Introduction

Pregnant females represent a critical subpopulation for which absorbed doses from radiological imaging procedures must be evaluated to make critical decisions regarding the outcome of the developing foetus. Over the period 1997–2006, the number of CT examinations performed on pregnant patients in the USA has increased by over 400% [1]. A similar trend is also observed in molecular and hybrid imaging technologies [2, 3]. This increase has led to increasing health concerns regarding the radiation risks to the foetus. Obviously, the best way to care about radiation protection of patients is to avoid radiation as much as possible. Alternative non-ionising imaging modalities, such as ultrasound (US) and magnetic resonance imaging (MRI), have to be considered first to diagnose the cause of symptoms. At foetal doses greater than 100 mGy, the potential hazard effects of radiation on the foetus include possible spontaneous abortion, intrauterine growth limitation, average intelligence quotient (IQ) loss, mental retardation, possible organ malformation and small head size [4]. Stochastic effects, such as cancer and organ-specific disease, might also occur at foetal doses below 50 mGy [5,6,7]. The foetal nervous system exhibits a longer period of sensitivity to ionising radiation than other foetal tissues and is also known to be affected by radiation exposure above 5 mGy [8]. The increased absolute incidence for childhood cancer is 0.017% per mGy [9], while the excess relative risk for leukaemia and brain tumours are 0.036 (95% confidence interval (CI) 0.005–0.120) and 0.023 per mGy (95% CI 0.0098–0.0494), respectively [10].

Accurate estimation of foetal absorbed dose and associated risk factors from CT examinations is highly desired owing to the high radiosensitivity of developing foetal organs. However, this constitutes an enormous challenge since direct measurements of energy deposition in the foetal body cannot be performed in human beings because of ethical and radiation protection issues, and alternative approaches using anthropomorphic phantoms can be used as a substitute [1]. Different approaches have been followed to estimate the foetal dose, including Monte Carlo simulations using computational anthropomorphic models [11,12,13,14,15,16,17,18,19,20,21] and measurements using physical phantoms with embedded dosimeters [22,23,24,25,26]. However, these approaches inherently bear a number of limitations, such as the difficulty of constructing patient-specific computational models and the difficulty of matching anthropomorphic physical phantoms to the size and location of the foetus within the maternal body [27]. Therefore, the assumptions inherent to the measurement and simulation setup might cause significant overestimation or underestimation of the foetal absorbed dose. While Monte Carlo calculations are considered as reference (gold standard) for dose estimation in diagnostic imaging [28], implementation of a framework for patient-specific estimation of foetal dose from CT examinations of pregnant patients in clinical setting is highly desired.

In this work, we propose a methodology for the construction of patient-specific computational models based on image segmentation and registration methods and previously developed standardised hybrid computational phantoms of pregnant female series [29]. The geometry of the Discovery CT 750 HD CT scanner (GE Healthcare, Waukesha, WI) was designed and integrated within the N-Particle eXtended (MCNPX) Monte Carlo code. The patient-specific maternal and foetal absorbed radiation doses were calculated and compared with the results provided by the Radimetrics™ (Bayer Healthcare) commercial dose-tracking software [30]. The correlation between patient anatomy, gestational age conceptus position and foetal organ-level dose from CT scans was investigated.

Materials and methods

Pregnant patient CT examinations

The institutional ethics committee approved this retrospective study. Written informed consent was waived. CT images of 30 pregnant patients referred to the emergency unit of Geneva University Hospital (HUG) for abdominal and pelvic CT imaging were included in the study protocol. US was systematically performed upon admission of every pregnant patient with acute abdominal pain. Further examination (MRI or CT) was obtained when US was either inconclusive or negative with a high clinical suspicion. Low-dose CT was performed when MRI was not immediately available. The studies were acquired on the GE 750HD CT scanner (120 kVp, slice thickness from 0.625 to 3 mm, pitch of 1.375, table speed of 55 mm/rot and 31.5 mAs tube current without current modulation corresponding to 22.9 effective mAs, i.e. mAs/pitch). Tube current modulation was not applied in this retrospective study approved by the ethics committee until the end of the initial study. All pregnant patients receiving low-dose CT scans (same tube current and CTDIvol of 1.8 mGy) between March 2011 and September 2015 were included. The 30 pregnant patients (age range: 19–45 years) had gestational ages ranging between 8 weeks and 35 weeks, with an average gestational age of 19 weeks. A representative CT slice of a pregnant patient is shown in Fig. 1a. The body contour, skeleton and lungs were automatically segmented using thresholding, whereas the uterus was manually identified (Fig. 1b). Thirty voxelised regional models corresponding to the axial scanned length of pregnant patients were created from the segmented CT image data of the maternal body, including skeleton, lung and uterus. The craniocaudal length of voxelised regional models of patients depends on the scanned length of CT image data and ranges at least from the lower thorax to the pubic symphysis of the patient’s anatomy. The maternal perimeter and conceptus depth were measured automatically from the constructed regional voxel model for each patient using an in-house C++ code as shown in Fig. 1b. The maternal perimeter was determined by the maximum outer perimeter of the patient on the image containing the uterus while the conceptus depth was measured as the average distance from the skin to the closest surface of the uterus because it is difficult to segment the foetal body on low-dose CT images. These two measurements were used to estimate the size of the maternal body and the location of the foetus within the mother’s abdomen, respectively.

Fig. 1
figure 1

(a) CT image with external body contour (red line) around the perimeter of a pregnant patient, the uterus (blue line) with a 29-week-old foetus. (b) Segmented CT image with external body contour, skeleton and uterus. The assessment of conceptus depth (averaged distance from skin to uterus [double arrow]) is also shown

Full size image

Patient-specific computational models

A series of ICRP-based standardised computational phantoms of pregnant females at the 8th, 10th, 15th, 20th, 25th, 30th, 35th and 38th weeks of gestations developed previously [29] were used as anchor models for the construction of patient-specific models. The corresponding anchor model of the closest gestational age was registered to the constructed voxelised regional patient-specific model using automatic affine registration to produce new personalised pregnant phantoms with well-defined anatomical structures, matching patient images obtained from CT examinations. The resulting models after registration are referred to as ‘patient-specific computational models’ below. Image registration was performed using the Insight Toolkit (ITK) [31]. Figure 2 shows two representative examples of the patient-specific computational phantoms together with clinical CT images of corresponding patients. Each patient-specific computational model includes 35 maternal organs and 25 foetal organs (Supplemental Table S1). The uterus of the patient-specific computational phantoms including the embedded foetus and other tissues were adjusted using rigid registration to the uterus region from patients’ CT images. The uterus of the standardised phantom, including the foetus, was moved and scaled to fit the centroid and size of patient’s uterus while the foetus was set in a head-down position.

Fig. 2
figure 2

Representative slices showing transverse (top), sagittal (bottom and left) and coronal (bottom and right) views of two examples of the developed computational phantoms at (A) 8 and (B) 35 weeks gestational age illustrating image registration between abdominal CT images of the patient and the computational phantom for the development of patient-specific pregnant computational models. A zoomed view of the 3D patient-specific pregnant computational models with the embedded foetus is also shown

Full size image

CT source model

Monte Carlo-based techniques were used to create a previously validated GE 750HD CT source and CT gantry geometry models [32]. This CT scanner was equipped with Performix Pro VCT 100 x-ray tube with a 56° fan-beam angle, 7° target angle and allows for a beam collimation of 40 mm. The quality equivalent filtration of the x-ray tube is calculated at a nominal thickness value of 4.3 mm of aluminium at 120 kVp. The measured half-value layer was 7.8 mm Al at 120 kVp. The source to isocentre distance on this CT scanner is 54 cm, while the source to detector distance is 95 cm. The scanner’s x-ray energy spectrum was generated using SpekCalc [33] and described as a function of the number of photons per kiloelectronvolt (keV) energy intervals in the Monte Carlo code. The Teflon bowtie filter and beam collimation were also included in the CT gantry model [32].

Monte Carlo simulations

The CT source model and computational models were integrated in the MCNPX code [34]. The simulated low-dose CT examinations were performed using the same parameters used for patients’ scanning with a helical source path and total collimation width of 64 × 0.625 mm. Simulations of x-ray photon transport were performed in the CT model using all 30 patient-specific computational phantoms as input. The energies deposited in the 35 maternal organs and 25 foetal organs were recorded. The simulations for each helical CT scan were repeated six times with the tube starting angle differing by 60° because of the unknown tube- starting angles in the actual examinations. The foetal tissue compositions were obtained from ICRP 89 [35] with the bone density set to 1.3g/cm3, whereas is was 0.99g/cm3 for soft tissue.

Experimental measurements and simulations of absorbed dose

To calibrate the developed CT model and Monte Carlo simulation code, a series of dose measurements and simulations involving spiral CT scanning were performed in air using a 10-cm RaySafe™ Solo pencil chamber (Unfors RaySafe GmbH, Germany). The measurements were performed by positioning the ion chamber at the isocentre of the CT scanner with the long axis aligned with the axis of the CT gantry rotation. CT examinations were performed on the active volume of the detector, and the energy deposited in its air cavity was obtained and recorded as \( {D}_{measured, in\; air} \). Monte Carlo simulations were then performed for the detector in air under the same geometrical conditions. The simulated free-in-air exposure value of the ion chamber was recorded as \( {D}_{simulated, in\; air} \). The ratio of the measured to the simulated in-air dose refers to the conversion coefficient from simulated results to physical absolute dose and is defined as the output correction factor (OCF) [36]:

$$ \mathrm{OCF}=\frac{D_{measured, in\kern0.5em air}}{D_{simulated, in\kern0.5em air}} $$
(1)

In this study, the measured and simulated free-in-air doses at 120 kVp tube voltage were 33.8 mGy/100 mAs and 4.8 mGy/100 mAs, respectively, with a calculated OCF value of 3.5.

To validate Monte Carlo-based dosimetric calculations, additional radiation dose measurements and simulations in similar conditions were performed using the head (16 cm diameter) and body (32 cm diameter) CT dosimetry phantoms made of polymethylmethacrylate. The measurements were performed by positioning the ion chamber at the centre of the cylindrical phantoms to obtain measured absorbed doses of \( {D}_{measured, head} \) and \( {D}_{measured, body,} \), respectively, after performing CT examinations along the length of the cylindrical phantoms. To include the scatter tails at the beginning and the end of CT scans of the CTDI phantom, the 5-cm axial field of view (FOV) was set along the centre of the 10-cm pencil chamber. The absorbed doses in the computerised CT dosimetry phantoms were calculated as:

$$ {D}_{estimated}={d}_{simulated}\times n\times \varOmega \times mAs\times OCF, $$
(2)

where \( {d}_{simulated} \) is the absorbed dose per photon emitted from the source obtained in simulations, n is the number of photons emitted from the source per solid beam angle per mAs, calculated from SpekCalc [33] simulated x-ray energy spectra, \( \varOmega \) is the solid angle of the fan beam, mAs is the effective tube current-time product value and OCF is the estimated output correction factor. The dose was reported both in mGy and normalised to CTDIvol.

Absorbed dose calculations for pregnant patients and foetus

The energy deposited in organs and tissues of the mother and foetus were recorded in the Monte Carlo code and used to calculate the absorbed dose of each organ and tissue from clinical CT scans according to Eq. (2). The effective dose was calculated according to the ICRP recommendations [37]:

$$ E=\sum_T{\omega}_T\sum_R{\omega}_R{D}_{R,T}, $$
(3)

where E is the effective dose, \( {\omega}_R \) is the radiation weighting factor for radiation type R, \( {D}_{R,T} \) is the contribution of radiation type R to the absorbed dose, and \( {\omega}_T \) is the tissue weighting factor for organ or tissue T reflecting its relative radiation sensitivity [37]. Since the effective dose recommended by the ICRP provides protection quantity for the whole population, the effective dose for the pregnant patient was calculated based on estimated maternal organ absorbed doses.

Subsequently, clinical CT images for each patient were imported into Radimetrics™ dose-tracking software [30] for calculation of maternal effective dose. Radimetrics™ collects CT scans directly from the hospital’s picture archiving and communication system and matches patient images with Cristy & Eckerman stylised computational phantoms [38] according to age, gender, weight and body size (as shown in Supplemental Fig. S1). The software extracts scanning parameters (tube voltage, mAs, scan range, etc.) from the DICOM files’ header information and calculates patient-specific absorbed dose at the organ level using Monte Carlo simulations.

The results obtained by Radimetrics™ and the effective dose calculated by multiplying the dose-length product (DLP) of CT examinations by appropriate conversion factors [39] were considered as reference values in this work. The results obtained using the patient-specific computational model and Monte Carlo calculations were further compared with these reference values [28]. This assumption is justified by the large number of users reporting on the validation of the MCNPX code and the more realistic computational models used in this work compared to stylised models adopted by Radimetrics™.

The absolute foetal dose is given in mGy for comparison between the results of this work and those reported in the literature. However, since the absolute foetal dose is scanner dependent, the CTDIvol-normalised-organ dose coefficient [40] was obtained by multiplying the foetal dose by the effective mAs and dividing the result by the CTDIvol (1.8 mGy) to normalise based on CTDIvol, which has been demonstrated to be effectively independent of the CT scanner [41].

Statistical analysis

The correlations between foetal organ doses for the cohort included in the study protocol and the three variables, namely gestational age, conceptus depth and maternal perimeter, were tested using SPSS [42]. Pearson correlation coefficients were calculated to determine statistical significance. Nonlinear curve-fitting was performed to assess the absorbed dose of foetal organs as a function of the three parameters (gestational age, conceptus depth and maternal perimeter) presenting with high and significant correlation (the Pearson correlation is significant at the 0.05 level).

Results

Validation of the Monte Carlo code

Table 1 compares the measured and Monte Carlo-simulated radiation doses in head and body CT dosimetry phantoms scanned on the Discovery CT 750HD CT scanner at 120 kVp with a collimation of 40 mm. The differences between measured and simulated absorbed doses for the head and body CT dosimetry phantoms are 0.03% and −3.5%, respectively.

Table 1 Comparison between Monte Carlo simulated and measured radiation doses in head and body CT dosimetry phantoms
Full size table

Patient-specific computational phantoms

The results of automated thresholding-based segmentation were compared with manual segmentation of all CT images with respect to the outline, skeleton and lungs. The Dice similarity coefficient [43] between the two methods varied between 0.92 to 0.97 with an average of 0.95 ± 0.01 (Table 2). The accuracy of the registration between the regional models of pregnant patients and the patient-specific computational models was evaluated by calculating Jaccard’s coefficient of similarity [43]. The latter varies between 0.45 and 0.65 with an average of 0.58 ± 0.05 among patients.

Table 2 Patient anatomical measurements together with the CTDIvol-normalised foetal total-body dose (mGy/CTDIvol) according to gestational age for the 30 pregnant patients involved in this study
Full size table

Absorbed and effective doses to pregnant patients

The absorbed doses of maternal organs were calculated for all patients according to Eqs. (2) and (3) using Monte Carlo simulations. Figure 3 compares the average absorbed dose for different maternal organs with estimates provided by the Radimetrics™ commercial dose-tracking system. The results provided by the commercial software are a bit different from our simulations but still in the same order of magnitude. For organs located in the region covered by abdominal and pelvic CT scans, the differences between simulated results and absorbed doses reported by Radimetrics™ dose-tracking software range between −47% and −16%. The organs located outside the acquisition FOV for abdominal CT scans, such as the brain, eye lens and thyroid etc., receive a significantly lower dose than other organs.

Fig. 3
figure 3

Comparison between absorbed doses to maternal organs estimated using Monte Carlo calculations and Radimetrics™ dose-tracking system

Full size image

The effective doses for the 30 pregnant patients varied between 1.1 mSv and 2.0 mSv, with an average of 1.6 ± 0.22 mSv, while Radimetrics™ reported an effective dose ranging between 1.7 mSv and 2.3 mSv with an average of 1.9 ± 0.14 mSv. The effective doses calculated using the DLPs and corresponding conversion factors of Huda et al.[39] range between 1.5 mSv and 1.9 mSv, with an average of 1.6 ± 0.11 mSv.

CT foetal dose

Table 2 summarises the calculated foetal doses normalised to CTDIvol according to gestational age of pregnant patients. The water-equivalent diameter [41] of the pregnant patients varies between 24.1 cm and 37.5 cm. The measurements of conceptus depth and maternal perimeter, which vary from 12.9 cm to 20.8 cm and 89.3 cm to 127.9 cm, respectively, are also shown. Table 3 compares the normalised foetal body doses with results reported in the literature. The average absolute dose to different foetal organs from CT examinations ranges between 1.12 mGy and 5.98 mGy (Table 4). The scanner independent CTDIvol-normalised foetal organ dose coefficients are shown in Table 4. The foetus brain and bone marrow receive an average dose of 3.11 ± 1.09 mGy and 1.44 ± 0.45 mGy, respectively.

Table 3 Comparison of foetal total-body absorbed doses from CT scans of pregnant patients between this study and results reported in the literature
Full size table
Table 4 Average absolute and CTDIvol-normalised foetal organ doses (organ dose/CTDIvol) for different foetal organs for the 30 patients
Full size table

The foetal total body dose is about 2.11 ± 0.37 mGy while Radimetrics™ reported 2.47 ± 0.42 mGy. The CTDIvol-normalised organ dose coefficients [40] of foetal organs vary between 0.62 and 3.32. Figure 4 shows the ratios of absorbed doses to the foetal body between the values estimated using Monte Carlo simulations and Radimetrics™. The average absolute differences of foetal doses between Monte Carlo simulations and Radimetrics™ are about 21%.

Fig. 4
figure 4

Ratios of absorbed doses to the foetal body between Monte Carlo calculations and Radimetrics™ dose-tracking system

Full size image

This study presents, for the first time, detailed organ-level radiation doses to the foetus from CT examinations. Figure 5 shows the average normalised absorbed dose of different foetal organs from CT examinations. It can be observed that the absorbed dose is non-uniformly distributed between foetal organs where the foetal skeleton and bone marrow receive about 3.8 and 2.0 times higher absorbed doses, respectively, than other foetal tissues. Pearson correlation coefficients between foetal organ doses and the three independent parameters (conceptus depth, maternal perimeter and gestational age) are summarised in Table 5. As shown in Fig. 6, linear regression analysis indicated no significant correlation between foetal total body dose and gestational age, while significant correlations were observed between foetal total body dose and the conceptus depth or maternal size. For the foetal brain, the correlation coefficients between absorbed dose and the gestational age, conceptus depth and maternal size are −0.83, −0.71 and −0.58, respectively, indicating a significant correlation between foetal brain dose and the above parameters. The absorbed doses to some organs, such as the kidney, gall bladder, large intestine (LI), small intestine (SI), urinary bladder (UB), etc., present insignificant correlations with both gestational age and conceptus depth. The foetal organ absorbed dose as a function of the anatomical and physiological parameters presenting with high correlations is parameterised using nonlinear curve-fittings (Table 5).

Fig. 5
figure 5

Normalised absorbed doses of foetal organs from CT scans performed on pregnant patients with the average values for each organ

Full size image
Table 5 Pearson correlation coefficients between foetal organ doses of the cohort and the three parameters (gestational age, conceptus depth and maternal perimeter). Nonlinear curve-fitting was used to estimate the foetal organ dose as a function of the three parameters presenting with significant correlation at the 0.01 level using a two-tailed statistical test. The R2 values for the non-linear curve-fitting are also shown
Full size table
Fig. 6
figure 6

Plots of radiation dose to the foetal body and foetal brain versus conceptus depth, gestational age and maternal perimeter. No significant correlation was observed between foetal total-body dose and gestational age

Full size image

Discussion

We implemented a methodological framework for assessment of patient-specific foetal dose suitable for application in clinical setting. Pregnant patients with a range of anatomical variability undergoing CT examinations can be adapted by segmenting CT images to produce a voxelised regional model and registering it to a standardised computational model to construct a patient-specific CT scan model suitable for incorporation in a dedicated Monte Carlo code. The radiation dose to the foetal body from CT examinations of pregnant patients varies within the range 6.7–13 mGy/100 mAs with a mean value of 9.2 mGy, while Felmlee et al. [44], Angel et al. [11], Damilakis et al. [12] and Gu et al. [21] reported foetal doses of 11.3 mGy/100 mAs (Picker 1200, Siemens DRH, GE 9800 and GE 8800), 10.8 mGy/100 mAs (GE HighSpeed CT/I, GE LightSpeed QX/i, GE LightSpeed Ultra, GE LightSpeed PRO, GE LightSpeed 16 and GE LightSpeed VCT), 22.6 mGy/100 mAs (Siemens Sensation 16 CT scanner) and 6.6 mGy/100 mAs (GE LightSpeed Pro16 and GE LightSpeed™ 16), respectively. The foetal doses normalised to CTDIvol were 1.17, 0.94, 3.32 and 0.80 for the body, soft tissue, skeleton and brain, respectively. The corresponding body and organ foetal doses reported by Gu et al. [21] are 1.4, 1.29, 4.47 and 1.00, respectively. The radiosensitivity and radiation risks of the developing foetal organs vary across the different trimesters or gestational age. Therefore, the study of correlations between radiation exposure to conceptus and organ-specific childhood cancer after birth requires estimation of organ-level radiation dose to the foetus. However, the foetal organ-level radiation dose from CT examinations is rarely reported in literature. In this work, the absorbed doses to 25 foetal organs were evaluated for pregnant patients undergoing CT examinations. The average foetal organ dose normalised to CTDIvol varies between 0.62 and 3.32 (Fig. 5). The absorbed dose is non-uniformly distributed within the different foetal organs. The foetal skeleton and bone marrow receive significantly higher absorbed dose than other soft tissues because the high density of bones results in higher energy deposition in the skeleton. Since radiation exposure of bone marrow and brain in childhood may increase the risk of leukaemia and brain tumours [10], the absorbed dose to foetal bone marrow and foetal brain during CT examinations of pregnant patients is an important matter of concern for the radiologist and the patient.

This work compares Monte Carlo simulated radiation doses with estimates reported by a commercial dose-tracking system, conventional DLP-based absorbed doses and results reported in the literature. Our results are in agreement with observations reported elsewhere confirming that the foetal total body dose has significant correlations with patient size but insignificant correlation with gestational age [11]. However, the absorbed radiation dose of the foetal brain shows significant correlations with patient size, conceptus depth and gestational age while some foetal organs, such as the kidneys, large intestine (LI), small intestine (SI), spleen, urinary bladder (UB), etc., present significant correlations with patient size but insignificant correlation with conceptus depth and gestational age.

Three main factors contribute to the uncertainty on the calculated foetal dose: segmentation of clinical CT images (<5%), image registration of the computational phantom to the clinical CT scan (<30%), which determines the accuracy of the represented anatomical geometry of the patient, and the Monte Carlo calculation uncertainty (<2%), which can be reduced by increasing the simulated particle histories.

The obtained normalised absorbed dose can be used for estimating foetal dose for pregnant patients undergoing CT examinations at other tube current-time product settings since higher mAs values are commonly used to scan heavy or large patients. We are actually expanding the capabilities of the developed framework to other patient populations and different CT scanner models.

The mean foetal total body dose estimated in this work for the investigated patient cohort is consistent with results reported in the literature, but is about 14% lower than the estimates provided by Radimetrics™ dose-tracking system. This discrepancy may be attributed to the utilisation of simplistic stylised pregnant female phantoms and many other approximations and assumptions made by Radimetrics™, which results in an overestimation of the foetal dose.

Among the limitations of this study is that the foetus in the standard pregnant phantoms is facing the downward position while the actual foetal position may vary among patients. This posture variation may affect the dose distributions in foetal organs from CT scan and bring uncertainties in the dose estimation. Another limitation related to the development of patient-specific computational models is the requirement of manual segmentation of the uterus region, which increases the workload for achieving pregnant patient-specific radiation dosimetry. The computation time depends on the number of simulated particles (about 7 min for 2 million simulated photons), the choice of the latter is related to the desired accuracy of the estimates or statistical uncertainty. Overall, the computational time is of the order of 14 h on a single core of Intel Xeon CPU E5-2630 to get less than 2% coefficient of variation. With advances in computer technology, the proposed approach could be implemented in the clinic by subdividing time-consuming simulations on a cluster of computer nodes or geographically distributed platforms taking advantage of the latest developments in grid technology.

Overall, the foetal doses were significantly lower than the consensus level for negligible risks (50 mGy) for deterministic effects. Still, ample justification to perform CT scans for pregnant patients and adequate documentation of the lack of alternative options is still necessary. Since patient size has a noticeable effect on foetal dose estimates, this work shows that patient-specific computational models can be created based on the available standardised computational phantoms of pregnant females to reflect personalised body morphometries. This enables radiologists to perform patient-specific dosimetry for a variety of radiation exposure situations and provide elaborate estimation of organ-level radiation dose to the foetus.