Patient-specific Monte Carlo-based organ dose estimates in spiral CT via optical 3D body scanning and adaptation of a voxelized phantom dataset: proof-of-principle

Objective. We present a method for personalized organ dose estimates obtained before the computed tomography (CT) exam, via 3D optical body scanning and Monte Carlo (MC) simulations. Approach. A voxelized phantom is derived by adapting a reference phantom to the body size and shape measured with a portable 3D optical scanner, which returns the 3D silhouette of the patient. This was used as an external rigid envelope for incorporating a tailored version of the internal body anatomy derived from a phantom dataset (National Cancer Institute, NIH, USA) matched for gender, age, weight, and height. The proof-of-principle was conducted on adult head phantoms. The Geant4 MC code provided estimates of the organ doses from 3D absorbed dose maps in the voxelized body phantom. Main results. We applied this approach for head CT scanning using an anthropomorphic voxelized head phantom derived from 3D optical scans of manikins. We compared the estimates of head organ doses with those provided by the NCICT 3.0 software (NCI, NIH, USA). Head organ doses differed up to 38% using the proposed personalized estimate and MC code, with respect to corresponding estimates calculated for the standard (non-personalized) reference head phantom. Preliminary application of the MC code to chest CT scans is shown. Real-time pre-exam personalized CT dosimetry is envisaged with adoption of a Graphics Processing Unit-based fast MC code. Significance. The developed procedure for personalized organ dose estimates before the CT exam, introduces a new approach for realistic description of size and shape of patients via voxelized phantoms specific for each patient.


Introduction
Estimation of radiation dose from radiological procedures in patients is a fundamental requirement for the related assessment of radiation exposure in the population. A recent report (based on the NCRP report 184, NCRP 2019) investigates this issue in the United States population for the period 2006-2016, showing that 74 million computed tomography (CT) scans in 2016 determined an average individual effective dose of 1.37 mSv, out of a total effective dose of 2.2 mSv from all radiologic and nuclear medicine procedures (Mettler et al 2020), so representing about 62% of the total individual dose burden. At the same time, it has been noted that the per capita dose decreased by nearly 20% over this period (Einstein 2020). In all cases, this outlines the importance of patient-based radiation dose monitoring for patient risk assessment, in particular for CT exams (Tsalafoutas et al 2020).
Patient dose estimates in CT via simulation software (Damilakis 2021) require both an appropriate in silico replication of the scanner and a refined computational model of the patient anatomy. Monte Carlo (MC) simulations of the scanner geometry and scan protocol include a description of the actual x-ray beam spectrum and conformation as shaped by the filter and bowtie compensator (Turner et al2009, 2010. Usually, digital anatomical models for patient dose estimates represent an 'average' (reference) patient (e.g. Papadakis and Damilakis 2022), but recently, large patient's model datasets have been produced for better representng the gender variability. Based on the scanner CT computational model and on a digital patient model, conversion coefficients for dose estimates from measurable values are computed via MC simulations (ICRP 2007.
The organ and effective dose to the patient undergoing CT examination depend both on the scanner type and protocol and on the physical conformation of the body (size and shape), e.g. as based on size-specific dose estimates and water-equivalent diameter (AAPM 2011, 2019a, 2019b, Abuhaimed and Martin 2020. In principle, dose estimates could be derived for all the scanner types and protocols, with in silico simulations (e.g. Li et al 2011, Fujii et al 2020. On the other hand, a digital patient cohort large enough to represent closely all possible patient anatomies, is unrealistic: hence, effective dose or organ dose estimates completely tailored to the actual patient under examination are not practical. To mitigate this limitation, a proposed approach is to tune and deform pre-calculated reference digital patient models (Zvereva et al 2017, Borbinha et al 2019, e.g. based on patient's weight and height (Sahbaee et al 2014, Lee et al 2015, 2019, ICRP 2020. In the approach of Lee et al (2019), conventional digital phantoms were deformed starting from secondary body parameters such as sitting height, head height, length, and breadth, sagittal abdominal diameter. Upper arm, waist, buttock, thigh, and calf circumferences were used for extending the patient cohort, also including 10th, 50th and 90th percentile standing heights and body weights of adult male and female Caucasian populations. In a different approach, the user interface application developed by NCI in USA (Lee et al 2015) permits the customization of the patient model starting from the weight and height of the patient, for the calculation of effective dose from a pre-calculated datasheet which includes the description of many different reference persons. In any case, the proposed customized digital phantoms are models of the exposed patient derived from epidemiological and statistical data, and they do not present the shape and silhouette of any actual patient under CT examination.
In this work, we propose a method for the customization of patient digital models based on the assessment of the external shape of the actual patient, for a customized phantom description of the individual patient, for input to a MC simulation of the CT scan for dose estimates. This might represent a more accurate approach to consider the effective anatomy of the patient, for minimization of patient dose assessment inaccuracy, with respect to present patient size metrics, in particular for pediatric patients (Sapignoli et al 2022). To this purpose, a new methodology and a new MC code was developed, for organ dose estimates using the patient-specific voxelized phantom. Such an approach relies on a pre-acquisition of the external shape of the patient via a 3D body scanner. Having determined in this way the patient's silhouette, the standard body model (digital phantom from a reference person dataset, selected appropriately for gender and class age) undergoes a deformation to fit the measured shape.
We point out that, while it is possible to derive a patient-specific voxelized phantom after the CT scan by segmentation of the corresponding CT slices, our goal is to provide a method for personalized CT organ dose estimates before the exam, by combining the 3D optical scan of the patient body, by creating a personalized (voxel) phantom and by executing a (fast) MC simulation for organ dose estimate, reproducing the characteristic of the CT scan protocol and the patient anatomy. This estimate could be used to tune the CT patient dose before the scan, to be compared with dose estimates using conventional CT dose estimates which adopt standard reference person phantoms (e.g. as provided by ICRP Publication 110, 2009, or by ICRP Publication 145, 2020. As a result, any discrepancy between the two organ dose estimates could be evidenced, to be possibly attributed to the specific description of the patient anatomy with respect to reference person-based dose estimates. Here we present the new method, validate the MC program, and evaluate, via MC dose estimates, the impact of the customization of the patient model (assembled from the measured body silhouette) on the estimated dose, in the case of head CT scanning. We also point out that our main goal in this work is to outline the entire method, to present and validate a MC code for organ dose estimate, and to investigate the feasibility of the proposed approach, rather than providing optimized solutions for each of its application steps. For example, the dosimetric method here investigated could be adopted with a different MC simulation code, or with a different optical scan technique, or with different and more efficient methods for phantom volume deformation than the rigid deformation approach here adopted. Moreover, the case of head CT scan was chosen, as a first approach for a proof-of-principle of the proposed method to pre-exam personalized CT dosimetry. The cross-sectional area of the head region showed variations of ±6% for male patients and ±7% for female patients (Huda et al 2004). CT dosimetry for more complex anatomies (e.g. chest) are under investigation (though preliminary addressed in the following) and will be reported in the next steps of this initial study.

Method
The flowchart in figure 1 describes the methodology steps proposed for the personalized MC estimate of organ doses; the processing steps are described in the following sections.

The MC simulation
The Monte Carlo code was derived from a previous version used for in silico x-ray breast imaging investigations (di Franco et al 2020, Sarno et al 2017aSarno et al , 2018Sarno et al , 2022. It is based on the Geant4 toolkit and physics list Option4. The new code version permits to perform spiral scanning geometry and to set the beam aperture at the detector, the length of the scan and the pitch. The MC software includes simulation of photoelectric, Compton and Rayleigh x-ray photon interactions. The electrons are not tracked but supposed to deposit energy locally. For the purpose of this work, the patient dose map was simulated: for each interacting event, the location and the released energy was computed and stored in a 3D matrix.

The beam model
The replicated scanner specifications and characteristics, adopted for this proof-of-principle study, were those of the Astelion CT scanner produced by Toshiba (model CGS-61A). The scanner permitted the use of tube voltages of 80 kV, 100 kV, 120 kV and 135 kV. To replicate the model of the x-ray beam in the MC simulations, we followed the approach described in Turner et al (2009). This permits to measure the relative beam fluence without precise knowledge of the shape of the bowtie embodied on the scanner. With this goal, a 3 cm 3 Radcal 20 × 6-3CT ionization chamber was placed at the center of the field of view (FOV) and the dependence of the beam intensity on the distance from the isocenter was investigated. The accuracy of the dosimeter was within ±4%, as reported in the calibration certificate. The axis of the ionization chamber was placed corresponding to the axis of the CT gantry and laterally centered. During the measurements, the x-ray source was kept stationary (scout image). Several measurements were performed with the ionization chamber moved far from the scanner isocenter along the radial direction of the gantry in the plane perpendicular to the beam propagation direction. The relative beam intensity was sampled with a sampling step of 1 cm up to 14 cm distance from the scanner isocenter. The evaluated beam profile was used in the MC simulations to describe the probability density function of the directions of the generated photons. On the other hand, in the axial direction, the beam was supposed to emit photons with a uniform probability. To model the x-ray spectra, the beam HVL was evaluated (Turner et al 2009) and the used spectra computed as suggested in Hernandez et al (2017); then, the measured HVL was matched by varying the added aluminum filtration. X-ray spectra were modelled for 80 kV, 100 kV, 120 kV and 135 kV tube voltages (i.e., all the spectra available from the used Astelion CT scanner).

The customized digital phantom
Customized phantoms were generated starting from the family of digital phantoms developed by the USA National Institute of Health (NCICT phantoms, Geyer et al 2014, Lee et al 2015). The dataset includes 370 phantoms of both genders and of various heights and weights and representative of a large range of patient ages, from newborns to adult individuals. Specifically, in this work we used one digital phantom from the dataset representing an adult male, and one representing an adult female individual. The first comprised 769 axial slices whose thickness was 2.207 mm; each slice comprised a matrix of 267 × 194 pixels, with pixel size of 1.579 mm × 1.579 mm. The female phantom comprised 610 axial slices with thickness of 2.700 mm and made by a matrix of 308 × 240 pixels with pitch of 1.260 mm. In this work, exclusively head scanning protocols were investigated and exclusively the heads of the two phantoms were employed.
To reproduce a digital phantom with the dimensions of the actual patient, we adapted the selected phantom to the silhouette derived from an optical 3D scan. Hence, an optical 3D scan of the actual patient physical phantom was acquired; then, the male or female digital phantom selected from the NCI database underwent a rigid adaptation, which consisted either of compressing or stretching the voxel sizes, to reduce the discrepancy between the dimension to the 3D patient rendering. In this proof-of-principle laboratory-based study, we used polystyrene manikins instead of actual patients, for the evaluation of the methods and of the impact of the model mismatch to the actual patient on the estimated dose. The silhouette of the manikin was acquired via the handheld Artec Eva 3D scanner (https://www.artec3d.com/it/portable-3d-scanners/artec-eva). This can be used to obtain an accurate and pre-textured 3D model of medium-sized objects, such as a human head or torso.
The Artec Eva 3D optical scanner works by structured light triangulation, that allows to receive 3D data from objects contactless, and by simply illuminating them with (unharmful) light flashes. It uses two cameras, mounted at different locations. For acquiring surface data via the active structured light technique, it is necessary to project pre-defined patterns on the surface of the object (usually in the form of parallel beams): the patterns become then distorted when projected onto the object's surface (the scanner captures these images using the cameras). Then, this information is given to the software that elaborates it via the method of triangulation to calculate the object's depth and surface information. The final output from the scanner is a digital 3D representation on the computer.
A structured-light 3D scanner can only take 3D images of what the cameras can see: therefore, to create a digital model of the entire object, optical scans must be taken at multiple angles (typically by rotating around the object). The scans are then cleaned up, merged, and stitched together (known as post-processing) to create a complete digital model (the user receives a color texturized 3D mesh made up of triangles). The Artec Eva 3D scanner was controlled by a notebook running the Artec Studio Professional ver. 16 software, used for reconstructions and elaborations of the models. Artec Studio performs the fusion of all the frames acquired during the rotation of the scanner and allows the user to 'correct' the 3D model, for example, by filling holes, erase undesired parts of the object etc. Figure 2 shows the 3D rendering of the 3D scans of body parts and manikins.
To evaluate alternative approaches for dose estimates for patients whose silhouette differs from the standard one, we have also employed the NCICT 3.0 software (NCI, NIH, USA) (Lee et al 2015). This software permits to customize the patient model on the basis of his/her height and weight; hence, both for female and male adult models and fixed heights, we tested the influence on the dose calculation of changing the phantom size by increasing at maximum its weight, this change being supposed to correspond to the maximum difference in the patient silhoutte and dimension.

Dose computations
The absorbed dose (D) to the kth organ (D k ) was computed as the average dose over the organ tissue volume as follows: where m k is the mass of the kth organ tissue, and the sum considers the energy deposited (E i ) for each of the ith interacting events occurring in the kth organ. Electrons are supposed to release energy locally and de-excitation processes are not simulated. With the intent of passing from the simulated dose to the patient model to dose released by the clinical scanner with a selected protocol, the CF factors were introduced ( where CTDI 100 is the CT dose index evaluated for a 100 mm long pencil ion-chamber. The superscript air indicates that it is evaluated at the scanner isocenter with the ion-chamber placed in air (with no phantoms). The subscripts 'm' and 's' indicate measured and simulated CTDI 100 , respectively. The simulated value is expressed in mGy per photon, while the measured one was evaluated for 1 mAs. The CF factor permits to convert dose to the digital model (in mSv per photon) to dose to the actual represented patient, in mSv per mAs. In order to calculate the CT dose index in the simulated scanner CTDI , 100,s air ( ) we modelled the pencil ion-chamber as a cylinder made of dry air whose diameter was 1.10 mm and the length was 100 mm: where m i is the mass of the air contained in the ion-chamber volume, N ev is the total number of generated primary photons (10 9 in our case), N the number of image sections irradiated in the single source rotation, and T is the width of the image sections (in cm). For this evaluation, the source was confined to an axial scan. CTDI 100,s air was measured on the clinical scanner with the same geometrical and protocol used for the simulated scan, and for 100 mAs.
To validate the MC code, we also computed and measured the CTDI w . For this reason, the pencil ionchamber was first placed along the central axis of a PMMA cylinder with a diameter of 16 cm and the axis of 10 cm as used in the evaluation, for the CTDI in the head examinations (AAPM 2019b). The CTDI 100,s at the center of the phantom (CTDI 100,s,c ) was computed as indicated in equation (3) for the case in air. Similar simulations were performed shifting the ion-chamber 15 cm laterally from the axis of the PMMA phantom. The simulations were repeated for the ion-chamber located at 3, 6, 9 and 12 o'clock positions over the circular perimeter of the PMMA cylinder (AAPM 2019b) and the simulated weighted CTDI (CTDI w,s ) was evaluated. Measurements for the same simulated geometry were repeated on the clinical scanner, and the CTDI w,m was evaluated.

Monte Carlo validation
To validate the new scanning geometry of the CT scanner used, and the specific beam conformation, we performed additional tests. A first simple test was performed to confirm that the simulated beam profile reflected that of the measured one. Secondly, the measured CTDI w to CTDI 100 ratio was compared to the simulated one at 80 kV, 100 kV, 120 kV and 135 kV tube voltage.
A physical male adult head phantom was 3D printed statring form the digital one from the NCI database (figure 3). The head digital phantom was digitally cut via a freeware CAD software (3D Builder, Microsoft Corporation Inc.) in 4 sections so to obtain 4 STL format files. These 4 sections were separately given as input to the 3D printing slicing software (Cura, Ultimaker) which returned a gcode format file used for the operations in the printing process. The physical phantom was manufactured via Ultimaker 2 fused deposition modelling (FDM) 3D printing technology. The phantom was made with white PLA material and printed through a 0.4 mm diameter nozzle, layer height of 0.2 mm, and infill factor of 40%. The printing time spent for each section of the phantom was about 18 h.
To measure the dose distribution, a large area of 20 cm × 25 cm GafChromic™ film piece of XR-QA2 type (Ashland Inc., USA) was then placed in a coronal plane between two halves of the phantom. The phantom was then scanned at 120 kV, 300 mA, pitch 1 and beam aperture 16 × 1. The GafChromic™ film lot was previously calibrated as described in previous works (Tomic et al 2010, Di Lillo et al 2016, Sarno et al 2017b, Valero et al 2022. For this reason, 3 cm × 5 cm film pieces were irradiated at known exposure levels. They were held by a PMMA frame and nylon wires and faced the x-ray source ( figure 4(a)).
They were irradiated via scout acquisitions whose length permitted to include the entire piece in the field. A 6 cm 3 RadCal ion chamber model 10 × 6-06 was then located at the same position of the GafChromic™ sheets and air kerma measured by repeating the previous scout images ( figure 4(b)). The accuracy of the dosimeter was within ±4%, as reported in the calibration certificate. To increase the measured air kerma and to cover a broad enough exposure range, both radiochromic film pieces and the ion chamber underwent from 1 to 14 consecutive irradiations. To reduce the acquisition time and scanner load, we selected three contiguous regions of interest of the exposed GafChromic™ piece, meant as independent acquisitions for the evaluation of the calibration function, as done in Tomic et al (2010).
Figure 5(a) shows the evaluated calibration curve; this was fitted with a rational curve (Di Lillo et al 2016). The uncertainty curve was evaluated as described in Di Lillo et al (2016) and Valero et al (2022) and reported in figure 5(b). The dose evaluations were performed for a dose level which guaranteed a total uncertainty lower than 5% ( figure 5(b)).  ion-chamber located to measure the air-kerma at which GafChromic weas exposed. Figure 6 shows the measured beam profiles in the radial direction in the axial plane. It was evaluated up to 14 cm from the scanner isocenter, intended to cover the entire FOV for head scanning and for tube voltages of 80 kV, 100 kV, 120 kV and 135 kV. Continuous lines represent sigmoidal fits used for simpler analytical model of the profile following the parametric formula:

Beam profile model
Here, x is the distance from the isocenter and A 1 , A 2 , x 0 and d are fitting coefficients. Table 1 reports sigmoidal curve fitting coefficients evaluated for the measured curves in figure 6. R 2 fitting parameters are also reported in the table, showing the high suitability of the proposed sigmoidal model. The beam profiles are supposed to be symmetric with respect to the central axis of the scanner, in the CT scanner simulations. A radiochromic film sheet was irradiated at the scanner isocenter for verifying the assumption of uniform probability distribution of the beam exposure in the axial direction ( figure 7).
The film was irradiated with the stationary x-ray source and attached at the gantry to be joined with the beam footprint ( figure 7(c)). The exposure profile was sampled at the scanner isocenter in the axial direction ( figure 7(a)) (in addition to the radial direction, figure 7(c)). A fit of the central part of the beam profile ( figure 7(b)) shows that the hypothesis of beam uniformity in the axial direction is reliable. Hence, a constant  function, used for fitting central profile points comprised in the 4 mm aperture of the beam, presented a R 2 fit parameter higher than 0.999. The beam profile also presents exposure tails that may be caused by the finite size of the focal spot or scatter from bowtie and collimators. These tails were not simulated. However, such an amount not comprised in the primary beam is considered in the CF factor thanks to the use of long ion-chamber for sensing the entire beam output in the dose computation. Figure 7(d) also shows a comparison between the simulated beam profile as a function of the distance from the isocenter, and that measured with the radiochromic film. Table 2 reports the measured beam HVL for the available tube voltages. The evaluation was performed at the scanner isocenter and at 50 mm from the isocenter, where the bowtie could also influence the beam quality. However, differences between HVL in the two positions are within the experimental error. In this work, the beam was modelled following the HVL evaluated at the scanner isocenter.

.1. Validation of the simulated beam profile
In the first validation test, summarized in figure 8, we compared simulated beam profiles (continuous curves) to the measured ones (data points). These beam profiles were evaluated as air kerma (normalized to the value at the isocenter) as a function of the radial distance, for tube voltages between 80 kV and 135 kV. As in the cases of evaluations in previous paragraphs, these profiles were evaluated in the radial direction up to 140 mm from the isocenter. Simulated beam profiles, evaluated with a spatial resolution of 1 mm, lie on the measured curves (differences much less than 1%). A second test compared simulated and measured CTDI w , normalized to the CTDI measured in air (CTDI air ), without using the head PMMA phantom. Results are reported in figure 9. Differences were contained within 4%, largely within the measurement's errors of the CTDI w /CTDI air ratio of about 8%. Figure 10 shows the simulated 3D dose distribution obtained for the male (figures 10(a), (b)) and female (figures 10(c), (d)) adult NCICT head phantoms (100 kV, 10 9 histories, pitch 1). These 3D dose distributions, replicated at 120 kV, were used for MC validation tests. Hence, the same evaluation was performed via a GafChromic™ film sandwiched in between the two halves of the 3D printed head phantom. To compare dose distribution profiles, the head was simulated with a homogenous phantom made of PLA material with a density of 0.408 g cm −3 . This value was obtained by calculating the mean of the measured densities of the four head sections. Figures 11(a) and (b) report the simulated and measured dose distributions in the middle plane of the head phantom, respectively. Simulated and measured axial profiles of dose distribution evaluated at the scanner isocenter are reported in figure 11(c). Simulated dose profiles were rescaled by means of the CF factor. Figure 11(c) shows an acceptable agreement between the measured and simulated dose levels. The spatial frequency of the quasi-sinusoidal curve (peak-to-peak frequency) was 0.068 mm −1 and 0.061 mm −1 for the simulated and measured profiles, respectively. The little difference may be ascribed to misalignments of the central plane of the phantom in the measurements. The discrepancy between consecutive peak-to-valley differences was 6.5%, in the order of the uncertainty of measured differences, which present the higher value.  3.6 ± 0.6 4.2  0.5 100 4.7 ± 0.7 5.4  0.6 120 6.8 ± 0.7 6.6  0.6 135 7.9 ± 0.3 7.7  0.9

Validation of the simulated dose
The dose evaluated via the MC software adopted in this study and that computed via NCICT (Lee et al 2015) were compared, and the percentage discrepancies were reported in tables 3 and 4, for male and female phantoms, respectively. The percentage discrepancies D f were calculated as: For validation purposes, we compared the dose estimated for four tissues contained in the NCI head phantoms: brain, eyeballs, eye lens and pituitary gland. Comparisons were done for all tube voltages allowed by  the considered Astelion CT apparatus (i.e., 80 kV, 100 kV, 120 kV and 135 kV) and for male (table 3) and female (table 4) adult NCI phantoms, respectively.
Maximum differences were found for the estimated dose to the pituitary gland at 80 kV in the case of the male phantom (23.9%), and for lens dose at 135 kV in the case of the female phantom (24.8%). These differences  were mainly due to the two different MC codes employed in the simulations. Main differences may reside in the beam shape model-based on profile measurements in this work-and in the calculated conversion factors (equation (2)); a little discrepancy may also derive from subtle differences in the simulated physics. However, in most of the cases, differences were lower than 15%, in absolute value. In analysing literature data comparing different software codes used for organ dose estimates, we note that differences as large as 160% were found (Ding et al 2015).

Personalized patient dose
We reported the evaluated CF coefficients, for the selected scan and tube voltages, in table 5. These values were used for converting simulated dose, evaluated in mSv per photon, to dose to the patient model for the specific exam, in mSv per mAs. For the female and male adult head phantoms provided by NCI, we calculated the dose for specific examinations whose characteristics were those evaluated for the Astelion CT scanner and previously reported. The scanning was performed with 4 × 1 beam aperture and for all the available tube voltages. This same scanner protocol was adopted for the computation of CF factors. Subsequently, we modified the standard female and male head phantoms in order to match the evaluated silhouette for two head manikins, representing actual patients head phantoms, as described in section 2.3. Differences in dose, evaluated for standard male adult head phantom and that evaluated for corresponding customized phantom,are reported in table 6.
Dose differences were evaluated for four selected tissues comprised in the head phantom: brain, eyeballs, eye lens and pituitary gland. The maximum difference was evaluated for lens dose at 135 kV (−13.7%). The customization of the male digital phantom on the basis of the measured silhouette, produced dimensions reductions of standard phantom of 13% in antero-posterior direction, and of 12% in lateral direction; in axial direction the standard phantom was extended by 8% to have the same length of the measured silhouette. The volume of the resulting head phantom was 18% smaller than the starting one. Similarly, dose differences evaluated for the female phantom are reported in table 7. In this case, the maximum discrepancy reached 38.0% for pituitary gland dose at 100 kV. These larger discrepancies may be ascribed to the stronger deformation to which the female standard digital phantom underwent. Indeed, to present the same dimensions of the measured  silhouette, the standard phantom was reduced by 26% in lateral direction, by 11% in antero-posterior direction and by 16% in axial direction, with a resulting volume reduction of 45%. Tables 8 and 9 report dose discrepancies evaluated with NCICT 3.0 software in the case of customization of the patient model based on its weight. The adopted scanning protocol was the same used in previous tests. For a male phantom of 175 cm height, increasing the weight from 75 kg to 135 kg determined a maximum dose difference of 8.9%, evaluated for the eye lens. In the case of the brain dose, the discrepancies were less than 1.0% (table 8). The same test was performed for female phantoms of 160 cm height, and discrepancies were reported in table 9. In this case, the model weight was increased from 70 kg to as much as 125 kg, and the related discrepancies ranged between 2.6% and −7.5%, evaluated for the brain and the eye lens, respectively, at 80 kV.

Discussion
Evaluation of radiation dose in CT is a topic of high interest due to the increasing use of CT examinations worldwide and the corresponding dose burden. In a CT scan the organ dose mainly depends on patient's anatomy, scan region and scanner output; organ dose estimates permit to compute the effective dose for the given CT scan. However, the dose to the organs is not an immediate information, and the general approach is to use a Monte Carlo algorithm associated with an anthropomorphic phantom to calculate it. To date, several software programs have been introduced to calculate the organ dose in CT (e.g., CTExpo, National Cancer Institute CT, NCICTX, NEXO[DOSE] ® , Virtual Dose). All these softwares allow to calculate a reference dose for standard phantoms or to evaluate the dose after the CT scan. We have proposed a personalized dose evaluation through a MC code and a customized phantom to evaluate a personalized dose estimate before the CT exam. This code is based on the phantom customization, realized through the adaptation of internal part of NCICT standard phantoms, to the external body silhouette derived from an optical 3D scan. The MC code was validated on a specific multi-detector spiral CT scanner (Toshiba Astelion), but it can be promptly adapted to replicate the characteristics of any clinical scanner. Our Geant4 based MC code for head CT simulation featured a processing time in the order of 4 × 10 5 primary photon histories s −1 , when running on a cluster of 128 processors on a multi-CPU server platform.
We point out that, in a recent work, we have demonstrated Graphics Processing Unit (GPU) accelerated fast MC simulations for processing times in the order of 10 9 photon histories s −1 , with a code running on a single NVIDIA GeForce RTX 3090 GPU card (10496 NVIDIA CUDA cores, 24 GB GDDR6X RAM). This GPU code (initially developed by, and in collaboration with, the group of Prof X Jia at University Texas Southwestern) was intended for dedicated cone-beam breast CT, digital mammography, and digital breast tomosynthesis imaging and dosimetry. It includes a 3D dose map output (voxel resolution about 0.03 mm 3 ), as well as projection and tomographic imaging output, with a resolution in the order of (0.3 mm) 3 for a phantom volume in the order of 10 6 mm 3 . When translated to the head geometry and the helical simulation scheme used in this work, the adoption of this GPU-based MC code is expected to feature a total processing time in the order of 10 s for dosimetry with statistical uncertainty less than a few percent. This predicted performance, still to be demonstrated for whole-body CT scans, would show the potential for the clinical implementation of the present approach for real-time pre-exam personalized organ dose estimates in multi detector CT. Tube current modulation (TCM) during the patient scan can be included in the MC code once the tube modulation function of the specific scanner type has been determined (e.g. retrospectively from archive scans by analyzing DICOM header's TCM info).

Code validation
The validation of the proposed code was realized by a comparison of simulated and measured beam profiles and CTDI. The Astelion CT scanner produced by Toshiba (model CGS-61A) was replicated in the code, and the beam was modelled using the HVL evaluate at the scanner isocenter. The hypothesis of beam uniformity in the axial direction was verified by radiochromic film measurements. We compared the simulated and measured beam profile evaluated as air-kerma normalized to the value at the isocenter, obtaining a discrepancy well below 1%. On the other hand, the comparison of simulated and measured CTDI w , shows an absolute difference of 4% (figure 9), which is in line with, if not better than, values reported in the literature (Kostou et al 2019).

NCICT Validation
The dose values for brain, eyeballs, lens and pituitary gland, provided by our MC code using an NCI head phantom, showed a maximum discrepancy for the pituitary gland at 80 kV in the case of the male phantom (23.9%), and for lens dose at 135 kV in the case of the female phantom (24.8%), when compared to the calculation perfomed via the NCICT software (Lee et al 2015). In the other cases the differences were lower than 15%, in absolute value (tables 4 and 5  (2020), where the evaluation of the different sources of uncertainty were analyzed.

Personalized patient dose
The developed MC code provided organ dose estimates using a customized phantom, realized through the adaptation of internal part of NCICT standard phantoms to the external body silhouette derived from an optical 3D scan. In the case of the male head phantom (table 6), where the phantom deformation led to a phantom's volume increase by 14%, the dose to the standard phantom was up to 14% higher than that evaluated for the customized phantom. The maximum differences were observed for the lens dose, with overestimations of customized phantoms ranging between 10.2% at 80 kV, and 13.7% at 135 kV. The lower differences were observed for the brain, not exceeding 5.1% in absolute value. The head phantom modification in the case of the female individual (table 7) produced an overall increase of the volume by 31%. The volume modification observed, larger than that for the male phantom, may be the main cause of the larger observed discrepancies between dose evaluated for the standard and customized phantoms. Hence, the former led to dose estimates down to 38.0% lower than the second (evaluated for the pituitary gland at 100 kV). For the brain, the standard phantom underestimation ranged between 17.9% and 36.3%; these values ranged between 8.9% and 30.8% in the case of dose to the eyeballs, and 3.0% and 28.2% in the case of lens dose. Moore et al (2014) proposed to customize the pediatric organ dose on the basis of the size specific dose estimates (AAPM 2011). Estimated values were within ±10% compared to measurements with pediatric phantoms. More sophisticated models for body deformations have been adopted for dose re-evaluation in image guided radiotherapy (Brock et al 2017, Paganelli et al 2018 with the maximum dimensionality of the transformations reaching up to 3 times the number of the voxels contained in the image. Such deformations permitted to recalculate delivered doses to the organs with the purpose of keeping target doses unchanged after patient modification and repositioning. The NCICT 3.0 softtware (Lee et al 2015), here adopted for our code validation,permits to customize the model of the patient in CT. In particular, it permits to manually input the height and the weight of the patient in order to select the most appropriate model over 370 digital phantoms representing both genders, children and adults. In a test performed in this paper, we considered two adult models, one for each gender, and evaluted the dose discrepancy obtained for the extreme of the allowed weight ranges, for a fixed model height. It is expected that such a weight range covers the maximum modification in silhouette for the selected phantoms, as well as the maximum dose discrepancies. In the case of the selected 175cm height male phantoms, going from 75 to 135 kg in weight does not produce substantial changes in absorbed dose, with maximum observed differences lower than 8.9% for eye lens at 80 kV. The customization proposed in this work, based on the measurement of the patient silhoutte, produced a difference of 13.7% for the same tissue, at 135 kV. In the case of 160cm female model-where the customization based on the measured silhouette led to dose differences up to 38.0% evaluated for pituitary gland-going from 70 kg to 120 kg model weight produced a dose discrepancy less than 7.5%, for the considered head tissues.
The results here presented for personalized CT head scans will be extended to (more complex) anatomical districts: preliminary data (shown in figure S1 in Supplementary material), implementing the MC code for CT chest scans, were considered promising in view of the application of the proposed methodology to pre-scan organ dose estimates in routine clinical CT scans of all anatomical districts. We point out that in the application of the proposed methodology, once a whole-body 3D optical scan is derived for the fist time for a given patient, the personalized phantom could be used, also with the same or different MC codes, for organ dose estimates, in successive exams.
We also envisage the possibility of adopting the proposed 3D optical scan of the patient body for nuclear medicine personalized dosimetry with MC codes (e.g. We are aware that presently, a limitation of this work-though not affecting the rationale of the new dosimetric method proposed-is related to the customization of the standard patient model to the silhouette of the actual one. Indeed, the adopted rigid adaptation scheme, based on the compression/stretching of the voxel sizes in the three spatial directions, may cause a deformation of those internal structures of the body which does not follow this proportional law (e.g. the eyballs, for the head phantom). In addition, deformation may also be caused by the presence of fat layers manly located in the outer borders of the body, causing a radiation shield effect and a dose reduction to the inner radiosensitive tissues. Application or development of more efficient deformable image registration methods should be investigated (e.g. Zvereva et al 2017, Borbinha et al 2019); moreover, the proposed methodology for personalized CT dosimetry should be compared to other approaches, as here done versus NCI tools for both the reference phantoms dataset and the organ dose estimate algorithm.

Conclusions
We carried out a validation of a new method for obtaining Monte Carlo-based personalized dose estimates in CT before performing the scan, which involves (i) to input a realistic patient body shape and size by 3D optical scanning, (ii) to compute a personalized voxelized body phantom by adaptation of reference phantoms, and (iii) to derive organ doses using a validated MC simulation. The proof-of-principle test shown here was conducted on adult head phantoms, with head organ dose estimates comparable with corresponding values obtained using reference body and organ phantoms provided by NIH/NCI dosimetry tool. Possible extension of the MC code to chest spiral CT scan was indicated.