Physics in Medicine &

Biology

PAPER • OPEN ACCESS You may also like

- A comparative study on automatic
The influence of daily imaging and target margin treatment planning for online adaptive
proton therapy of esophageal cancer:
reduction on secondary cancer risk in image- which combination of deformable
registration and deep learning planning
tools performs the best?
guided and adaptive proton therapy C Draguet, P Populaire, M Chocan Vera et
al.

To cite this article: A Smolders et al 2024 Phys. Med. Biol. 69 225004 - Quantitative assessment of areal bone
mineral density using multi-energy
localizer radiographs from photon-counting
detector CT
Sören Jasper, Joseph Swicklik, Francis
Baffour et al.
View the article online for updates and enhancements.
- The effect of intra- and inter-fractional
motion on target coverage and margins in
proton therapy for uveal melanoma
Daniel Björkman, Riccardo Via, Antony
Lomax et al.

This content was downloaded from IP address 80.39.202.176 on 13/02/2025 at 11:54

 Phys. Med. Biol. 69 (2024) 225004 https://doi.org/10.1088/1361-6560/ad8da3

Physics in Medicine & Biology

PAPER

The influence of daily imaging and target margin reduction on

OPEN ACCESS
secondary cancer risk in image-guided and adaptive proton
RECEIVED
24 July 2024 therapy
REVISED
22 October 2024 A Smolders1,2,∗, K Czerska1, Z Celicanin1, A Lomax1,2 and F Albertini1
ACCEPTED FOR PUBLICATION 1
Center for Proton Therapy, Paul Scherrer Institute, Villigen, Switzerland
31 October 2024 2
Department of Physics, ETH Zurich, Zurich, Switzerland
PUBLISHED ∗
Author to whom any correspondence should be addressed.
11 November 2024
E-mail: [email protected]
Original Content from Keywords: robustness, imaging dose, proton therapy, radiation-induced cancer, image-guided, adaptive, target margin reduction
this work may be used
under the terms of the
Creative Commons
Attribution 4.0 licence.
Abstract
Any further distribution
of this work must Objective. Image-guided and adaptive proton therapy rely on daily CBCT or CT imaging, which
maintain attribution to
the author(s) and the title increases radiation dose and radiation-induced cancer risk. Online adaptation however also
of the work, journal
citation and DOI.
reduces setup uncertainty, and the additional risk might be compensated by reducing the setup
robustness margin. This work developed a framework to investigate how much this robustness
margin should be reduced to offset the secondary cancer risk from additional imaging dose and
applied it to proton therapy for head-and-neck cancer. Approach. For five patients with
head-and-neck cancer, voxel-wise CT and CBCT imaging doses were estimated with Monte Carlo
radiation transport simulations, calibrated with air and PMMA phantom measurements. The total
dose of several image-guided and adaptive treatments protocols was calculated by summing the
planning CT dose, daily and weekly CBCT or CT dose, and therapy dose, robustly optimized with
setup margins between 0 and 4 mm. These were compared to a reference protocol with 4 mm setup
margin without daily imaging. All plans further used 3% range robustness. Organ-wise excess
absolute risk (EAR) of cancer was calculated with three models to determine at which setup margin
the total EAR of image-guided and adaptive treatment protocols was equal to the total EAR of the
reference. Results. The difference between the simulated and measured CT and CBCT doses was
within 10%. Using the Monte Carlo models, we found that a 1 mm setup margin reduction was
sufficient for most patients, treatment protocols, and cancer risk models to compensate the
additional risk imposed by daily and weekly imaging. For some protocols, even a smaller reduction
sufficed, depending on the imaging frequency and type. The risk reduction by reducing the margin
was mainly due to reducing the risk for carcinomas in the brain and, for some patients, the oral
cavity. Significance. Our framework allows to compare an increased imaging dose with the reduced
treatment dose from margin reductions in terms of radiation-induced cancer risk. It is extendable
to different treatment sites, modalities, and imaging protocols, in clinic-specific or even
patient-specific assessments.

1. Introduction

In image-guided radiotherapy (IGRT), an image is acquired each day to reduce setup uncertainty (Chen et al
2009). Based on the image, the patient’s position is corrected by applying shifts and rotations to the
treatment couch which improves the correspondence between the planned and treated patient position
(Dawson et al 2005, Hong et al 2005). This allows the patient to be treated with a lower robustness setting,
for example, with a smaller planning target volume (PTV) margin or lower setup uncertainty in robust
optimization, referred to in the following as a smaller setup robustness margin (Chen et al 2009, Maund et al
2014). This in turn reduces the total dose delivered to the patient, limiting radiation-induced adverse events.

© 2024 The Author(s). Published on behalf of Institute of Physics and Engineering in Medicine by IOP Publishing Ltd
Phys. Med. Biol. 69 (2024) 225004 A Smolders et al

The rigid patient alignment can, however, not account for non-rigid setup variations. Additionally,
anatomical variations cannot be accounted for; hence, the use of IGRT still requires treatment with relatively
large setup robustness margins (Maund et al 2014, Lim-Reinders et al 2017). Online adaptive radiotherapy
aims to overcome these limitations by adapting and reoptimizing the treatment on the daily image
(Lim-Reinders et al 2017). This additionally accounts for anatomical and non-rigid setup variations,
allowing to treat with smaller setup robustness margins (Paganetti et al 2021).
IGRT mostly uses daily cone-beam CTs (CBCT), both for photon and particle therapy. Online adaptation
requires higher-quality images and current photon therapy implementations use daily MRI or high-quality
CBCT (Winkel et al 2019, Byrne et al 2022). Both can also be used for particle therapy, but their application
is less straightforward. The depth of the particles is highly sensitive to the tissue densities along the beam
path and accurate density estimates are necessary to perform dose calculation (Paganetti 2012). MRI or
CBCT images require advanced postprocessing before they can be converted to stopping power ratios, which
introduces additional uncertainties (Paganetti et al 2021). Combining an MRI with a proton therapy system
also introduces other difficulties, like the need for incorporating the magnetic field in the dose calculation
(Duetschler et al 2023) and the limited space inside the gantry (Oborn et al 2017). Whereas adaptation based
on CBCT and MRI will possibly be feasible in the future, as an alternative, a daily CT can be acquired using a
CT-on-rails system installed inside the treatment room (Albertini et al 2020, Nesteruk et al 2021, Paganetti
et al 2021). The conversion of CT numbers to particle stopping power ratios is well-developed within the
particle therapy community. Adaptation based on a daily CT is therefore more straightforward to
implement, and a first clinical implementation was reported recently (Albertini et al 2024).
The downside of acquiring daily CBCT or CT images is that it increases the imaging dose delivered to
patients (Albertini et al 2020). Even though the imaging dose is substantially below the therapy dose, it
covers a much larger volume (Buckley et al 2018), especially in particle therapy because of the high dose
conformality with the target and sharp dose fall-off. Therefore, image guidance and online adaptation lead to
an increased dose in the low-dose regions and, simultaneously, reduce the dose in some high-dose regions
because of the small setup robustness margin.
The reduction in therapy dose in the vicinity of the target volume and increase in imaging dose
everywhere have opposing effects on radiation-induced side effects. Most commonly reported
radiation-induced side effects only occur at doses substantially above the imaging dose (Brodin et al 2018, De
Ruysscher et al 2019, Wang and Tepper 2021), i.e. the reduction of the therapy dose will most likely dominate
over the increase in imaging dose. In contrast, secondary cancer induction by radiation is a stochastic
process, and even low-dose exposure may be important (Wang and Tepper 2021). Moreover, due to cell
killing at high doses, the marginal increase in radiation-induced cancer risk at low dose is higher than at high
dose, i.e. the slope of the dose-effect relation decreases with increasing dose (Schneider 2009, Schneider et al
2011). It is therefore possible that image-guided or adaptive particle therapy increases the risk of
radiation-induced cancers, despite the potential to reduce the setup robustness margin.
In this work, we developed a framework to assess the radiation-induced cancer risk in image-guided and
adaptive therapy and applied it to proton therapy for head-and-neck cancer patients. First, we developed and
calibrated Monte Carlo models to estimate the voxel-wise imaging dose delivered by CT and CBCT scanners.
Secondly, we evaluated at what level of setup margin reduction the reduced therapy dose compensates the
additional radiation-induced cancer risk introduced with daily CT or CBCT imaging. Because of the
non-linear dose-effect relation, inhomogeneous dose distribution, fractionation effects, and different
radiosensitivity of organs, we used radiation-induced cancer risk models instead of evaluating the integral
dose.

2. Materials and methodology

To assess the combined radiation-induced cancer risk of imaging and therapy dose, we first need to quantify
the voxel-wise imaging dose. This is achieved by creating and calibrating Monte Carlo radiation transport
models for our CT and CBCT scanner. With the calibrated models, the imaging dose in patients can be
simulated and summed up voxel-wise with the therapy dose, to use as an input for several organ-specific
secondary cancer risk models.

2.1. CT scanner modeling

A Siemens SOMATOM Sensation Open CT scanner (Siemens Healthineers, Erlangen, Germany) was
modeled using the Monte Carlo code Gate 9.3 (Jan et al 2011, Sarrut et al 2022). We modeled our clinical
standard-dose acquisition protocol (120 kVp, 320 mAs reference exposure), which includes CareDose4D, a
protocol that uses tube current modulation to limit imaging dose (Kalra et al 2004). Gate can run
time-dependent simulations, allowing the tube current to be varied while the x-ray source moves around the

2
Phys. Med. Biol. 69 (2024) 225004 A Smolders et al

Figure 1. Setup of the CT experiments. Experiment 1: with a static x-ray source, the dose was measured in air at 17 locations
between the isocenter and the edge of the fan beam. Experiment 2: with a rotating x-ray source, the dose was measured in the 5
holes of PMMA CTDI phantoms with 16 and 32 cm diameter.

patient for simulating CT protocols involving tube-current modulation. The x-ray source energy spectrum
was modeled using the SpekPy package (Poludniowski et al 2021) with a 120 kVp tube potential, 7◦ anode
angle and 7.5 mm inherent aluminum filtration, according to the CT manual. The resulting spectrum was
cross-validated with the manufacturer.
CT scanners employ bowtie filtration to limit the imaging dose (Boone 2010). However, the material and
thickness of these filters are mostly proprietary. Therefore, instead of including the bowtie filter in the
simulation, we modeled an x-ray source with angle-dependent intensity (Boone 2010) using an approach that
can be applied to any CT scanner without requiring proprietary information. The lateral, i.e. perpendicular
to the patient axis, intensity profile was in our case obtained from the manufacturer, but this can be
measured as well, as described in the next section. The bowtie filter however also affects the x-ray spectrum: a
higher bowtie thickness filters the softer part of the spectrum out more, causing beam hardening. This was
modeled by applying aluminum filtration for each fan angle individually with SpekPy with a thickness that
yielded the corresponding attenuation of the intensity provided by the manufacturer. This filtration was then
applied to the initial spectrum for each angle, resulting in an x-ray source with different intensities and
different x-ray spectra for each fan angle. In the parallel direction, i.e. parallel to the patient axis, the source
was modeled with uniform intensity between angles corresponding to the nominal 6 mm collimation.
The accuracy of the angle-dependent x-ray source was validated with experimental measurements
(figure 1, experiment 1). While keeping the x-ray source static, the dose in air was measured at 17 locations
between the isocenter and edge of the fan beam (26 cm) with a RaySafe X2 CT dose ion chamber (Unfors
RaySafe AB, Hovas, Sweden). The active length of this chamber is 100 mm, which is substantially longer than
the 6 mm nominal collimation of the x-ray source. This device automatically starts and stops its data
acquisition when detecting radiation, simplifying synchronization with the beginning and end of the CT
acquisition. This experimental setup was further modeled with Gate and the experimentally measured dose
was compared with the simulated dose.
A second set of experiments was used to calibrate the simulation and assess its quality (figure 1,
experiment 2). For these, the CT scanner was set in rotating mode, with 0.5 s rotation time, 12 mm s−1 table
translation speed, 2.5 s acquisition time and 40 mA constant tube current (i.e. CareDose4D turned off).
Time-dependent dose-rate profiles were acquired with the RaySafe X2 CT dose ion chamber. Only
measurements between 0.75 and 1.75 s were used for calibration, to avoid start-up and shut-down effects in
the CT and ionization chamber. We introduced both 16 and 32 cm PMMA CTDI phantoms (Sun Nuclear,
Melbourne, Australia) at the isocenter, and acquired a measurement in the center hole and each of the holes
on four sides of the phantom (figure 1, experiment 2). Despite the symmetry of the CT, the measured dose in
each hole varied because the phantoms were supported by the treatment couch. Additionally, we acquired a
similar measurement in air, with the ion chamber at the isocenter, and moved 7 and 15 cm to the left, i.e. the
same locations where the left hole is in the 16 and 32 cm CTDI phantoms, respectively. Note that this is
different from experiment 1 because of the rotation of the x-ray source. For these measurements in air, the
ion chamber was mounted on the head of the treatment couch, so that the couch was not within the field of
view (FOV) and not affecting the measurement. This makes the acquisitions symmetric, so they do not need

3
Phys. Med. Biol. 69 (2024) 225004 A Smolders et al

Figure 2. Setup of the CBCT experiments. Experiment 1: with a rotating x-ray source, the projection images were acquired for an
acquisition in air. Experiment 2: with a rotating x-ray source, the dose was measured in the 5 holes of PMMA CTDI phantoms
with 16 and 32 cm diameter.

to be repeated to the right of the isocenter. The measurement at the isocenter was used for absolute
calibration. All experiments were modeled with Gate, all without the treatment couch for simplicity, and the
measured doses were compared to the simulated ones.

2.2. CBCT scanner modeling

We further modeled the imaging dose originating from the gantry-mounted CBCT scanner in a Varian
Probeam 4.1 gantry (Varian Medical Systems, Palo Alto, USA). This CBCT provides two protocols relevant
for the head-and-neck region, namely the Head protocol (110 kVp, 847 mAs) and the Image Gently protocol
(80 kVp, 548 mAs). The x-ray sources for both protocols were modeled similarly to the CT scanner, with
angle-dependent intensity and x-ray spectra. The x-ray spectra were modeled with SpekPy with a 80 and
110 kVp tube potentials, 7◦ anode angle, 2.7 mm inherent aluminum filtration, 1.3 mm equivalent
aluminum filtration from the collimator polycarbon window and 0.89 mm titanium beam hardening filter,
all taken from the manual (Varian Medical Systems 2020). No intensity profile could be obtained from the
manufacturer. Instead, a CBCT was acquired in air and the projection images acquired with the flat panel
detector were exported from the system (figure 2, experiment 1). Because of the bowtie filtration, the
intensity at the edges of these images was lower than at the center (figure 3). Using the average projection, the
relative drop in intensity at each lateral angle with respect to the center was calculated and used to determine
the equivalent bowtie thickness. This process was performed both with the Image Gently and Head protocols.
This yielded a lateral angle-dependent x-ray spectrum and intensity, with the maximal angles corresponding
to the ± 14.5 cm nominal collimation. In the parallel direction, a uniform intensity was taken between the ±
10.9 cm nominal collimation.
As for the CT, a second set of experiments was used for absolute calibration (figure 2, experiment 2). The
clinical CBCT acquisition includes a 191◦ rotation in 3 s, with 385 projections. First, the dose in air with the
ion chamber at the isocenter was measured for absolute calibration. This measurement was also repeated
with the ion chamber moved 8 cm to the left. Secondly, we introduced the 16 cm PMMA phantom, and
measured the dose with the ion chamber in all the holes. Since the CBCT is not symmetric, and the dose in
the different holes on the edges is different, all measurements were used in the comparison. This process was
performed for both protocols.

2.3. Imaging dose simulation in patients

Using the angle-dependent x-ray sources and absolute calibrations, the voxel-wise CT and CBCT imaging
dose received by patients can be simulated. The imaging dose can be simulated for the planning CT by
converting the CT numbers to material composition and density following the method described by
Schneider et al (2000). For the CT, the pitch factor, table speed, and rotation speed can be extracted from the
DICOM header of the planning CT images, as well as the slice-wise tube current which varies because of the
tube current modulation. For the CBCT, the clinical protocol and exposure can be simulated, assuming the

4
Phys. Med. Biol. 69 (2024) 225004 A Smolders et al

Figure 3. Average intensity of the 385 flat panel projections of a CBCT acquired in air.

Figure 4. Exposure as a function of slice position for the standard-dose and low-dose CTs for an example patient. The distance is
measured from the start of the standard-dose CT scan, slightly above the top of the skull. Exposure increases strongly at the
shoulders (at a distance of ∼ 300 mm), which are within the field of view because of the involvement of the lymph nodes.

same isocenter as the CT and that the scanner was centered on the clinical target volume (CTV) in the
superior-inferior patient axis.
The first clinical implementation of online adaptive proton therapy used a low-dose CT protocol (Nenoff
et al 2021, Albertini et al 2024), and other authors have also proposed similar approaches (Nesteruk et al
2022). Therefore, we also included the low-dose protocol of Nenoff et al (2021) in our study, which lowers
the exposure but still uses 120 kVp, so the same x-ray source as the standard-dose CT can be used. For each
slice, the exposure of the low-dose protocol is substantially below the standard-dose CT (figure 4).
Additionally, for our standard-dose CT protocol, the scan covers the region from the manubrium of the
sternum to the top of the head. Given the coplanar beam arrangement used for the patients studied in this
work, this is unnecessary for daily dose calculation, and the low-dose scan length can be shortened by only
acquiring the low-dose CT from 3 cm above to 3 cm below the CTV.

2.4. Patient data for testing

The data used in this work to test the framework consisted of five patients with sinonasal or nasopharyngeal
tumors with lymph node involvement, previously treated at the Centre for Proton Therapy in Switzerland.
Each patient was treated with intensity-modulated proton therapy (IMPT) with 54, 66, and 70-72 GyRBE
being delivered to the low-, intermediate-, and high-dose CTVs, respectively. For this study, the clinical
treatment plans were emulated in Raystation (RaySearch Laboratories AB, Stockholm, Sweden) using robust
optimization on the different CTVs with 4 mm setup and 3% range robustness (figure 5). Note that, in the
following, we will refer to the setup robustness setting as the setup margin, which does not refer to an
isotropic expansion like the CTV-to-PTV, but to the magnitude of the setup shifts used in robust
optimization. The same optimization objectives were used as in the clinical treatment planning system
(Varian Eclipse, Varian Medical Systems, Palo Alto, USA), but their weights were retuned to achieve similar

5
Phys. Med. Biol. 69 (2024) 225004 A Smolders et al

Figure 5. Proton therapy dose distribution for patient 3.

Table 1. Overview of the type of daily and weekly imaging as well as the potential setup margins for the treatment protocols analyzed in
this work.

Daily image Weekly image Setup margin [mm]

Baseline None None 4

IGRT CBCT (Image Gently) CT 4, 3, 2, 1, or 0
IGRT CBCT (Image Gently) Low-dose CT 4, 3, 2, 1, or 0
Adaptive CBCT (Head) CT 4, 3, 2, 1, or 0
Adaptive CBCT (Head) Low-dose CT 4, 3, 2, 1, or 0
Adaptive Low-dose CT None 4, 3, 2, 1, or 0
Adaptive Full-range low-dose CT None 4, 3, 2, 1, or 0
Adaptive CT None 4, 3, 2, 1, or 0

target coverage (V95) and organ-at-risk sparing as the clinical plan. It was further assumed that all patients
were treated in 33 fractions in 7 weeks with simultaneous integrated boost.

2.5. Treatment protocols

The total dose, i.e. the sum of the therapy and imaging dose, must be computed to evaluate the
radiation-induced cancer risk. However, this largely depends on the type and frequency of the imaging, and
the treatment robustness, and is therefore clinic-dependent. To be comprehensive and not clinic-specific, we
compared several IGRT and adaptive treatment protocols.
First, we simulated a baseline (non-IGRT) protocol, which consisted of one planning CT and treatment
optimized with 4 mm setup and 3% range robustness, assuming that the daily imaging dose was negligible
(table 1). This is representative of daily positioning with dose-free imaging, like MRI, or very low-dose
imaging, such as planar radiographs. Even though this is not the standard of care, it represents a scenario
with negligible imaging dose, useful as a baseline to compare different protocols.
This baseline protocol was compared to several IGRT and adaptive treatment protocols optimized with
3% range robustness and varying setup margins between 0 and 4 mm (table 1). For IGRT, the Image Gently
CBCT protocol can be used for daily positioning and we further included either a low-dose or standard-dose
weekly control CT. Furthermore, we simulated CBCT-guided adaptive treatment protocols with daily CBCTs
applying the higher quality Head protocol together with weekly low-dose or standard-dose control CTs.
Lastly, we also simulated CT-guided adaptation with daily CT images, applying either a low-dose CT, a
low-dose CT without shortening the scan length, i.e. keeping the same length as the planning CT, or a
standard-dose CT. If the scan length was not shortened, we refer to the protocol as the full-range low-dose CT.

2.6. Secondary cancer risk modeling

For each of the different protocols, the radiation-induced cancer risk was estimated using the organ-specific
models described by Schneider et al (2011). These models convert an inhomogeneous dose distribution
within an organ to an organ equivalent dose (OEDorg ), which relates linearly to the excess absolute risk
(EARorg ) of radiation-induced cancers:

EARorg = β EAR OEDorg µ (agex , agea ) (1)

with β EAR the initial slope of the dose-effect relation derived from the Japanese atomic bomb survivors and
µ(agex , agea ) a factor to correct for the age at exposure (agex ) and age attained (agea ). In the following, we

6
Phys. Med. Biol. 69 (2024) 225004 A Smolders et al

Figure 6. Relation between absorbed and risk equivalent dose for the salivary gland with the three secondary cancer models. The
parameters were taken from Schneider et al (2011).

assumed agex = 30 and agea = 70, which were the parameters used to fit the models (Schneider et al 2011),
making µ(agex , agea ) = 1.
The OEDorg is defined as the volumetric average of the voxel-wise risk equivalent doses (REDi )
∑
org vi REDi
i∑
OED = (2)
i vi

with i the indices of the voxels inside the organ and vi the volume of voxel i. The REDi depends on the dose
and accounts for cell killing using the α and β parameters of the linear-quadratic model and fractionation
effects using a repopulation parameter R (Schneider et al 2011). The REDi was calculated with three different
models: the bell model with R = 0, plateau model with R = 1, and the full model, for which the parameter R
was fitted together with α (Schneider et al 2011). For sarcomas, we included models with R = 0.1, 0.5, and 1,
as in Schneider et al (2011). The organ-specific values for R and α can be found in table A1 in the appendix,
considering only those organs that were within the FOV of the planning CT and for which fitted parameters
were available. Following Schneider et al (2011), α/β = 3 for all organs. The parameters of these models
were mostly fitted with doses up to 40 Gy, but the patients in this work were treated with up to 72 Gy. For
some organs, the three models extrapolate differently to such high doses (e.g. for the salivary glands, figure 6)
and it is unclear which model is most appropriate. We therefore included all three. The thyroid and
esophagus only had fitted parameters for one model, and this model was used for all calculations.
The mouth and pharynx were considered as the union of the oral cavity and pharynx segmentation; the
brain and central nervous system (CNS) as the union of the brain, brainstem and spinal cord; and the
salivary glands as the union of the parotid and submandibular glands. If the relevant contours were created
for treatment planning, these were also used in the models. However, some contours were not available.
Therefore, the pharynx was manually segmented with region growing and the esophagus and submandibular
glands were segmented using Limbus Contour 1.7 (AI Limbus Inc. Regina, Canada). The bone was
segmented using the open-source TotalSegmentator (Wasserthal et al 2023) and the soft tissue as the entire
body contour minus the bone. We assumed that the parts of the esophagus and spinal cord that were not
inside the FOV of the CT scan did not receive any dose.
To assess the trade-off between imaging dose and setup margin for the IGRT and adaptive treatment
protocols, we evaluated the difference between the total EARBL of the baseline (BL) with the total EAR for the
IGRT and adaptive protocols, for different levels of setup margin for each patient and model individually.
The total EAR was calculated as the sum of the EARorg for all considered organs. Furthermore, for each
treatment protocols, patient and cancer risk model, we calculated the Required Margin Reduction (RMR),
i.e. the setup margin reduction for which the total EAR was equal to that of the baseline.

3. Results

3.1. Monte Carlo simulation calibration

The normalized simulated lateral dose profile in the CT using the angle-dependent x-ray source matched
closely with the experimentally measured values (figure 7). The measured dose profile seemed to fall off
slightly faster than the simulated one, but we opted to keep using the intensity data provided by the
manufacturer and not adjust the source in the simulation to better match these experiments.

7
Phys. Med. Biol. 69 (2024) 225004 A Smolders et al

Figure 7. Measured and simulated normalized dose profiles of the x-ray source after the bowtie filtration as a function of distance
to the isocenter. The Monte Carlo simulation was normalized to the measured dose at the isocenter.

Table 2. Comparison between the measured Dmeas and simulated dose Dsim for the CT scanner at various locations in air and the PMMA
phantoms (16 and 32 cm diameter). Only the measurement in the top hole of the phantom is reported, as it was least affected by the
presence of the table and the simulation was perfectly symmetric.

Phantom location Dmeas /Dsim

Air center (calibration) 1.00

Air center + 7 cm 0.99
Air center + 15 cm 0.94
PMMA 16 center 0.94
PMMA 16 top 0.95
PMMA 32 center 0.97
PMMA 32 top 0.91

Table 3. Comparison between the measured Dmeas and simulated dose Dsim for the Head and Image Gently protocols of the CBCT
scanner at various locations in air and the 16 cm PMMA phantom. Because the CBCT only rotated 191◦ , the imaging dose was not
symmetric and the dose in all holes of the phantom was reported.

Dmeas /Dsim
Phantom location Head Image Gently

Air center (calibration) 1.00 1.00

Air center + 8 cm 0.91 1.05
PMMA 16 center 0.98 0.93
PMMA 16 top 1.02 0.94
PMMA 16 left 0.97 1.01
PMMA 16 down 0.90 0.96
PMMA 16 right 0.96 1.00

The absolute calibration yielded 1.087 × 1010 particles/mAs for the CT, 1.619 × 109 particles/mAs for the
CBCT with Image Gently and 3.495 × 109 particles/mAs with the Head protocol. The calibration of the two
CBCT protocols is different because of the difference in tube voltage. Using this calibration, we found that all
our simulations were within 10% of the measured values (tables 2 and 3), similar to values reported in the
literature Abuhaimed et al (2014, Tahiri et al 2021). In general, the dose measured in the phantoms was lower
than the one simulated, which was partly due to the treatment table absorbing some energy in the
measurements while it was not included in the simulations.
In patients, the low-dose CT protocol indeed substantially lowered the imaging dose compared to the
standard-dose CT (figure 8). The standard-dose CT dose clearly showed the effect of tube current
modulation: the dose strongly increased in the shoulder region. The integral dose of the low-dose CT was on
average 11 times lower than the standard-dose CT, partly due to the lower exposure and partly to the lower
scan length. The integral low-dose CT dose was on average 3 times higher than the one of the CBCT Head
protocol and 9 times higher than the one of the CBCT Image Gently protocol. The dose was highest in the
bones due to the importance of the photoelectric effect at kV energies and reached up to 65 mGy for a single
standard-dose CT.

3.2. Secondary cancer risk

We evaluated the trends in the total EAR when reducing the setup margin. We consider the example case of
low-dose CT-guided adaptive proton therapy, as this has been implemented clinically for selected patients

8
Phys. Med. Biol. 69 (2024) 225004 A Smolders et al

Figure 8. Voxel-wise imaging dose from the calibrated Monte Carlo simulation for the CT and CBCT protocols for an example
patient. Note that the color scaling for the standard-dose CT is 4 times larger than for the others.

Figure 9. Relative difference in total excess absolute risk between the low-dose CT-guided adaptive (EARAPT ) and baseline
(EARBL ) treatment protocols for various levels of setup margin reduction with adaptive therapy for all five patients. Trends are
shown individually for the full (full line), bell (dashed line), and plateau models (dotted line) models..

(Albertini et al 2024), but the analysis of other treatment protocols is similar. Without reducing the setup
margin, i.e. with 4 mm setup margin, EAR of the adaptive treatment was larger than for the baseline because
of the daily CTs (figure 9). Furthermore, without setup margin reduction, the difference between the full,
bell, and plateau models was very small. This is because the only difference between the total doses of the two
protocols was the daily CT dose. The EAR increase was therefore mainly due to a small dose increase in a
large, low-dose region. Since all models approximately behave linearly at low dose and use the same β EAR , the
differences between them were small.
The adaptive treatment’s EAR decreased with decreasing setup margin for 4 out of 5 patients and the
differences between the models increased. This larger spread was due to the change in setup margin mainly
affecting the high-dose region, where the models yield different results. With a setup margin reduction larger
than 0.26 mm, the adaptive EAR was below the baseline in 50% of all the patient-model combinations
(figure 9 and table 4). Reducing the setup margin by more than 0.5 mm resulted for 4 out of 5 patients in a
lower EAR for the adaptive treatment for all models.
For patient 4, however, more than 2 mm reduction in setup margin was necessary in the adaptive
treatment to reach an equal EAR as the baseline. This difference was not due to the imaging dose, but due to
a different trend in EAR when reducing robustness. This can be understood by looking at the contributions

9
Phys. Med. Biol. 69 (2024) 225004 A Smolders et al

Figure 10. Difference in organ-specific (and total) excess absolute risk between the low-dose CT-guided adaptive (EARAPT ) and
baseline (EARBL ) treatments for various levels of setup margin reduction with adaptive therapy for all five patients. Only the full
model is shown for clarity, but the other models had similar trends.

Figure 11. Total absorbed dose and corresponding risk equivalent dose using the secondary cancer risk parameters of the full
model for the salivary glands (contoured in magenta) for patient 3. The setup margin in this plan is 4 mm.

of the organ-specific EAR to the total EAR (figure 10). For all patients, the total EAR mainly reduced with
setup margin because of a decrease in EAR in the brain and CNS. Additionally, the EAR in the oral cavity
contributed substantially to the total EAR, but its trend with reducing setup margin varied: for some
patients, the EAR in the oral cavity increased, whereas for others, it decreased. This occurred because, for
some organs like salivary glands and oral cavity, the EAR peaks at intermediate doses and decreases at high
doses due to cell killing. For example, in the salivary glands, the EAR peaks around 10 Gy (figure 6). For
patient 3, reducing the setup margin reduced the dose in a part of the salivary gland where it was above
10 Gy, effectively decreasing cell killing and increasing the risk of radiation-induced cancer there. This
observation can be appreciated from the low risk equivalent dose in the high-dose regions (figure 11). The
plan in figure 11 uses a setup margin of 4 mm. If a smaller margin were used, the red areas with high-risk
equivalent dose move towards the salivary glands, increasing the radiation-induced cancer risk. For patient 4,
a similar effect happened in the oral cavity, and it was stronger than the reduction in EAR in the other
organs, leading to a net increase in total EAR when reducing the setup margin with 1 mm from the baseline.
Reducing the setup margin even more yielded a similar trend as for the other patients.
The RMR hinged strongly on assumptions about the type and frequency of imaging in IGRT and
adaptive treatment protocols (table 4). Independently of the model, the median RMR was below 0.6 mm for
all protocols, except for CT-guided adaptive therapy with standard-dose CT. Disregarding patient 4 and
standard-dose CT-guided adaptive therapy, the RMR was below 1 mm for all patients, protocols, and cancer
risk models. Interestingly, the RMR in low-dose CT-guided adaptive therapy was not substantially different
from the one in IGRT with weekly CT imaging, often used in current clinical practice. Furthermore, the
results showed a substantial advantage of acquiring weekly low-dose CTs instead of standard-dose CTs, both
for IGRT and CBCT-guided adaptive therapy. Lastly, simply limiting the FOV of the low-dose CT to the
tumor site reduced the median RMR by more than a third.

10
Phys. Med. Biol. 69 (2024) 225004 A Smolders et al

Table 4. Required margin reduction (RMR) for the IGRT and adaptive treatment protocols. The median RMR over all patients is
reported for each model individually.

Median RMR (mm)

Daily image Weekly image Full Bell Plateau

IGRT CBCT CT 0.31 0.34 0.42

IGRT CBCT Low-dose CT 0.08 0.09 0.09
Adaptive CBCT CT 0.38 0.42 0.51
Adaptive CBCT Low-dose CT 0.16 0.18 0.18
Adaptive Low-dose CT None 0.26 0.29 0.33
Adaptive Full-range low-dose CT None 0.41 0.45 0.58
Adaptive CT None 1.27 1.37 1.67

4. Discussion

In this work, we analyzed the risk of secondary cancer in image-guided and adaptive proton therapy for
head-and-neck cancer patients. We found that for most treatment protocols, patients and cancer risk models,
a reduction of setup robustness margin with 1 mm was sufficient to compensate the additional risk
introduced by daily imaging. By reducing the setup margin more, the secondary cancer risk in IGRT and
adaptive treatments would become considerably lower than in conventional treatments. Alongside other risk
reductions for more acute side effects (Lalonde et al 2023), this shows the advantage of treatments with
reduced setup margins.
The required setup margin reduction however strongly depended on the assumptions about the protocol
and frequency of CT and CBCT imaging, both for IGRT and adaptive treatment protocol. For example,
compared to an IGRT treatment with weekly standard-dose CTs, the RMR was lowered by a factor of 4 by
simply acquiring weekly low-dose CTs, while adaptation based on daily standard-dose CTs would increase
the RMR by a factor of 4. Since both the imaging protocols and frequency vary between clinics (Nabavizadeh
et al 2016, Kearney et al 2020), the RMR values found here cannot be taken as a reference, but should be
re-evaluated by each clinic based on their imaging protocols. The framework presented here can be used for
such evaluations, also when looking at other treatment sites, treatment modalities, or imaging devices.
This work focused on the development and validation of the CT and CBCT Monte carlo models and on
the applicability of cancer risk models to compare the trade-off between imaging dose and target robustness.
We focused on a relatively small patient cohort, which purposely had similar indications and treatment sites.
Nevertheless, besides its dependency on imaging type and frequency, the RMR also depended on the patient.
More specifically, it depended on how the setup margin reduction affected the specific dose levels in the
several organs, which depends on factors like tumor size, shape, and location, but also on the treatment plan
and its optimization functions. It would be interesting to establish which factors are most important and
how they affect the RMR, but this was out of the scope of the current study.
The complex patient and treatment plan dependency of the RMR implies that it could be interesting to
take into account radiation-induced cancer risk when deciding to treat a patient with image guidance or
adaptive therapy. Indeed, in this cohort, there was one patient who needed a much larger setup margin
reduction than the others, so this patient would benefit less from image guidance or adaptation than the
others in terms of radiation-induced cancer risk. Since the risk can be calculated solely with the planning CT
and treatment plan, the secondary cancer risk could be evaluated prospectively for each patient and
compared to the risk in a conventional treatment. In such a comparison, it is important to include also the
imaging of the conventional treatment, e.g. for offline replanning, instead of the best-case scenario as used in
this work. Together with other normal tissue complication models, this difference could be used in a
model-based selection system (Langendijk et al 2013) for adaptive therapy patients. Analogously, cancer risk
could be taken into account when deciding to treat with protons or photons.
A limitation of our framework is that it depends on organ-specific models, but such models do not exist
for all organs and cancer types. We considered all organs with fitted parameters within the FOV of the
planning CT, but no parameters were available for many other organs, e.g. the eyes. Their risk for
radiation-induced cancer was therefore ignored, even though it could affect the results we found in this
study. Our results however also show that it would not be advisable to use non-organ-specific models, as the
differences in cancer risk between organs were very large.
The cancer risk models are based on several assumptions (e.g. constant α/β, linear dose-effect relation at
low dose,...). A full discussion about these assumptions and their effect on the models is beyond the scope of
this paper, and the reader is referred to Schneider et al (2011) for an improved understanding. Some
uncertainty about the results was alleviated by using three models for cell repopulation, including the

11
Phys. Med. Biol. 69 (2024) 225004 A Smolders et al

extreme cases. However, other uncertainties could not be eliminated, and our results should be interpreted
carefully.
Another limitation of the current work is that it ignores the contribution of neutrons in the therapy dose.
For IMPT, neutron dose mainly originates from nuclear interactions within the patient, and less from nozzle
scatter (Hälg and Schneider 2020). Whereas this dose could in principle be accounted for using full Monte
Carlo simulations, the dose engine in Raystation does not track neutron transport. Preliminary simulations
with TOPAS showed that the neutron dose from nuclear interactions was more than a factor 10 below the
imaging dose within the primary tumor region, and even lower in the rest of the FOV of the CT. This means
ignoring neutron contribution is likely valid within the FOV of the CT. Outside the CT’s FOV, these scattered
neutrons are likely the main contribution to the dose. The scattered neutron dose is proportional to the total
monitor unit count (Hälg and Schneider 2020), and since these doses are low, a linear dose-effect relation is
valid. Therefore, also the EAR in organs outside the FOV of the CT would be proportional to the total
monitor unit count, which drops approximately 10% when reducing the setup margin from 4 to 0 mm for
our patients. Still, these neutron doses are very low, and their contribution to the total EAR is likely small
(Hälg and Schneider 2020).

5. Conclusion

In this work, we developed a framework to analyze the effects of target robustness and imaging dose on
radiation-induced cancer risk and applied it to image-guided and adaptive proton therapy for head-and-neck
cancer patients. For most patients and treatment protocols studied, we found that a 1 mm setup margin
reduction was sufficient to compensate the radiation-induced cancer risk from additional imaging. However,
we found that the required setup margin reduction strongly depended on imaging frequency, device and
protocol, which suggests that clinics should re-evaluate this reduction for their specific protocols.

Data availability statement

The data cannot be made publicly available upon publication because they contain sensitive personal
information. The data that support the findings of this study are available upon reasonable request from the
authors.

Acknowledgments

This project has received funding from the European Union’s Horizon 2020 Marie Skłodowska-Curie Actions
under Grant Agreement No. 955956, the European Union’s Horizon 2020 research and innovation program
under the Marie Skłodowska-Curie Grant Agreement No. 884104 (PSI-FELLOW-III-3i) and the Krebsliga
Grant (KFS-5660-08-2022). The authors would like to thank Siemens Healthineers for providing
information regarding the CT scanner. We would further like to express our gratitude to Claire-Louise
Chapple for her help in modeling the CT and CBCT scanners, and Uwe Schneider and Rosalind Perrin for
their help in the cancer risk modeling.

Ethical statement

The data used in this work was based on patients previously treated at the Center for Proton Therapy. All
patients gave consent to the use of their data for research, as standardly requested in our protocols. No new
patient data has been acquired for this study.

12
Phys. Med. Biol. 69 (2024) 225004 A Smolders et al

Appendix. Secondary cancer risk model parameters

Table A1. Parameters for the radiation-induced cancer risk models considered in this work. β EAR is the initial slope of the dose-effect
relation, R the repopulation parameter, and α the cell killing parameter of the linear-quadratic model. The values for thyroid and
esophagus were taken from D’Arienzo et al (2012), Cella et al (2013), Murray et al (2015), all others from Schneider et al (2011).

Organ Model β EAR R (-) α (Gy−1 )

Mouth and pharynx Full 0.73 0.97 0.043

Bell 0.73 0 0.017
Plateau 0.73 1 0.045

Esophagus Full 3.20 0.5 0.46

Bell 3.20 n/a n/a
Plateau 3.20 n/a n/a

Brain and CNS Full 0.70 0.93 0.018

Bell 0.70 0 0.009
Plateau 0.70 1 0.021

Thyroid Full 0.40 n/a n/a

Bell 0.40 0 0.0318
Plateau 0.40 n/a n/a

Salivary gland Full 0.73 0.23 0.087

Bell 0.73 0 0.059
Plateau 0.73 1 0.282

Bone (sarcoma) Intermediate R (full) 0.2 0.5 0.067

Low R (bell) 1.7 0.1 0.019
High R (plateau) 0.1 1 0.078

Soft tissue (sarcoma) Intermediate R (full) 0.6 0.5 0.06

Low R (bell) 3.3 0.1 0.04
High R (plateau) 0.35 1 0.093

ORCID iDs

A Smolders  https://orcid.org/0000-0003-2874-3634
F Albertini  https://orcid.org/0000-0002-2547-1269

References
Abuhaimed A, Martin C J, Sankaralingam M, Gentle J D and McJury M 2014 An assessment of the efficiency of methods for
measurement of the computed tomography dose index (CTDI) for cone beam (CBCT) dosimetry by Monte Carlo simulation
Phys. Med. Biol. 59 6307–26
Albertini F et al 2024 First clinical implementation of a highly efficient daily online adapted proton therapy (DAPT) workflow Phys.
Med. Biol. 69 215030
Albertini F, Matter M, Nenoff L, Zhang Y and Lomax A 2020 Online daily adaptive proton therapy Br. J. Radiol. 93 20190594
Boone J M 2010 Method for evaluating bow tie filter angle-dependent attenuation in CT: theory and simulation results Med. Phys.
37 40–48
Brodin N P, Kabarriti R, Garg M K, Guha C and Tomé W A 2018 Systematic review of normal tissue complication models relevant to
standard fractionation radiation therapy of the head and neck region published after the QUANTEC reports Int. J. Radiat. Oncol.
Biol. Phys. 100 391–407
Buckley J G, Wilkinson D, Malaroda A and Metcalfe P 2018 Investigation of the radiation dose from cone-beam CT for image-guided
radiotherapy: a comparison of methodologies J. Appl. Clin. Med. Phys. 19 174–83
Byrne M et al 2022 Varian ethos online adaptive radiotherapy for prostate cancer: early results of contouring accuracy, treatment plan
quality and treatment time J. Appl. Clin. Med. Phys. 23 e13479
Cella L, Conson M, Pressello M C, Molinelli S, Schneider U, Donato V, Orecchia R, Salvatore M and Pacelli R 2013 Hodgkin’s lymphoma
emerging radiation treatment techniques: trade-offs between late radio-induced toxicities and secondary malignant neoplasms
Radiat. Oncol. 8 22
Chen G T Y, Sharp G C and Mori S 2009 A review of image-guided radiotherapy Radiol. Phys. Technol. 2 1–12
D’Arienzo M, Masciullo S G, Sanctis V D, Osti M F, Chiacchiararelli L and Enrici R M 2012 Integral dose and radiation-induced
secondary malignancies: comparison between stereotactic body radiation therapy and three-dimensional conformal radiotherapy
Int. J. Environ. Res. Public Health 9 4223–40
Dawson L A, Eccles C, Bissonnette J-P and Brock K K 2005 Accuracy of daily image guidance for hypofractionated liver radiotherapy
with active breathing control Int. J. Radiat. Oncol. Biol. Phys. 62 1247–52

13
Phys. Med. Biol. 69 (2024) 225004 A Smolders et al

De Ruysscher D, Niedermann G, Burnet N G, Siva S, Lee A W M and Hegi-Johnson F 2019 Radiotherapy toxicity Nat. Rev. Disease
Primers 5 13
Duetschler A, Winterhalter C, Meier G, Safai S, Weber D C, Lomax A J and Zhang Y 2023 A fast analytical dose calculation approach for
MRI-guided proton therapy Phys. Med. Biol. 68 195020
Hälg R A and Schneider U 2020 Neutron dose and its measurement in proton therapy-current uedge Br. J. Radiol. 93 20190412
Hong T S, Tomé W A, Chappell R J, Chinnaiyan P, Mehta M P and Harari P M 2005 The impact of daily setup variations on
head-and-neck intensity-modulated radiation therapy Int. J. Radiat. Oncol. Biol. Phys. 61 779–88
Jan S et al 2011 GATE V6: a major enhancement of the GATE simulation platform enabling modelling of CT and radiotherapy Phys.
Med. Biol. 56 881–901
Kalra M K, Maher M M, Toth T L, Schmidt B, Westerman B L, Morgan H T and Saini S 2004 Techniques and applications of automatic
tube current modulation for CT Radiology 233 649–57
Kearney M, Coffey M and Leong A 2020 A review of image guided radiation therapy in head and neck cancer from 2009-2019 - best
practice recommendations for RTTs in the clinic Tech. Innov. Patient Support Radiat. Oncol. 14 43–50
Lalonde A, Bobić M, Sharp G C, Chamseddine I, Winey B and Paganetti H 2023 Evaluating the effect of setup uncertainty reduction and
adaptation to geometric changes on normal tissue complication probability using online adaptive head and neck intensity
modulated proton therapy Phys. Med. Biol. 68 115018
Langendijk J A, Lambin P, De Ruysscher D, Widder J, Bos M and Verheij M 2013 Selection of patients for radiotherapy with protons
aiming at reduction of side effects: the model-based approach Radiother. Oncol. 107 267–73
Lim-Reinders S, Keller B M, Al-Ward S, Sahgal A and Kim A 2017 Online adaptive radiation therapy Int. J. Radiat. Oncol. Biol. Phys.
99 994–1003
Maund I F, Benson R J, Fairfoul J, Cook J, Huddart R and Poynter A 2014 Image-guided radiotherapy of the prostate using daily CBCT:
the feasibility and likely benefit of implementing a margin reduction Br. J. Radiol. 87 20140459
Murray L, Sethugavalar B, Robertshaw H, Bayman E, Thomas E, Gilson D and Prestwich R J D 2015 Involved node, site, field and
residual volume radiotherapy for lymphoma: a comparison of organ at risk dosimetry and second malignancy risks Clin. Oncol.
27 401–10
Nabavizadeh N, Elliott D A, Chen Y, Kusano A S, Mitin T, Thomas C R and Holland J M 2016 Image guided radiation therapy (IGRT)
practice patterns and igrt’s impact on workflow and treatment planning: results from a national survey of american society for
radiation oncology members Int. J. Radiat. Oncol. Biol. Phys. 94 850–7
Nenoff L, Matter M, Charmillot M, Krier S, Uher K, Weber D C, Lomax A J and Albertini F 2021 Experimental validation of daily
adaptive proton therapy Phys. Med. Biol. 66 205010
Nesteruk K P, Bobic M, Sharp G C, Lalonde A, Winey B A, Nenoff L, Lomax A J and Paganetti H 2022 Low-dose computed tomography
scanning protocols for online adaptive proton therapy of head-and-neck cancers Cancers 14 5155
Nesteruk K P, Bobićbobić M, Lalonde A, Winey B A, Lomax A J and Paganetti H 2021 CT-on-rails versus in-room CBCT for online daily
adaptive proton therapy of head-and-neck cancers (available at: https://doi.org/10.3390/cancers13235991)
Oborn B M, Dowdell S, Metcalfe P E, Crozier S, Mohan R and Keall P J 2017 Future of medical physics: Real-time MRI-guided proton
therapy Med. Phys. 44 e77–e90
Paganetti H 2012 Range uncertainties in proton therapy and the role of Monte Carlo simulations Phys. Med. Biol. 57 R99
Paganetti H, Botas P, Sharp G C and Winey B 2021 Adaptive proton therapy Phys. Med. Biol. 66 22TR01
Poludniowski G, Omar A, Bujila R and Andreo P 2021 Technical Note: SpekPy v2.0-a software toolkit for modeling x-ray tube spectra
Med. Phys. 48 3630–7
Sarrut D et al 2022 The OpenGATE ecosystem for Monte Carlo simulation in medical physics Phys. Med. Biol. 67 184001
Schneider U 2009 Mechanistic model of radiation-induced cancer after fractionated radiotherapy using the linear-quadratic formula
Med. Phys. 36 1138–43
Schneider U, Sumila M and Robotka J 2011 Site-specific dose-response relationships for cancer induction from the combined Japanese
A-bomb and Hodgkin cohorts for doses relevant to radiotherapy Theor. Biol. Med. Modelling 8 27
Schneider W, Bortfeld T and Schlegel W 2000 Correlation between CT numbers and tissue parameters needed for Monte Carlo
simulations of clinical dose distributions Phys. Med. Biol. 45 459–78
Tahiri M, Mkimel M, Benameur Y, El Baydaoui R, Mesradi M R and El Rhazouani O 2021 Organ dose estimation for adult chest CT
examination using GATE Monte Carlo simulation Phys. Part. Nuclei Lett. 18 502–9
Varian Medical Systems 2020 ProBeam connect reference guide 4.1 Technical Report Palo-Alto,California,USA
Wang K and Tepper J E 2021 Radiation therapy-associated toxicity: etiology, management and prevention Cancer J. Clin. 71 437–54
Wasserthal J et al 2023 Totalsegmentator: robust segmentation of 104 anatomic structures in CT images Radiology 5 24
Winkel D et al 2019 Adaptive radiotherapy: the Elekta unity MR-linac concept Clin. Transl. Radiat. Oncol. 18 54–59

14