Physica Medica 127 (2024) 104837

Contents lists available at ScienceDirect

Physica Medica
journal homepage: www.elsevier.com/locate/ejmp

Influence of X-ray spectrum and bowtie filter characterisation on the

accuracy of Monte Carlo simulated organ doses: Validation in a whole-body
CT scanning mode
Gwenny Verfaillie a,*, Jeff Rutten a , Lore Dewulf a , Yves D’Asseler b,c, Klaus Bacher a
a
Department of Human Structure and Repair, Ghent University, Proeftuinstraat 86 – Building N7, 9000 Ghent, Belgium
b
Department of Nuclear Medicine, Ghent University Hospital, Corneel Heymanslaan 10, 9000 Ghent, Belgium
c
Department of Diagnostic Sciences, Ghent University, Corneel Heymanslaan 10, 9000 Ghent, Belgium

A R T I C L E I N F O A B S T R A C T

Keywords: Purpose: For patient-specific CT dosimetry, Monte Carlo dose simulations require an accurate description of the
Computed Tomography (CT) CT scanner. However, quantitative spectral information and information on the bowtie filter material and shape
Monte Carlo from the manufacturer is often not available. In this study, the influence of different X-ray spectra and bowtie
Patient-specific dosimetry
filter characterisation methods on simulated CT organ doses is studied.
Organ dose
Methods: Using ImpactMC, organ doses of whole-body CTs were simulated in twenty adult whole-body voxel
models, generated from PET/CT examinations previously conducted in these patients. Simulated CT organ doses
based on the manufacturer X-ray spectra and bowtie filter data were compared with those obtained using
alternative characterisation models, including spectrum generators and experimentally measured dose data. A
total of four different X-ray spectra and one bowtie filter model were defined based on these data.
Results: For all X-ray spectra and bowtie filter combinations, estimated CT organ doses are within 6% from those
resulting from simulations with the CT characterisation models provided by the manufacturer. While varying the
bowtie filter model results in CT organ dose differences smaller than 1%, dose differences up to 6% are observed
when X-ray spectra are not based on the quantitative data from the manufacturer.
Conclusions: Estimated organ doses slightly depend on the applied CT characterisation model. When manufac-
turer’s data are not available, half-value layer and dose measurements provide sufficient input to obtain
equivalent X-ray spectra and bowtie filter profiles, respectively.

1. Introduction volume CT dose index (CTDIvol) can be scaled to incorporate the size of
the patient resulting in a size-specific dose estimate (SSDE). Neverthe-
Over the past decades, the use of computed tomography (CT) has less, accurate individual organ dose estimations to assess potential ra-
increased significantly. Due to new techniques, protocols and technol- diation risks are needed. For this purpose, easy-to-use dose calculation
ogies its application exceeded beyond diagnostic imaging towards tools such as CT-Expo [5], WAZA-ARI [6], VirtualDose [7] and NCICT
screening for lung and colon cancer, and minimally invasive procedures. [8] were developed. However, they are often limited in the number of
In addition, the use of CT in hybrid nuclear medicine imaging (PET/CT available phantoms or lack accurate implementation of automatic tube
and SPECT/CT) is growing as well. This widespread use for different current modulation. CT-Expo, for instance, only employs mathematical
clinical purposes together with the growing concern about the long-term phantoms, including the adult Adam and Eva phantom, a child and a
effects of radiation exposure, especially the risk of cancer, leads to the baby phantom. Although WAZA-ARI already uses voxel phantoms to
need for accurate dose estimations [1,2]. estimate organ doses, the adult male and female phantoms only repre-
Using the effective or water-equivalent diameter metric, introduced sent the average Japanese body type [9]. In addition, there are only four
by the AAPM Task Groups 204 and 220 [3,4], the CT dose indicator adult and five paediatric phantoms available for each gender.

* Corresponding author.
E-mail addresses: [email protected] (G. Verfaillie), [email protected] (J. Rutten), [email protected] (L. Dewulf), [email protected]
(Y. D’Asseler), [email protected] (K. Bacher).

https://doi.org/10.1016/j.ejmp.2024.104837
Received 8 February 2024; Received in revised form 5 September 2024; Accepted 18 October 2024
Available online 25 October 2024
1120-1797/© 2024 Associazione Italiana di Fisica Medica e Sanitaria. Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).
G. Verfaillie et al. Physica Medica 127 (2024) 104837

VirtualDose, on the other hand, includes 25 virtual patients in total. whole-body PET/CT examination on a 40-slice Siemens Biograph mCT
These include a set of male and female adult voxel phantoms of various Flow (Siemens Healthineers, Germany), were collected retrospectively
heights and weights, paediatric male and female phantoms at five ages, from the institutional Picture Archiving and Communication System
the RPI adult male and female phantom and pregnant females at three (PACS). The ten male and female patients were chosen in such a way as
gestational stages. Meanwhile, NCICT incorporates a library of 351 to assure a wide variety in Body Mass Index (BMI) (Table 1). To be
paediatric and adult male and female voxel phantoms of various heights suitable for accurate dose estimations, the reconstructed Field of View
and weights (paediatric: 85 male and 73 female; adult: 100 male and 93 (FOV) of the CT scans included the entire cross-section of the patient. All
female), the ICRP reference paediatric and adult phantoms, and eight CT data was anonymised according to the hospitals anonymization
pregnant phantoms containing detailed foetus models at various gesta- policy to comply with the current General Data Protection Regulation
tions. Automatic tube current modulation can be activated in all four (GDPR) rules. Patient-related information, represented by unique iden-
tools. However, its implementation can be different. In CT-Expo it is only tifiers (tags), in the DICOM header of the images was thus completely
available for adults while it has only recently been made accessible in removed or replaced by de-identified information. Only data concerning
NCICT. Nevertheless, organ sizes and positions differ from patient to patient sex, age, length and weight was kept. The retrospective use of the
patient making it, even with a large library of voxel phantoms, impos- CT images was approved by the institutional ethical committee.
sible to estimate patient-specific organ doses. Therefore, dedicated Based on the data of the 512 x 512 DICOM images, a patient-specific
Monte Carlo (MC) frameworks using patient-specific voxel geometries 3D whole-body voxel model was created for each patient with 0.9727 x
were established. These individualised 3D voxel models can be created 0.9727 x 3 mm3 voxel size.
based on clinically available CT data of the patient.
Monte Carlo frameworks allow, next to the implementation of
patient-specific anatomical models, an accurate description of the X-ray 2.2. Monte Carlo dose simulations
modality. For CT examinations, characterisation of the CT scanner in-
cludes describing the geometrical, spectral and shaped filter character- To estimate patient-specific CT organ doses, Monte Carlo (MC)
istics. The necessary geometrical information can easily be found in the simulations were performed with ImpactMC 1.6 (CT Imaging GmbH,
technical reference manual of the system. However, to model the X-ray Erlangen Germany). This patient-specific dose calculation tool combines
spectrum and bowtie filter the situation is different. Ideally, quantitative Monte Carlo algorithms with scanner specific parameters such as geo-
data is provided by the manufacturer. This means the number of photons metric, spectral and shaped filter characteristics, and patient-specific
at each energy level for the X-ray spectrum, while for the bowtie filter voxel models based on patient CT images. In this way, the software
the varying thicknesses with increasing fan angle position are described calculates individualised 3D dose distributions, considering all relevant
for each material out of which the bowtie filter is made up. Based on photon interaction processes [16,17]. To calibrate the simulation soft-
non-disclosure agreements, some manufacturers also provide this ware, the air kerma free-in-air in the isocenter of the CT was measured
quantitative X-ray spectrum and bowtie filter information. Nevertheless, using a pencil beam ionisation chamber (Model 10X6-3CT, Radcal
in most cases, these data are not available. Fortunately, other method- Corporation, USA). The Monte Carlo software ImpactMC was validated
ologies exist to determine X-ray spectra and model shaped filters. by several research groups. Schmidt et al. [18], Deak et al. [17], Myr-
Research of for example Tucker et al. [10] and Poludniowski et al. [11] onakis et al. [19] and Chen et al. [16] all validated the software based on
resulted into generators to create an artificial X-ray spectrum based on the comparison of simulated and measured CTDI values using either the
information about the tube potential, anode angle and amount of IEC CT body and/or head dosimetry phantom. While Schmidt et al. [18]
filtration. The latter may be specified by the manufacturer separately for also compared their results with those obtained from Monte Carlo pro-
all inherent tube filtration present after signing a non-disclosure grams based on the EGS4 platform and published values, Deak et al. [17]
agreement. On the other hand, Turner et al. [12] presented an equiva- and Myronakis et al. [19] also performed validation measurements
lent source model to describe the energy spectrum and filtration based using anthropomorphic phantoms of various sizes.
on half-value layer and bowtie filter profile measurements, respectively, In this study, whole-body CT examinations, from head to mid-thigh,
without the need of manufacturer’s data. Meanwhile Boone [13] and were simulated using the scan parameters of a diagnostic whole-body CT
McKenney et al. [14] developed the COBRA method to compute the on a Siemens Biograph mCT Flow. Helical scans were simulated at 120
angle-dependent bowtie filter attenuation and thickness while Kramer kV with a rotation time of 0.5 s, a beam collimation of 19.2 mm and a
et al. [15] created a mathematical bowtie model. pitch of 0.7 (Table 2). To integrate tube current modulation (TCM), the
Although newly developed methodologies to characterise the X-ray tube current value from the DICOM header of each reconstructed image
spectrum or bowtie filter of a CT scanner are evaluated by comparing the was extracted using an in-house developed Fiji/ImageJ macro. Because
accuracy of CTDI simulations with measured CTDI values, their results of the TCM system available on the simulated CT scanner (CARE
are almost never compared with those obtained when quantitative Dose4D), each tube current value is the average of the angularly and
manufacturer’s data is used instead. In addition, when simulation results longitudinally modulated values applied over the gantry rotation used to
were compared, this was done using the IEC CT dosimetry phantoms reconstruct this image [20–25]. To ensure the speed and accuracy of the
[12] or, in a rare case, using an anthropomorphic phantom [15]. Monte Carlo simulation, the number of interacting photons was chosen
Therefore, the purpose of this study was to estimate the influence of to be 1010 for all simulations. In order to convert the CT values of the
various X-ray spectrum and bowtie filter modelling techniques, input whole-body patient CT images into density values the standard
including those based on the use of quantitative manufacturer’s data, on conversion curve incorporated in the ImpactMC software was used [26].
Monte Carlo simulated CT organ doses. By using voxel models created The assumed relationship is shown in Fig. 1.
based on clinical whole-body (WB) CT images, the accuracy of patient-
specific CT organ doses obtained through Monte Carlo simulations per- Table 1
formed with X-ray spectrum and bowtie filter models that differ from the Summary of mean (minimum – maximum) age, length, weight and BMI of the
quantitative models from the manufacturer was studied. study population.
Study Age Length (m) Weight (kg) BMI (kg/
2. Materials and methods population (years) m2)

10 females 64 (25 – 1.63 (1.50 – 63 (40 – 87) 24 (16 –

2.1. Patients and voxel models 86) 1.70) 34)
10 males 65 (45 – 1.77 (1.62 – 75 (51 – 24 (16 –
79) 1.93) 105) 33)
Whole-body CT images of twenty adult patients, acquired during a

2
G. Verfaillie et al. Physica Medica 127 (2024) 104837

Table 2 schematic overview of these five models and the data that was used to
Summary of exposure parameters for a diagnostic whole-body CT obtain them is shown in Fig. 2. Subsequent paragraphs explain this in
at a Siemens Biograph mCT Flow PET/CT. more detail.
Parameter Siemens
whole-body CT 2.3.2.1. Quantitative spectral information from the manufacturer. The
Tube voltage (kV) 120 first X-ray model was based on the quantitative spectral information for
Tube current (mA) ATCM* 120 kV provided by the manufacturer which was specified as the
Rotation time (s) 0.5 number of photons at each energy level (Fig. 2 – X-ray spectrum model
Pitch 0.7
Beam collimation (mm) 19.2
(1)). Because normalisation of the number of photons is done by the
Scan FOV (mm) 500 Monte Carlo software, if necessary, the provided spectral information
Scan start head could be directly used as input for the dose simulations. A graphical
Scan end mid-thigh visualisation of the spectrum is shown in Fig. 3.
* Automatic Tube Current Modulation
2.3.2.2. Spectrum generators. Secondly, two artificial X-ray spectra were
created using so called spectrum generators. For the first model the
ImpactMC integrated spectrum generator (based on work of Tucker et al.
[10]) was used, while for the second spectrum SpekCalc (based on work
of Poludniowski et al. [11]) was applied. In both tools, the user needs to
select the tube potential, the anode angle and the amount of filtration in
mm (Fig. 2 – X-ray spectrum model (4) and (5)). The latter can be
defined for different materials separately. In this study, 120 kV was
selected as tube potential. Information on the anode angle and amount
of filtration (all materials and their corresponding thicknesses) of the X-
ray tube was provided by the manufacturer. The resulting spectrum
models are visualised in Fig. 3. Due to a non-disclosure agreement we
cannot disclose the specific details about the anode angle and amount of
filtration. However, some of this information can be found in the tech-
nical specifications of the CT scanner. Nevertheless, the inherent tube
filtration will often be described in equivalent aluminium thickness
instead of the thickness of each separate filter material.

2.3.2.3. Half-value layer measurements. Finally, the methodology as

Fig. 1. Standard conversion from CT values in Hounsfield Units (HU) to density described by Turner et al. [12] for equivalent energy spectra in CT was
values in g/cm3 as performed by the ImpactMC software (from the ImpactMC used. Because it only requires physical measurements and calculations
user guide [26]). no information from the manufacturer is needed. First, the half-value
layer of the 120 kV spectrum was experimentally derived. To do this,
2.3. CT scanner characterisation a calibrated pencil beam ionisation chamber (Model 10X6-3CT, Radcal
Corporation, USA) was placed free-in-air at the isocenter of the CT. Thin
For scanner-specific Monte Carlo (MC) dose computations, infor- aluminium slabs with a thickness of 2, 1 and 0.5 mm were then stepwise
mation on the geometric, spectral and shaped filter characteristics of the added to the beam path, until the measured air kerma was less than half
CT scanner are needed as input parameters. This section describes how the initial air kerma measured without extra aluminium. In this way, the
these characteristics were determined to model the CT part of a Siemens half-value layer was determined as the amount of aluminium needed to
Biograph mCT Flow PET/CT. halve the initial air kerma. The measurements were carried out at a fixed
tube current–time product. To keep the X-ray tube stationary, mea-
2.3.1. Geometric specifications surements were performed in the 2D projection mode. Because CT couch
The geometrical specifications, such as the focus to isocenter dis- movement is inevitable in this scan mode, the ionisation chamber was
tance (595 mm) and fan angle (0.7955 rad), were derived from the placed in the isocenter of the CT using a tripod positioned at the bottom
specific data elements, DICOM tags, in the DICOM header of the CT of the gantry. In this study, table movement in the field of view could be
images. However, they could be extracted from the technical reference avoided. As a result the X-ray tube was positioned at the bottom and the
manual of the system as well. aluminium slabs could simply be placed on the gantry. However, if this
is not possible, measurements can also be performed in lateral 2D pro-
2.3.2. X-ray spectrum determination jection mode. The experimental set-up is illustrated in Fig. 4.
An important input parameter for Monte Carlo dose simulations is Secondly, an equivalent spectrum was generated with an in-house
the spectrum of the X-ray beam, which is determined by its tube po- developed MATLAB code (Mathworks, USA) with added SPEKTR tool
tential and the first half-value layer (HVL). The ImpactMC software [27]. The code started from a soft spectrum and iteratively added layers
expects a text file wherein the spectrum is specified as the number of of aluminium until the difference between the simulated and measured
photons at each energy level [26]. If necessary, the number of photons, half-value layer was minimal. The resulting spectrum was binned in 1
binned in 1 keV steps, will be normalised by the software. However, to keV steps to be used in a Monte Carlo simulation. A flowchart of this
specify the X-ray beam spectrum different possibilities exist. Which iterative process is shown in Fig. 5.
method can be used depends on the data made available by the manu- Next to the tube potential and the amount of added filtration, the
facturer. This has an impact on the accuracy of simulated CT organ SPEKTR tool also expects the ripple factor (percentage voltage ripple) as
doses. Therefore, in this study, five X-ray beam spectra models were input parameter for the generation of equivalent energy spectra. Turner
generated using different methodologies for the tube voltage of 120 kV, et al. [12] used a 25% voltage ripple, while the study of Yang et al. [28]
which is the standard tube voltage for diagnostic whole-body CT scans at started with a ripple factor of 0%. In this study, equivalent energy spectra
the Siemens Biograph mCT Flow PET/CT as described before. A for CT were created applying a voltage ripple of both 0% and 25% (Fig. 2

3
G. Verfaillie et al. Physica Medica 127 (2024) 104837

Fig. 2. Schematic overview of the five created X-ray beam spectrum models: (1) – X-ray spectrum model as provided quantitatively by the manufacturer; (2) and (3)
– equivalent energy spectrum based on half-value layer measurements and created with the SPEKTR tool applying a 0% and 25% voltage ripple, respectively; (4) and
(5) – X-ray spectrum model generated using the ImpactMC integrated spectrum generator and SpekCalc, respectively, based on the anode angle and filtration data
provided by the manufacturer.

Fig. 3. Graphical overview of the X-ray spectrum models for a tube voltage of 120 kV. The X-ray spectra models were, respectively, provided quantitatively by the
manufacturer, created using the SPEKTR tool applying a 0 % or 25 % voltage ripple after half-value layer measurements, and generated with the ImpactMC inte-
grated spectrum generator or SpekCalc based on the anode angle and filtration data provided by the manufacturer.

− X-ray spectrum model (2) and (3)). These two resulting X-ray spec- 2.3.4. Impact of CT scanner characterisation model on estimated organ
trum models are given in Fig. 3 together with the other three models. doses
To evaluate the effect of the X-ray spectrum and bowtie filter char-
2.3.3. Shaped bowtie filter model acterisation model on estimated CT organ doses, each of the previously
To characterise the shaped filter, two bowtie filter models were determined models was used in the Monte Carlo simulations. Because
created. The first model was based on data provided by the manufac- five X-ray spectrum and two bowtie filter models were defined, this
turer, which defined the attenuation of the bowtie (w.r.t. the detector means that ten Monte Carlo dose simulations were performed for each of
signal) as a function of the fan angle. This information was converted to the twenty patients included in this study. An overview of these different
an input file suitable for the Monte Carlo software. scenarios is given in Table 3.
The second model characterised the bowtie filter based on dose
measurements performed free-in-air. For this purpose, a calibrated 2.4. Organ dose calculation
pencil beam ionisation chamber (Model 10X6-3CT, Radcal Corporation,
USA) was used. Dose measurements were incrementally obtained by 2.4.1. Delineation of organs
moving the ionisation chamber in 1 cm intervals from the isocenter To delineate the radiosensitive organs and tissues of interest, the
while keeping the X-ray tube stationary (Fig. 6). Since the bowtie filter is open source software tools Fiji/ImageJ [29,30] and 3D Slicer [31] were
symmetric, only one side of the bowtie filter needed to be defined used. For the lungs, bones (ribs/spine) and liver, the regions of interest
together with the focus to isocenter distance and the increment distance (ROIs) were obtained semi-automatically. The breast (female patients),
between the measurement points. However, due to uncertainties in heart, kidneys, thyroid and oesophagus on the other hand were delin-
positioning, dose measurements were performed in both the +x and − x eated manually. This was done by a medical physicist.
direction. The dose at each increment position was then calculated as the
mean of the measured dose values in both directions at the same dis- 2.4.2. Patient-specific organ doses
tance from the isocenter. A Monte Carlo dose calculation with ImpactMC results in a 3D dose
distribution based on the physical properties (i.e. attenuation, compo-
sition and size) of the input patient CT scan. Overlaying the contours of
each organ on the corresponding slices of the dose distribution results in

4
G. Verfaillie et al. Physica Medica 127 (2024) 104837

Fig. 4. Set-up of half-value layer measurement (based on Turner et al. [12]).

The X-ray tube is kept stationary at the 6o’clock position and a calibrated
ionisation chamber is placed at the isocenter of the CT. Aluminium slabs are
stepwise added to the beam path, by placing them on the gantry, until the
measured air kerma is less than half the initial air kerma measured without
extra aluminium slabs in the beam path. All CT exposures are performed at a
fixed tube current–time product.

an estimation of patient-specific organ doses DT which were determined

as follows:
N (
∑ ) Ai,T
DT = fi,T • Mi,T with fi,T = ∑N
i=1 i=1 Ai,T

where Mi,T is the mean dose within the contour at slice i of organ T, N the
total number of slices that contain contours of organ T and fi,T the
fractional area of each organ contour (with Ai,T the area within the
contour at slice i of organ T). To enable unsupervised organ dose
calculation, an algorithm was implemented in Fiji/ImageJ. An overview
of the complete workflow is given in Fig. 7.
Fig. 5. Flowchart of the iterative process for the generation of equivalent X-ray
spectra implemented in MATLAB: using the ‘spektrSpectrum’ function a soft
2.5. Comparison of organ dose estimations spectrum si is created for an energy E of 120 kV without extra aluminium (Al)
filtration and with a percentage voltage ripple of 0 % or 25 %. Secondly, 1 mm
For each patient and each Monte Carlo simulation, using a different of aluminium filtration is added and by using the ‘spektrBeers’ function a new
combination of X-ray spectrum and bowtie filter model, organ doses energy spectrum sp is created of which the first half-value layer HVLsim is
were calculated as described before. Next, the mean organ doses and determined with the ‘spektrHVLn’ function. The simulated HVL value is then
their corresponding standard deviations were determined for the study compared with the measured HVL value resulting from the ionisation chamber
measurements. As long as the simulated HVL is lower, 1 mm extra aluminium
population. For each organ, percentage dose differences were obtained
filtration is added and a new energy spectrum sp and corresponding HVLsim is
by comparing the doses to those obtained using the quantitative model
calculated. When the simulated HVL is larger, 1 mm of aluminium is removed
(s) provided by the manufacturer, which are assumed to be the ground and the iterative process is repeated subsequently for the addition of 0.1 mm,
truth. From these deviations the maximum value over all organs was 0.01 mm and 0.001 mm aluminium. Finally, an equivalent spectrum is created
determined for each studied situation. This was done to study the in- with a HVL that differs minimally from the measured HVL (E – energy (e.g. 120
fluence of both the used X-ray spectrum and bowtie filter model sepa- kV), Al – aluminium, HM – hardening material (e.g. aluminium), HVLsim –
rately and their combinations. simulated half-value layer, HVLmeasured – measured half-value layer).

3. Results the spectrum generated with SpekCalc was used. Applying the manu-
facturer’s bowtie filter model seemed to result in lower doses for most
The mean CT organ doses and corresponding standard deviations of organs.
the breast, heart, liver, lungs, kidneys, ribs, thyroid, oesophagus and
spine are shown in Fig. 8. As expected, deviations in organ doses were
observed when a different combination of X-ray spectrum and bowtie 3.1. Influence of bowtie filter model
filter model was used in the Monte Carlo simulation. For all organs, the
estimated organ doses were the smallest when the X-ray spectrum pro- For each organ, dose differences were calculated when using the
vided by the manufacturer was applied while they were the largest when bowtie filter model based on dose measurements instead of the model

5
G. Verfaillie et al. Physica Medica 127 (2024) 104837

Fig. 6. Set-up of CT bowtie profile dose measurements, with L the focus to

isocenter distance and li the distance between the measurement points (based
Fig. 7. Overview of the workflow – Whole-body CT images were used to create
on [12]). The X-ray tube is kept stationary at the 12 o’clock position. Dose
patient-specific 3D voxel models. Individual organs were delineated on the
measurements are performed using a calibrated ionisation chamber that is
original CT images. The organ segmentations were used to calculate mean
incrementally moved from the isocenter in +x and − x direction.
organ doses on the output images of the Monte Carlo simulations.

provided by the manufacturer, which is assumed to be the most accurate

4. Discussion
model. Fig. 9 shows the maximum percentage organ dose difference
over all organs for each X-ray spectrum model. Maximum dose differ-
Nowadays, Monte Carlo frameworks are the gold standard to
ences ranging from around 0.90% to 0.94% were observed.
perform patient-specific CT dosimetry. Besides allowing the imple-
mentation of anatomical 3D models, it also allows an accurate descrip-
3.2. Influence of X-ray spectrum model
tion of the CT scanner. Technical reference manuals already report
specifications regarding the X-ray tube such as the anode angle and
Fig. 10 presents the maximum percentage difference in calculated CT
material. The inherent tube filtration, on the other hand, is often spec-
organ dose between Monte Carlo simulations performed with and
ified in equivalent aluminium thickness at a certain tube voltage. Similar
without the X-ray spectrum model based on the quantitative data from
conclusions can be made for the bowtie filter. If reported at all, the
the manufacturer. Irrespective from the used bowtie filter model, organ
available bowtie filters are described in terms of quality equivalent
dose differences up to 6% were observed. For Monte Carlo simulations
filtration. However, information about the exact shape of the bowtie
performed with the equivalent X-ray spectrum with a 0% voltage ripple
filter, or the varying filter thickness with increasing fan angle, is mostly
a maximum difference of 2.4% was found.
missing. Specific details about the inherent filtration and bowtie filters,
including the used material(s) and respective thickness(es), are the
3.3. Influence of X-ray spectrum and bowtie filter model
manufacturer’s secret. Based on non-disclosure agreements, it is some-
times possible to obtain the necessary data to quantitatively model the
Estimated CT organ doses were within 6% from those resulting from
X-ray spectrum and bowtie filter for a certain CT scanner from a specific
Monte Carlo simulations with the CT characterisation models provided
manufacturer. Nevertheless, there are many CT scanners and gaining
by the manufacturer (Fig. 11). Disregarding all scenarios applying
manufacturer’s data is not always successful. Fortunately, methods exist
quantitative manufacturer’s data, organ dose differences are within 5%.
to model X-ray spectra and bowtie filters based on little or no pro-
Dose differences smaller than 3% were found when the equivalent en-
prietary data.
ergy spectrum applying a 0% voltage ripple is used.
As for Monte Carlo simulation tools, newly developed methodologies
to characterise the X-ray spectrum and bowtie filter of a CT scanner are
evaluated by comparing the accuracy of multiple CTDI simulations with

Table 3
Overview of the ten X-ray spectra and bowtie filter model combinations (scenarios).
X-ray spectrum and bowtie filter combination 120 kV X-ray spectrum Bowtie filter
model model

Sc 1a Manufacturer Manufacturer
Sc 1b Manufacturer Dose measurements
Sc 2a Equivalent – 0% kV ripple Manufacturer
Sc 2b Equivalent – 0% kV ripple Dose measurements
Sc 3a Equivalent – 25% kV ripple Manufacturer
Sc 3b Equivalent – 25% kV ripple Dose measurements
Sc 4a ImpactMC generator Manufacturer
Sc 4b ImpactMC generator Dose measurements
Sc 5a SpekCalc Manufacturer
Sc 5b SpekCalc Dose measurements

6
G. Verfaillie et al. Physica Medica 127 (2024) 104837

Fig. 8. Estimated mean organ doses of a diagnostic whole-body CT scan for each X-ray spectrum and bowtie filter model combination (spectrum: 1 –manufacturer, 2
– equivalent with 0% voltage ripple, 3 – equivalent with 25% voltage ripple, 4 – ImpactMC integrated generator, 5 – SpekCalc; bowtie filter: a – manufacturer, b –
dose measurements).

Fig. 10. Maximum percentage difference in calculated mean CT organ dose for
the different X-ray spectrum models compared to the X-ray spectrum model
provided by the manufacturer. Note that the assumption was made that the X-
Fig. 9. Maximum percentage difference in calculated mean CT organ dose ray spectrum from the manufacturer is the most accurate model. Results are
between the two bowtie filter models, for each X-ray spectrum model. Note that given for simulations performed with the bowtie filter model based on data
the assumption was made that the bowtie filter model provided by the manu- from the manufacturer and dose measurements, respectively.
facturer is the most accurate model.
model, organ doses of a whole-body CT scan were calculated for each
measured CTDI values. Depending on the specific study at hand, this is combination of five X-ray spectrum models and two bowtie filter
done for either or both standard IEC CT dosimetry phantoms [32]. models, including those based on quantitative manufacturer’s data. The
Turner et al. [12], Kramer et al. [15] and Belinato et al. [33] reported an observed standard deviations are related to the wide range of BMI in the
accuracy of approximately 95% while a deviation of even less than 4% study population and the use of automatic tube current modulation.
was observed by Adrien et al. [34]. Nonetheless, the number of studies For all organs, as expected, deviations in simulated CT organ doses
looking at the difference in simulated CT dose between Monte Carlo are observed when the X-ray spectrum and bowtie filter were modelled
simulations performed using their own X-ray spectrum and bowtie filter in a different way. However, looking at the order of magnitude of the
models and those obtained from the manufacturer is limited. Using the organ doses the impact is rather small. Modelling the bowtie filter based
IEC CT dosimetry phantoms, Turner et al. [12] found a difference of on dose measurements instead of using the one provided by the manu-
approximately 12% between the simulated CTDI values. Kramer et al. facturer leads to dose differences that are within 1%, irrespective of the
[15], on the other hand, showed a maximum deviation of about 6% applied X-ray spectrum model. This model is thus a very good alterna-
between CTDI values simulated using their own developed bowtie filter tive when no manufacturer’s data about the bowtie filter is available.
model and the proprietary data made available by the manufacturer. Considering only a variation in used X-ray spectrum determination
In this study, the accuracy of patient-specific organ doses obtained method, CT organ doses within 6% from those resulting from simula-
through Monte Carlo simulations applying different CT characterisation tions with the manufacturer spectrum are found. Even smaller de-
models was studied. Therefore, whole-body CT images of twenty adult viations, smaller than 4%, are observed when the spectrum is modelled
patients were used as anatomy-specific voxel models. For each patient based on the methodology described by Turner et al. [12]. Similar

7
G. Verfaillie et al. Physica Medica 127 (2024) 104837

scanner from another manufacturer. Although obtaining the necessary

proprietary data, even with a non-disclosure agreement, can take several
weeks to months it can thus also be completely unsuccessful. Further-
more, the influence of the X-ray spectrum model on the accuracy of CT
organ doses from Monte Carlo simulations was studied for one tube
voltage, namely 120 kV which is the standard in diagnostic whole-body
CT protocols in nuclear medicine applications. However, to obtain the X-
ray spectrum model for other CT tube voltages the same methodologies
can be used and similar results in accuracy are expected.
In order to implement patient-specific Monte Carlo dose simulations
within the clinic, some requirements need to be fulfilled. First, the CT
scanner needs to be characterised. As described before, the focus to
isocenter distance and fan angle can be found in the technical reference
manual or calculated from specific DICOM tags in the DICOM header of
the CT images. Modelling the X-ray spectrum and bowtie filter must be
done only once at acceptance of your device or, if preferred, whenever
an X-ray tube or a bowtie filter is replaced. In a few hours it is possible to
perform all necessary measurements to model the X-ray spectrum for
Fig. 11. Maximum percentage difference in calculated mean CT organ dose for each tube voltage and every available bowtie filter. At the same time, the
the different X-ray spectrum and bowtie filter combinations (Sc 1b – Sc 5b) air kerma free-in-air can be measured for each combination of tube
compared to scenario 1a (Sc 1a) that combines the X-ray spectrum and bowtie voltage, collimation and bowtie filter, which is needed to calibrate the
filter model provided by the manufacturer. Monte Carlo software. Secondly, to perform patient-specific MC simu-
lations of a CT scan, the CT scan parameters need to be given. The
results are seen for both bowtie filter models. In the end, it is important rotation time, pitch and applied beam collimation can be extracted from
to look at the combined effect of the applied bowtie filter and X-ray the examination protocols on the CT scanner. However, they can also
spectrum model. For all possible scenarios, organ doses are within 6% easily be extracted from the DICOM header of the images together with
from the most accurate simulations using the models based on the data the tube current values of each reconstructed image to integrate the tube
from the manufacturer. Disregarding all situations that make use of the current modulation. The time needed to run one Monte Carlo simulation
quantitative data provided by the manufacturer, the best results are depends on several factors. First of all, it depends on whether the
obtained by determining equivalent X-ray spectra with a voltage ripple computer on which the simulations run has a GPU or not. As can be
of 0%. Organ dose differences are then within 3%. Half-value layer and expected, simulations run faster when a GPU is available instead of only
dose measurements are thus good alternative methods to obtain equiv- a CPU. Also the generation of GPU plays a role. Next, the simulation time
alent X-ray spectra and bowtie filter profiles, respectively. Moreover, increases linearly with the number of simulated photons. Although the
these measurements can be performed without special equipment and relative error in estimated organ dose decreases with the number of
without the need to go into service mode. simulated photons, it is important to find a balance between the number
Similar to Fig. 9 and Fig. 10, Fig. 11 presents the maximum per- of simulated photons and a realistic simulation time as the decrease in
centage difference in calculated mean CT organ doses between different relative error eventually reaches a platform. Besides, the simulated scan
X-ray spectra and bowtie filter model combinations and the one length plays a role. The shorter the scan length, the faster the Monte
combining the models provided by the manufacturer. Looking in more Carlo simulation is done. Finally, organ segmentation may be very time
detail to the organ dose percentage differences for the scenarios that do consuming. However, the development of automatic segmentation tools,
not make use of any quantitative manufacturer’s data, the following such as TotalSegmentator [36], and advances in deep learning create a
ranges are observed: 1.20% to 2.81%, 1.58% to 4.43%, 3.26% to 4.61% lot of opportunities to speed up this process.
and 3.62% to 5.04% for scenarios 2b, 3b, 4b and 5b, respectively. The
lowest values are found for the liver, heart and thyroid and they are 5. Conclusions
within 0.5% of each other for each scenario. Also for the lungs a small
deviation from the observed minimal percentage difference (≤ 0.7%) is When manufacturer’s data are not available, half-value layer and
seen. For scenario 2b, 3b and 4b the largest percentage dose differences dose measurements, which can be performed without special equip-
are related to the ribs, while for scenario 5b the largest difference is ment, provide sufficient input to obtain equivalent X-ray spectra and
observed for the oesophagus. In general, these differences are quite bowtie filter profiles, respectively. Monte Carlo simulations then result
small especially when taking into account the order of magnitude of the in estimated CT organ doses that deviate less than 3% from the most
organ doses. accurate simulations.
To our knowledge only Kramer et al. [15] eventually compared CT
organ doses obtained through Monte Carlo simulation using their and 6. Author agreement
the manufacturer’s characterisation models for both the X-ray spectrum
and bowtie filter. Organ dose differences within 8% and 6% were found This article was setup in the framework of the MEDIRAD and
for, respectively, simulations of a CT abdomen and CT thorax on a SINFONIA projects. Data collection, Monte Carlo simulations and anal-
Philips Brilliance 64 CT scanner. Looking at the same organs as ysis of the results were performed by GV. Monte Carlo simulations and
segmented in our study, their observed minimal organ dose differences segmentations were partially performed by LD. The first draft was
were around 2.2% for both CT scans. These results are quite similar to written by GV and all authors commented on previous versions of the
those obtained in our research. However, the results of Kramer et al. manuscript. All authors read and approved the final manuscript.
[15] are calculated in the anthropomorphic MASH and FASH phantom
for, respectively, a CT thorax and a CT abdomen scan [35]. Ethical Approval
One of the limitations of this study is that the results are based on one
CT scanner from one manufacturer. Despite all attempts made, no The retrospective use of the CT images was approved by the medical
quantitative spectral and bowtie filter profile data was obtained for a CT ethics committee of Ghent University Hospital, Belgium (No. BC-06610
E01).

8
G. Verfaillie et al. Physica Medica 127 (2024) 104837

Funding [15] Kramer R, Cassola VF, Andrade MEA, de Araújo MWC, Brenner DJ, Khoury HJ.
Mathematical modelling of scanner-specific bowtie filters for Monte Carlo CT
dosimetry. Phys Med Biol 2017;62:781–809. https://doi.org/10.1088/1361-6560/
This study has received funding from the Euratom research and aa5343.
training programme 2014–2018 under grant agreement No. 755,523 [16] Chen W, Kolditz D, Beister M, Bohle R, Kalender WA. Fast on-site Monte Carlo tool
and the Euratom research and innovation programme 2019–2020 under for dose calculations in CT applications. Med Phys 2012;39:2985–96. https://doi.
org/10.1118/1.4711748.
grant agreement No. 945196. [17] Deak P, van Straten M, Shrimpton PC, Zankl M, Kalender WA. Validation of a
Monte Carlo tool for patient-specific dose simulations in multi-slice computed
tomography. Eur Radiol 2008;18:759–72. https://doi.org/10.1007/s00330-007-
Declaration of competing interest 0815-7.
[18] Schmidt B, Kalender WA. A fast voxel-based Monte Carlo method for scanner- and
patient-specific dose calculations in computed tomography. Phys Medica 2002;18:
The authors declare that they have no known competing financial 43–53.
interests or personal relationships that could have appeared to influence [19] Myronakis M, Perisinakis K, Tzedakis A, Gourtsoyianni S, Damilakis J. Evaluation
the work reported in this paper. of a patient-specific Monte Carlo software for CT dosimetry. Radiat Prot Dosim
2009;133:248–55. https://doi.org/10.1093/rpd/ncp051.
[20] Mulkens TH, Bellinck P, Baeyaert M, Ghysen D, Van Dijck X, Mussen E, et al. Use of
Acknowledgements an automatic exposure control mechanism for dose optimization in multi-detector
row CT examinations: Clinical evaluation. Radiology 2005;237:213–23. https://
doi.org/10.1148/radiol.2363041220.
The authors thank the Belgian team of Siemens Healthineers, Ger- [21] Rizzo S, Kalra M, Schmidt B, Dalal T, Suess C, Flohr T, et al. Comparison of angular
many, for their efforts to provide quantitative spectral and bowtie filter and combined automatic tube current modulation techniques with constant tube
data from the Siemens Biograph mCT Flow PET/CT. current CT of the abdomen and pelvis. Am J Roentgenol 2006;186:673–9. https://
doi.org/10.2214/Ajr.04.1513.
[22] Rego SL, Yu L, Bruesewitz MR, Vrieze TJ, Kofler JM, McCollough CH. CARE
References Dose4D CT automatic exposure control system. physics principles and practical
hints. Mayo Foundation for Medical Education and Research; 2008.
[1] Alsafi KG. Radiation Protection in X-Ray Computed Tomography: Literature [23] Soderberg M, Gunnarsson M. The effect of different adaptation strengths on image
Review. International Journal of Radialogy and Imaging. Technology 2016;2. quality and radiation dose using Siemens Care Dose 4D. Radiat Prot Dosim 2010;
[2] Franck C, Vandevoorde C, Goethals I, Smeets P, Achten E, Verstraete K, et al. The 139:173–9. https://doi.org/10.1093/rpd/ncq098.
role of Size-Specific Dose Estimate (SSDE) in patient-specific organ dose and cancer [24] Papadakis AE, Perisinakis K, Damilakis J. Automatic exposure control in CT: the
risk estimation in paediatric chest and abdominopelvic CT examinations. Eur effect of patient size, anatomical region and prescribed modulation strength on
Radiol 2016;26:2646–55. https://doi.org/10.1007/s00330-015-4091-7. tube current and image quality. Eur Radiol 2014;24:2520–31. https://doi.org/
[3] Aapm. Size-Specific Dose Estimates (SSDE) in pediatric and adult body ct 10.1007/s00330-014-3309-4.
examinations (Task Group 204). American Association of Physicists in Medicine [25] Soderberg M. Overview, Practicaltips and Potential Pitfalls of Using Automatic
2011. Exposure Control in Ct: Siemens Care Dose 4d. Radiat Prot Dosim 2016;169:84–91.
[4] Aapm. Use of water equivalent diameter for calculating patient size and Size- https://doi.org/10.1093/rpd/ncv459.
Specific Dose Estimate (SSDE) in CT (Task Group 220). American Association of [26] User Guide ImpactMC. CT Imaging GmbH; 2010-2016.
Physicists in Medicine 2014. [27] Siewerdsen JH, Waese AM, Moseley DJ, Richard S, Jaffray DA. Spektr: a
[5] Stamm G, Nagel HD. CT-Expo - a novel program for dose evaluation in CT. Rofo computational tool for x-ray spectral analysis and imaging system optimization.
2002;174:1570–6. https://doi.org/10.1055/s-2002-35937. Med Phys 2004;31:3057–67. https://doi.org/10.1118/1.1758350.
[6] Ban N, Takahashi F, Sato K, Endo A, Ono K, Hasegawa T, et al. Development of a [28] Yang K, Li XH, Xu XG, Liu B. Direct and fast measurement of CT beam filter profiles
web-based CT dose calculator: WAZA-ARI. Radiat Prot Dosimetry 2011;147:333–7. with simultaneous geometrical calibration. Med Phys 2017;44:57–70. https://doi.
https://doi.org/10.1093/rpd/ncr333. org/10.1002/mp.12024.
[7] Ding AP, Gao YM, Liu HK, Caracappa PF, Long DJ, Bolch WE, et al. VirtualDose: a [29] Schindelin J, Arganda-Carreras I, Frise E, Kaynig V, Longair M, Pietzsch T, et al.
software for reporting organ doses from CT for adult and pediatric patients. Phys Fiji: an open-source platform for biological-image analysis. Nat Methods 2012;9:
Med Biol 2015;60:5601–25. https://doi.org/10.1088/0031-9155/60/14/5601. 676–82. https://doi.org/10.1038/nmeth.2019.
[8] Lee C, Kim KP, Bolch WE, Moroz BE, Folio L. NCICT: a computational solution to [30] Schneider CA, Rasband WS, Eliceiri KW. NIH Image to ImageJ: 25 years of image
estimate organ doses for pediatric and adult patients undergoing CT scans. J Radiol analysis. Nat Methods 2012;9:671–5. https://doi.org/10.1038/nmeth.2089.
Prot 2015;35:891–909. https://doi.org/10.1088/0952-4746/35/4/891. [31] Fedorov A, Beichel R, Kalpathy-Cramer J, Finet J, Fillion-Robin JC, Pujol S, et al.
[9] Sato K, Noguchi H, Emoto Y, Koga S, Saito K. Japanese adult male voxel phantom 3D Slicer as an image computing platform for the Quantitative Imaging Network.
constructed on the basis of CT images. Radiat Prot Dosim 2007;123:337–44. Magn Reson Imaging 2012;30:1323–41. https://doi.org/10.1016/j.
https://doi.org/10.1093/rpd/ncl101. mri.2012.05.001.
[10] Tucker DM, Barnes GT, Chakraborty DP. Semiempirical model for generating [32] IEC. IEC 60601-2-44:2009+AMD1:2012+AMD2:2016 ed3.2 Medical electrical
tungsten target x-ray spectra. Med Phys 1991;18:211–8. https://doi.org/10.1118/ equipment - Part 2-44: Particular requirements for the basic safety and essential
1.596709. performance of X-ray equipment for computed tomography2016.
[11] Poludniowski G, Landry G, DeBlois F, Evans PM, Verhaegen F. SpekCalc: a program [33] Belinato W, Santos WS, Paschoal CMM, Souza DN. Monte Carlo simulations in
to calculate photon spectra from tungsten anode x-ray tubes. Phys Med Biol 2009; multi-detector CT (MDCT) for two PET/CT scanner models using MASH and FASH
54:N433. https://doi.org/10.1088/0031-9155/54/19/N01. adult phantoms. Nucl Instrum Meth A 2015;784:524–30. https://doi.org/10.1016/
[12] Turner AC, Zhang D, Kim HJ, DeMarco JJ, Cagnon CH, Angel E, et al. A method to j.nima.2014.09.036.
generate equivalent energy spectra and filtration models based on measurement for [34] Adrien C, Le Loirec C, Dreuil S, Bordy JM. A new Monte Carlo tool for organ dose
multidetector CT Monte Carlo dosimetry simulations. Med Phys 2009;36:2154–64. estimation in computed tomography. Radioprotection 2020;55:123–34. https://
https://doi.org/10.1118/1.3117683. doi.org/10.1051/radiopro/2020006.
[13] Boone JM. Method for evaluating bow tie filter angle-dependent attenuation in CT: [35] Cassola VF, Kramer R, Brayner C, Khoury HJ. Posture-specific phantoms
theory and simulation results. Med Phys 2010;37:40–8. https://doi.org/10.1118/ representing female and male adults in Monte Carlo-based simulations for
1.3264616. radiological protection. Phys Med Biol 2010;55:4399–430. https://doi.org/
[14] McKenney SE, Nosratieh A, Gelskey D, Yang K, Huang SY, Chen L, et al. 10.1088/0031-9155/55/15/014.
Experimental validation of a method characterizing bow tie filters in CT scanners [36] Wasserthal J, Breit HC, Meyer MT, Pradella M, Hinck D, Sauter AW, et al.
using a real-time dose probe. Med Phys 2011;38:1406–15. https://doi.org/ TotalSegmentator: robust segmentation of 104 anatomic structures in CT images.
10.1118/1.3551990. Radiol-Artif Intell 2023;5. ARTN e23002410.1148/ryai.230024.