed detecting shifts in Bragg peak position with a 1–2 mm precision at clinical beam currents in homogeneous phantoms [
      • Smeets J.
      • Roellinghoff F.
      • Prieels D.
      • et al.
      Prompt gamma imaging with a slit camera for real-time range control in proton therapy.
      ,
      • Perali I.
      • Celani A.
      • Bombelli L.
      • et al.
      Prompt gamma imaging of proton pencil beams at clinical dose.
      ]. However, the accuracy of this PG imaging system has not been studied in realistic anatomies.
      In this paper, we propose a method to evaluate the impact of patient setup and CT conversion errors on the Bragg peak position for realistic treatment plans and anatomies, and to evaluate the potential decrease in PG-based shift retrieval accuracy due to heterogeneities. This was performed by simulating PG emission and PG detection in the presence of setup and CT conversion errors on an anthropomorphic phantom. Because of the large number of spots to be analyzed, we implemented a fast analytical PG simulation tool. In order to validate the simulation tool, the PG detection profiles simulated in the reference conditions were compared to high dose measurements acquired in the reference conditions on the anthropomorphic phantom in a clinical proton therapy facility.

      Materials and methods

      Treatment planning

      A whole body anthropomorphic phantom (PBU-60, Kyoto Kagaku Hyohon Co.) was used in this study. The phantom was imaged using a Brilliance Big Bore CT scanner (Philips). Arbitrary targets were delineated in the brain, the nasal cavity and the lung of the phantom. The treatment planning was performed with RayStation (Raysearch Laboratories Inc.). For each target, a single beam angle was used for planning. A lateral beam was used for the brain case and an anterior beam was used for the nasal cavity and lung cases.

      Prompt gamma imaging

      The PG imaging prototype used in this study has been described in [
      • Perali I.
      • Celani A.
      • Bombelli L.
      • et al.
      Prompt gamma imaging of proton pencil beams at clinical dose.
      ]. The PG camera consisted of a photodetection system and a knife-edge slit tungsten collimator mounted on a dedicated trolley positioning system. The photodetection system was composed of two rows of 20 LYSO crystal slabs and the slabs were 4 mm wide along the beam axis, 10 cm high and 3.15 cm deep. The energy detection window of the photodetectors was set to 3–6 MeV. The distance from the collimator center to the center of the crystals was set to 16 cm. The collimator was placed at 15 cm from the beam axis for monitoring brain and nasal cavity cases, and at 20 cm for the lung case, resulting in 7.5 and 10 cm field-of-view, respectively.
      The experimental study was performed at the Proton Therapy Center in Prague. For each treatment site, the phantom was aligned using orthogonal X-rays. Then, the first five distal layers of each treatment plan were delivered. This corresponds to 496 spots for the brain case, 288 for the nasal cavity case and 510 for the lung case.
      For the purpose of this study, the whole treatment dose (60 Gy) was delivered at once, in order to evaluate the accuracy of the simulation tool using measurements with high statistics and therefore limit the impact of Poisson noise on the detected profiles. The average number of delivered protons per spot was of 2.2 × 108, 2.2 × 108 and 5.4 × 108 for brain, nasal cavity and lung, respectively. The spot size in air was roughly 4 mm (one sigma) at isocenter. The delivery was monitored by the PG imaging system and a PG profile was retrieved for each spot of the delivered layers. The PG profile resulting from complete layer delivery was also computed by accumulating the profiles from the different spots after projection on the beam axis.

      Analytical prompt gamma prediction algorithm

      Since the analytical model has been extensively described elsewhere [
      • Sterpin E.
      • Janssens G.
      • Smeets J.
      • et al.
      Analytical computation of prompt gamma ray emission and detection for proton range verification.
      ], only a very brief description is given here. The first idea is that distributions of PG emission in depth could be handled to some extent as depth-dose distributions in dose calculation algorithms based on pencil beam convolution. Practically, the PG emission signal is computed through a composition-weighted addition of Monte Carlo PG emission profiles previously generated in 12C, 14N, 16O, 31P or 40Ca. For the generation of reference PG emission profiles, the MC code used was PENELOPE extended to protons (PENH) [
      • Sterpin E.
      • Sorriaux J.
      • Vynckier S.
      Extension of PENELOPE to protons: simulation of nuclear reactions and benchmark with Geant4.
      ].
      The PG emission profile is convolved with a transfer function of the camera to generate the PG detection profile. The transfer function is obtained by interpolating a table of analytical fits to previously Monte Carlo simulated responses of the camera. Finally, optical and spectral properties of the incident proton beam are also taken into account in the analytical model [
      • Grevillot L.
      • Bertrand D.
      • Dessy F.
      • Freud N.
      • Sarrut D.
      GATE as a GEANT4-based Monte Carlo platform for the evaluation of proton pencil beam scanning treatment plans.
      ]. The incident spot has a Gaussian shape and 49 ray-tracings per spot were used in this study to simulate the size of the spot.
      The performance and the limitations of the analytical model are described in detail in [
      • Sterpin E.
      • Janssens G.
      • Smeets J.
      • et al.
      Analytical computation of prompt gamma ray emission and detection for proton range verification.
      ]. For all the experiments performed during that study in heterogeneous phantoms, the analytical model has shown an accuracy comparable to PENH that was previously benchmarked with MCNPX [
      • Sterpin E.
      • Janssens G.
      • Smeets J.
      • et al.
      Analytical computation of prompt gamma ray emission and detection for proton range verification.
      ].
      The Hounsfield unit (HU) to density conversion curve was based on [
      • Schneider W.
      • Bortfeld T.
      • Schlegel W.
      Correlation between CT numbers and tissue parameters needed for Monte Carlo simulations of clinical dose distributions.
      ]. The relative stopping power conversion curve was first computed based on the Bethe equation using an analytical estimation of the ionization potential [
      • Sterpin E.
      • Sorriaux J.
      • Vynckier S.
      Extension of PENELOPE to protons: simulation of nuclear reactions and benchmark with Geant4.
      ]. It was then adjusted for both types of phantom materials (SZ-50 for soft tissue and organs, EZ-100 for bones) by a correction factor computed based on 2 high-statistics pencil beams crossing mostly SZ-50 and mostly EZ-100, respectively.

      Sensitivity and error in shift estimation

      The purpose of this PG imaging system is to detect potential discrepancies between the expected Bragg peak positions (i.e. as computed by the TPS) and the actual positions (i.e. indirectly measured by the PG camera). In this study, the position of the beam relative to the room isocenter in the plane orthogonal to its axis was assumed to be known. Therefore, the position of the Bragg peak refers hereafter to its position along the beam axis. Several sources of uncertainty can lead to discrepancies in Bragg peak position and therefore in dose distribution. These uncertainties can be modeled in the simulation framework. By analyzing the modification of the expected Bragg peak position as a function of such uncertainties, one can identify the spots that will be the most sensitive to each specific source of uncertainty. Furthermore, the analytical prediction algorithm can be used to simulate the response of the camera as a function of the uncertainties. It is then possible to estimate how anatomical heterogeneities are expected to hamper the Bragg peak position retrieval from PG measurements.
      In this study, setup and CT conversion errors were considered. In addition to the nominal scenario (i.e. assuming no error), 6 scenarios were simulated: errors of ±3 mm in directions perpendicular to the beam axis, as well as systematic ±5% error in Hounsfield units to stopping power conversion curve. For each scenario, the expected Bragg peak position was computed for all spots of the first five layers using a continuous slowing down approximation (CSDA). As each spot was modeled using 49 ray-tracings, the Bragg peak position was computed as the weighted average (according to spot spatial distribution [
      • Sterpin E.
      • Janssens G.
      • Smeets J.
      • et al.
      Analytical computation of prompt gamma ray emission and detection for proton range verification.
      ]) of Bragg peak position from each of these trajectories. In the following, the CSDA-based calculation was used as reference for evaluating the PG-based Bragg peak position retrieval.
      The sensitivity of each spot was derived from the mean absolute difference in CSDA-based Bragg peak position induced by the simulated errors. A separate sensitivity was computed for each source of error in order to identify to which uncertainty each spot is sensitive. The shifts in Bragg peak position computed using the CSDA calculation are referred to as S.
      PG detection profiles were simulated for all spots. The shifts in Bragg peak position were also estimated by applying a matching algorithm between the reference profiles and the corresponding profiles in the simulated scenarios. Shifts were retrieved using a 1D least square matching on profiles after Gaussian smoothing (sigma of 20 mm) and subtraction of their respective average value. The estimated shifts are referred to as Ŝ. The error in shift estimation was then computed as the difference between the CSDA-based calculation S and the matching-based shift estimation Ŝ.
      This error does not take into account potential uncorrected measurement noise, e.g. due to Poisson noise, neutron background, detection uniformity and camera dead time. Indeed, the focus of this study was the marginal impact of heterogeneities on the shift retrieval process and not the precision of the PG imaging system regarding different sources of uncertainty such as camera positioning and imaging noise, the latter being discussed in [
      • Perali I.
      • Celani A.
      • Bombelli L.
      • et al.
      Prompt gamma imaging of proton pencil beams at clinical dose.
      ] and [
      • Smeets J.
      • Roellinghoff F.
      • Prieels D.
      • et al.
      Prompt gamma imaging with a slit camera for real-time range control in proton therapy.
      ].

      Results

      The first five layers of each treatment plan were successfully delivered and monitored by the PG imaging system. Good agreement was found between simulation and measurement. The shift retrieved from complete layer profiles (i.e. aggregation of all spot profiles from a layer) was sub-millimeter for all layers in all cases, with an average absolute value of 0.4, 0.3 and 0.5 mm for brain, nasal cavity and lung cases, respectively. A comparison between measured and simulated profiles is depicted in Fig. 1.
      Figure thumbnail gr1
      Fig. 1Simulated (solid) and measured (dotted) PG profiles from full energy layers (1st layer in dark red, 3rd layer in green and 5th layer in light blue) for a delivered dose of 60 Gy on the anthropomorphic phantom (top: brain; middle: nasal cavity; bottom: lung). The profiles were centered on zero by subtracting their respective average value. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
      Because of the regular entrance surface and the homogeneity of the tissues in the brain, the sensitivity of the Bragg peak position with respect to the setup errors was very low for the brain case, with sub-millimeter variations in Bragg peak position, as illustrated in the histograms of error-induced shifts in Bragg peak position in Fig. 2. The sensitivity to setup errors was higher in more complex anatomies (up to 6 mm in the nasal cavity and up to 20 mm in the lung). All cases were sensitive to errors in CT conversion, with average shifts of 3.9, 2.7 and 7.0 mm for brain, nasal cavity and lung, respectively.
      Figure thumbnail gr2
      Fig. 2Histograms of absolute shifts in Bragg peak position |S| (green) and absolute error in shift estimation |SŜ| (dashed red) for the simulated sources of error (top: brain; middle: nasal cavity; bottom: lung; left and center: setup errors according to each axis of the beam-eye-view plane; right: error in CT conversion). The thin vertical lines indicate the mean values. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
      The average accuracy in shift estimation was better than 1 mm in all cases except the lung, as illustrated by the histogram of errors in shift retrieval in Fig. 2. The accuracy in shift estimation for all error scenarios was better than 1 mm for 85%, 81% and 19% of the spots for brain, nasal cavity and lung, respectively. The accuracy in shift estimation for all error scenarios was better than 2 mm for 100%, 98% and 40% of the spots for brain, nasal cavity and lung, respectively. In the lung case, the spots showing large errors in shift estimation were localized in some specific areas in the beam, as highlighted in Fig. 3 for the CT conversion errors. Although all spots are sensitive to errors in CT conversion, spots on the border of the irradiation field showed larger range shifts due to the lung geometry. The error in range estimation is very large for such spots, while it remains relatively low (<2 mm) in the center of the beam.
      Figure thumbnail gr3
      Fig. 3Maps of absolute shifts in Bragg peak position |S| (top) and absolute error in shift estimation |SŜ| (bottom) for an error in CT conversion for the lung case. Layers 1–5 are displayed from left to right.

      Discussion

      PG profiles were successfully simulated using a fast analytical algorithm and the simulations were compared to high dose measurements on an anthropomorphic phantom for validation. The comparison between simulated and measured profiles of complete layers indicates that the amplitude and shape of the correlated part of the PG detection profiles are accurately simulated. This indicates also that with this specific PG imaging system, the neutron-induced background has no significant impact on the shape of the profiles and can therefore be removed from the measurement by subtracting the average profile value. This makes the simulation of the neutron contribution unnecessary. The purpose of the measurement was to assess the quality of the simulation. Therefore, high dose was delivered in order to improve the statistics of the measured profiles.
      In practice, several sources of noise can decrease the quality of the measured profiles, limiting the precision in shift estimation. The most important source of noise is the statistical noise, which is proportional to the number of protons per spot. Simple one-dimensional profile matching was shown to be robust against statistical noise. Indeed, previous studies on passively collimated PG cameras demonstrated a precision better than 2 mm (sigma) for clinical pencil beams from typical 2 Gy/fraction treatment plans in homogeneous phantoms. This was demonstrated for single slit [
      • Smeets J.
      • Roellinghoff F.
      • Prieels D.
      • et al.
      Prompt gamma imaging with a slit camera for real-time range control in proton therapy.
      ,
      • Perali I.
      • Celani A.
      • Bombelli L.
      • et al.
      Prompt gamma imaging of proton pencil beams at clinical dose.
      ] and multi-slit [
      • Roellinghoff F.
      • Benilov A.
      • Dauvergne D.
      • et al.
      Real-time proton beam range monitoring by means of prompt-gamma detection with a collimated camera.
      ] collimator designs. Another method based on the inflection point of a sigmoid fit demonstrated similar precision in Bragg peak position estimation [
      • Janssen F.
      • Landry G.
      • Cambraia Lopes P.
      • et al.
      Factors influencing the accuracy of beam range estimation in proton therapy using prompt gamma emission.
      ]. However, these studies were performed in non-realistic phantom geometries. In heterogeneous anatomies, setup and CT conversion errors can result not only in shifts, but also in distortions of the profile shapes. In such cases, the accuracy of the profile matching-based shift estimation has to be assessed. A recent study on a 1D-heterogeneous cylindrical phantom [
      • Priegnitz M.
      • Helmbrecht S.
      • Janssens G.
      • et al.
      Measurement of prompt gamma profiles in inhomogeneous targets with a knife-edge slit camera during proton irradiation.
      ] showed that, in some specific cases, accurate shift estimation was not possible. Therefore, it is necessary to identify, among all the spots, the good candidates for range assessment (i.e. the spots for which the accuracy in shift estimation is better than the 1–2 mm precision of the PG monitoring system). Performing the Bragg peak shift analysis on simulated error scenarios allows identifying the spots for which shifts cannot be retrieved properly, even without noise.
      The average accuracy in shift estimation was generally better than 1 mm, except in some cases. In particular, large errors were observed on the border of the irradiation field in the lung case. The areas that PG will not be able to monitor accurately can be clearly identified based on such analysis. There are mainly two causes for these large errors. First, because of the very low density of the lung, some over-ranged spots stopped at the very border of – or even outside – the field of view of the camera, which was limited to 10 cm with this setup. Second, the PG emission in lung tissue is very low, so that spots stopping in the lung or a few millimeters after the lung do not emit sufficient photons at the end of their range compared to the signal emitted upstream in soft tissues. In such cases, the transition from soft tissues to lung tissue is actually interpreted by the PG imaging system as the fall-off in the PG emission, which is the surrogate for the Bragg peak position. Therefore, a limitation of the PG imaging system is that the spots stopping within low-density lung tissue or a few millimeters beyond lung tissue should be excluded from the Bragg peak position monitoring. This is in agreement with the results in [
      • Priegnitz M.
      • Helmbrecht S.
      • Janssens G.
      • et al.
      Measurement of prompt gamma profiles in inhomogeneous targets with a knife-edge slit camera during proton irradiation.
      ].
      All spots were sensitive to errors in CT conversion. On the contrary, the spots from the brain treatment were not sensitive to setup errors due to the smoothness of the head surface and the lack of heterogeneities. PG imaging will therefore not be able to detect setup errors in the directions orthogonal to the beam. In the nasal cavity and lung cases, only a subset of spots was sensitive to setup errors. These were spots located close to heterogeneities in the densities crossed by the proton beam.
      Only setup errors orthogonal to the beam axis were investigated. Indeed, a setup error in the direction of the beam results roughly in a simple translation of the dose and PG profiles. In that case, the shift in Bragg peak position is simply equal to this setup error for all spots. Furthermore, the matching between the two translated profiles is trivial and gives a result equal to the setup error.
      Regarding the possible strategies for treatment adaptation based on Bragg peak position monitoring, the use of a few probe spots at the beginning of the treatment is a promising approach for online range verification and adaptation. In this case, probe spots can be selected based on the sensitivity to the different errors verified by the therapist. For instance, as illustrated in Fig. 4, setup errors in two different directions will not impact every spot in the same way. Some spots will be more sensitive to certain errors and thus will be better candidates for probing. Alternatively, if the strategy is to monitor all delivered spots during a treatment for retrospective analysis, the pattern of estimated shifts could be used to retrieve the most probable sources of error. In Fig. 4, one can clearly see that the distribution in shift amplitudes differs according to the direction of the setup error.
      Figure thumbnail gr4
      Fig. 4Maps of absolute shifts in Bragg peak position |S| for the nasal cavity case (anterior beam), for setup errors in the left–right (left part of the figure) and superior–inferior (right part of the figure) directions.
      In this study, only setup and CT conversion errors were investigated. Other sources of errors, such as rotations, local CT conversion errors and anatomical variations, could be studied in the same way. Furthermore, more advanced range analysis, such as machine learning-based dose monitoring [
      • Gueth P.
      • Dauvergne D.
      • Freud N.
      • et al.
      Machine learning-based patient specific prompt-gamma dose monitoring in proton therapy.
      ], could benefit from fast PG simulation in order to generate larger sets of training data in reasonable amount of time.

      Conflict of interest statement

      G. Janssens, J. Smeets, F. Vander Stappen, D. Prieels are employees of IBA.
      The appointment of E. Clementel and E. Hotoiu is partially funded by IBA.
      E. Sterpin is partially supported by IBA.

      Acknowledgments

      The authors thank the Proton Therapy Center of Prague for hosting and supporting the PG measurement.

      Appendix A. Supplementary data

      Figure thumbnail fx1
      Supplementary Fig. 1Left: axial view of the target volume and planned dose on the CT for brain, nasal cavity and lung cases (from top to bottom). Right: picture of the PG monitoring setup for each case. The PG slit opening was positioned orthogonally to the beam axis.

      References

        • Paganetti H.
        Range uncertainties in proton therapy and the role of Monte Carlo simulations.
        Phys Med Biol. 2012; 57: R99-117
      1. Stichelbaut F, Jongen Y. Verification of the proton beam position in the patient by the detection of prompt gamma-rays emission. In: 39th Meet. of the Particle Therapy Co-Operative Group; 2003.

        • Smeets J.
        • Roellinghoff F.
        • Prieels D.
        • et al.
        Prompt gamma imaging with a slit camera for real-time range control in proton therapy.
        Phys Med Biol. 2012; 57: 3371-3405
        • Bom V.
        • Joulaeizadeh L.
        • Beekman F.
        Real-time prompt gamma monitoring in spot-scanning proton therapy using imaging through a knife-edge shaped slit.
        Phys Med Biol. 2011; 57: 297-308
        • Peterson S.W.
        • Robertson D.
        • Polf J.
        Optimizing a three-stage Compton camera for measuring prompt gamma rays emitted during proton radiotherapy.
        Phys Med Biol. 2010; 55: 6841-6856
        • Min C.H.
        • Kim C.H.
        • Youn M.Y.
        • Kim J.W.
        Prompt gamma measurements for locating the dose falloff region in the proton therapy.
        Appl Phys Lett. 2006; 89: 183517
        • Min C.H.
        • Lee H.R.
        • Kim C.H.
        • Lee S.B.
        Development of array-type prompt gamma measurement system for in vivo range verification in proton therapy.
        Med Phys. 2012; 39: 2100-2107
        • Testa M.
        • Bajard M.
        • Chevallier M.
        • et al.
        Real-time monitoring of the Bragg-peak position in ion therapy by means of single photon detection.
        Radiat Environ Biophys. 2010; 49: 337-343
        • Verburg J.M.
        • Seco J.
        Proton range verification through prompt gamma-ray spectroscopy.
        Phys Med Biol. 2014; 59: 7089-7106
        • Golnik C.
        • Hueso-González F.
        • Müller A.
        • et al.
        Range assessment in particle therapy based on prompt γ-ray timing measurements.
        Phys Med Biol. 2014; 59: 5399-5422
        • Pinto M.
        • Dauvergne D.
        • Freud N.
        • et al.
        Design optimisation of a TOF-based collimated camera prototype for online hadrontherapy monitoring.
        Phys Med Biol. 2014; 59: 7653-7674
        • Perali I.
        • Celani A.
        • Bombelli L.
        • et al.
        Prompt gamma imaging of proton pencil beams at clinical dose.
        Phys Med Biol. 2014; 59: 5849-5871
        • Sterpin E.
        • Janssens G.
        • Smeets J.
        • et al.
        Analytical computation of prompt gamma ray emission and detection for proton range verification.
        Phys Med Biol. 2015; 60: 4915-4946
        • Sterpin E.
        • Sorriaux J.
        • Vynckier S.
        Extension of PENELOPE to protons: simulation of nuclear reactions and benchmark with Geant4.
        Med Phys. 2013; 40: 111705
        • Grevillot L.
        • Bertrand D.
        • Dessy F.
        • Freud N.
        • Sarrut D.
        GATE as a GEANT4-based Monte Carlo platform for the evaluation of proton pencil beam scanning treatment plans.
        Phys Med Biol. 2012; 57: 4223-4244
        • Schneider W.
        • Bortfeld T.
        • Schlegel W.
        Correlation between CT numbers and tissue parameters needed for Monte Carlo simulations of clinical dose distributions.
        Phys Med Biol. 2000; 45: 459-478
        • Roellinghoff F.
        • Benilov A.
        • Dauvergne D.
        • et al.
        Real-time proton beam range monitoring by means of prompt-gamma detection with a collimated camera.
        Phys Med Biol. 2014; 59: 1327-1338
        • Janssen F.
        • Landry G.
        • Cambraia Lopes P.
        • et al.
        Factors influencing the accuracy of beam range estimation in proton therapy using prompt gamma emission.
        Phys Med Biol. 2014; 59: 4427-4441
        • Priegnitz M.
        • Helmbrecht S.
        • Janssens G.
        • et al.
        Measurement of prompt gamma profiles in inhomogeneous targets with a knife-edge slit camera during proton irradiation.
        Phys Med Biol. 2015; 60: 4849-4871
        • Gueth P.
        • Dauvergne D.
        • Freud N.
        • et al.
        Machine learning-based patient specific prompt-gamma dose monitoring in proton therapy.
        Phys Med Biol. 2013; 58: 4563-4577