Abstract. Doses from CT examinations are difficult to estimate. However, they are requested more frequently due to the increase in CT examinations. In particular ...
T he British Journal of Radiology, 70 (1997 ), 272–278
© 1997 The British Institute of Radiology
Estimation of fetal and effective dose for CT examinations E J ADAMS, BSc, MSc, D S BRETTLE, BSc, MSc, A P JONES, BSc, MSc, A R HOUNSELL, BSc, MSc and D J MOTT, BSc, MSc North Western Medical Physics, Christie Hospital, Wilmslow Road, Withington, Manchester M20 4BX, UK Abstract. Doses from CT examinations are difficult to estimate. However, they are requested more frequently due to the increase in CT examinations. In particular, fetal dose estimations are frequently required for patients who have discovered, subsequent to the examination, that they were pregnant when the examination was conducted. A computer model has been developed to facilitate such dose calculations. This model combines empirical beam data with anatomical information. The model has been verified using thermoluminescent dosemeter (TLD) readings of internal and surface dose from both phantoms and patients, including intrauterine doses for patients undergoing afterloading gynaecological intracavitary treatment. Although only limited experimental data were available, the results indicate that the model accurately predicts uterine doses within acceptable errors. This approach has been validated for fetal dose estimation. The model was also used in a comparison with the nationally available CT dose data from the National Radiological Protection Board (NRPB). The two models were found to be in agreement for fetal dose estimations.
Recently there has been an increase in the number of CT scanners in the North West of England and also a general increase in awareness of patient doses following publication of the national protocol for patient dose measurements [ 1]. This has led to a rise in the number of enquiries relating to radiation doses to patients resulting from CT examinations. These questions range from quantifying the dose to a specific organ in a given patient to estimating the effective dose to the population from a range of CT examinations. The most common request is for the estimation of dose to the fetus. This request usually arises after a patient discovers that she might have been pregnant at the time of a CT examinations. Because of the high doses involved in CT examinations relative to the majority of diagnostic radiological examinations, the potential risk to the developing fetus is considered to be higher. Advice from the National Radiological Protection Board (NRPB) states that examinations giving uterine doses of a few tens of milligray should be restricted to the early part of the menstrual cycle [ 2]. Hence it is useful to be able to calculate doses from potentially high dose examinations before they are carried out. A computer model was developed to assist in the routine calculation of uterine doses required Received 22 July 1996 and in revised form 29 October 1996, accepted 7 November 1996. Address correspondence to E J Adams, Physics Department, Royal Marsden NHS Trust, Downs Road, Sutton, Surrey SM2 5PT, UK. 272
for fetal dose estimations. This model could also be used to provide information for routine patient dose estimation as well as allowing different protocols to be evaluated prior to the examination. At the onset of this work the majority of published literature on patient radiation doses from CT examinations was limited to the estimation or measurement of the radiation dose under defined conditions in terms of both the CT scanner and the examination parameters [e.g. 3–5 ]. Early Monte Carlo simulations [ 6] had similar limitations. At the same time as our computer model was completed, the NRPB published three reports on CT practice in the UK, in which the dosimetry was based on Monte Carlo simulations of CT examinations [7–9]. Although these data greatly assist in CT dose estimations and have the advantage that they are nationally available, little has been done to clinically verify the data. There are two fundamental differences between the two models. Firstly, the NRPB model relies on a fixed hermaphrodite anatomy whereas our model separates dose from male/female anatomy and allows variation in anatomical structure. Secondly, the NRPB model relies on in-air measurements and a detailed knowledge of the scanner geometry, including details of beam-shaping filters, whereas our method simply requires measurements carried out in a water phantom. This paper presents our model and describes a comparison of predicted and measured doses for the pelvic region. Results will also be presented comparing estimations from both models. T he British Journal of Radiology, March 1997
Dose estimation in CT
Methods Our computer model combines measured beam data with anatomical information for an idealized average patient containing organs considered important in dosimetry. This includes those organs required to calculate effective dose. As the majority of CT scans are 360 ° rotations, the beam data and anatomical model are treated as being circularly symmetrical. Previous experience with measuring CT beam data had indicated that the dose distribution is a slow function of the depth within a right circular cylindrical phantom [ 10]. The anatomical model and the beam data were therefore subdivided into three radial annuli the dimensions of which were chosen so that, for a 28 cm diameter cylindrical phantom, the dose range within each annulus was within ±20% of the surface dose for a range of scanners.
Thermoluminescent dosimetry measurements All thermoluminescent dosemeter (TLD) measurements were performed using lithium fluoride (LiF) TLD rods. The rods were calibrated against a caesium source and a calibration factor of 1.2 applied to convert to diagnostic energies. The caesium source was calibrated using a PTW (PTW-Freiburg, Freiburg, Germany) ionization chamber which was calibrated against a primary standard at the National Physical Laboratory (NPL, Teddington, Middlesex, UK). All doses are quoted as dose in water.
Beam data Beam data were measured using a previously described technique which uses depth dose from within the slice combined with scatter from outside the slice [10]. The dose distribution from a single 1 cm CT section was measured in a 28 cm diameter right circular water filled phantom of length 44 cm using TLD rods. In the z axis (i.e. perpendicular to the scan plane), measurements were taken at 2 cm intervals from the slice plane out to 20 cm, with an additional measurement at 1 cm from the scan plane. At each z position, measurements were made at radii of 0, 4.7, 7.0, 9.3 and 12.6 cm. To achieve significant results outside the scan plane multiple scans were used. Surface dose within the scan plane was measured by placing four equally spaced TLDs around the circumference of the phantom. Surface mounted film was used to verify that the scanner accurately performed a 360° rotation with no under/over scan. Hence, the assumption of a circularly symmetric dose distribution is valid. In addition to the beam data already acquired on the Philips Tomoscan 350 (Philips Medical Systems, Best, Germany) [ 10], T he British Journal of Radiology, March 1997
complete beam data sets were measured on a GE CT PACE (GE Medical Systems, Milwaukee, Wisconsin, USA) and a Philips TX scanner. For all three machines, the radial dose distribution within the scanned area was very similar, which is to be expected as they all have beamshaping filters. The radial dose distribution, normalized to the dose on the surface, from the three machines is illustrated in Figure 1. This figure also includes the boundaries of the three radial compartments into which the dose distribution and anatomical database were subdivided. The range of the relative dose within each compartment was within 30% of the surface dose. Beam data are stored as normalized doses relative to the surface dose. Surface dose itself is also stored, along with the mAs and kV for which the data are appropriate.
Anatomical model The anatomical model was based on Gray’s anatomy [11], Reference Man [12] and an examination of typical CT scans. The percentage of each organ within each 1 cm section was determined and then subdivided into each of the three radial compartments depending again on normal anatomical distribution (Table 1 ). The red bone marrow distribution was based on published data [13 ]. The organ distribution is based on a 170 cm tall adult but it can be re-scaled according to height to give a more accurate estimate of the dose to a specific patient. Re-scaling assumes that the relative positions and sizes of organs, for adults, are proportional to the patient’s height. If further
Figure 1. Radial dose distribution normalized to the surface dose. For clarity the error bars shown are the maximum limits only. C, central; M, mid; and S, surface compartment. X, Philips Tomoscan 350; $, GE CT PACE; and +, Philips TX. 273
97
20 20
93 94
20
95
96
20
98
20
99
Central only Central only Central only 50/50 mid/central 50/50 mid/central Surface only
Radial distribution
E J Adams, D S Brettle, A P Jones et al
90 91
92
modifications to the anatomical model are required to match a specific patient’s anatomy then an editor is provided to facilitate this. The anatomical database includes other information necessary to calculate the effective dose as follows:
89
$ Identification of whether the organ exists in the male or female. This results in the final calculation of the effective dose/effective dose equivalent being able to quote results for either sex, if required.
Bladder Uterus Ovaries L. Intestine S. Intestine Testes
16.1 16.1 16.1 16.1 16.1 16.1 14.0 21.0 23.0 22.0 14.0 6.0 33.3 33.3 33.3 3.81 3.81 5.76 3.07 2.79 2.51 0.76 0.76 0.76 0.89 0.89 0.89 0.89 0.890.89 0.89 6.43 6.43 6.43
87 88 86 85 84 83 82 81 80 79 78 77 76 75 74 73 72 71
Percentage of organ in cm sections at distance from the top of the head Organ
Table 1. Organ distributions within the pelvic region
274
Figure 2. Relative EDE doses for 1 cm slices along the computer model.
$ Tissue weighting factors as required in the calculation of effective dose/effective dose equivalent [14, 15]. Figure 2 shows the relative effective dose equivalent (EDE) for 1 cm slices along the model for a 170 cm male and female which illustrates the anatomically dependent dose distribution. Although the tissue weighting factors used in the model were derived from a mixed population, separating the male and female organs may result in more pertinent risk assessment for individual patients for examinations in the pelvic region. Our model can be modified to give results for an hermaphrodite.
Method of operation Before a dose estimation can be made, the user inputs patient details including height and sex. Examination details are also required, namely machine type, start slice, stop slice, slice width (pre-patient collimated width), number of slices, couch increment and mAs. Modifications to the anatomical distribution of the patient can be made at the patient detail entry stage. The irradiated volume is treated as being composed of 1 cm single slices. To allow for noncontiguous slices, a packing factor is used to spread the dose evenly throughout the scan length; this packing factor is the same as that used by the NRPB [8]. Absorbed doses are calculated by multiplying the anatomical and dose distributions for each of the 1 cm sections and integrating over T he British Journal of Radiology, March 1997
Dose estimation in CT
the length of the scan. Corrections are made for differences between the mAs used for the examination and the beam data collection. Organ equivalent dose is the absorbed dose multiplied by the radiation weighting factor which for photons is 1 [ 14]. Effective dose (ED) and EDE are then calculated by summation of the product of the equivalent dose for each organ and the appropriate tissue weighting factor [ 14, 15 ]. Organ equivalent doses, the ED and the EDE are quoted for the examination as defined and for the examination shifted up and down by 1 cm. This gives the user a feel for the sensitivity of the results relative to the specified anatomy.
Measurements to verify the model Surface and internal dose measurements were made for an abdominal examination on an anthropomorphic Alderson phantom (Radiology Support Devices Inc., California, USA) on both the Philips TX and the GE PACE. The use of an anthropomorphic phantom provided a realistic body shape when checking internal measurements. The examination consisted of 10 mm contiguous slices from slice 20 to 41 inclusive, using 260 mAs and 1350 mAs for the PACE and TX, respectively. TLDs were positioned both within and outside the scanned region, with internal measurements being made at positions corresponding to all three compartments within the region, as well as the central and surface compartments outside the region. TLDs were positioned in anterior, posterior and lateral positions within each compartment. To increase the measured doses and hence reduce the influence of background fluctuation on the reading, surface TLDs were left on the phantom for three scans and internal TLDs for five scans. These spot dose measurements were compared with the model prediction for the dose in the appropriate compartment and slice. Surface dose measurements were also made within the scan region for 24 patients undergoing routine CT examinations of the abdomen and pelvis. For each patient one TLD was placed on the anterior/posterior surface and one on a lateral surface. These were then averaged to give a mean surface dose. Additionally, a series of uterine dose
measurements was made on a small number of patients being treated for carcinoma of the cervix using the Nucletron Selectron afterloading system (Nucletron, Nucletron House, Tattenhall, Chester, UK). These patients are routinely scanned to ensure correct positioning of the steel intrauterine tube and so TLD rods were inserted into the tube prior to the scan. To increase the measured dose, one set of TLDs was used for five patients. Two further sets were used for individual patients. All patient dose measurements were performed on a GE CT PACE.
Comparison of models To compare the results of our computer model with the NRPB data, the CTDOSE program was used (National Radiation Laboratory, Christchurch, New Zealand). This program provides a user interface to the NRPB data [16 ], allowing examination details to be entered and doses to be calculated. Equivalent examinations were entered into both programs for typical lower body examinations. The parameters were for mean examinations on the GE CT PACE derived from the NRPB report [8 ] and are tabulated in Table 2. The predictions from the NRPB model were obtained using the specific CTDI [ 17] for our scanner, measured using the method described in the NRPB report [8 ]. The results for uterus and EDE doses from both programs were then compared. Predicted doses from the NRPB model were also obtained for comparison with the measured uterine doses mentioned above.
Results The anthropomorphic phantom dose measurements are illustrated in Figures 3 and 4 and the patient skin dose data in Figure 5. Dotted lines indicate ±20% deviation from the predicted doses. Table 3 shows the results of the uterine dose measurements. All measured dose values have been increased by an empirical factor of 1.8 to correct for attenuation in the steel intrauterine tube. This correction factor was obtained by comparing measured doses in a phantom with TLDs either within the steel tube or simply encased in plastic tubing.
Table 2. Examinations used to compare data obtained for the GE CT PACE Examination
mAs
Number of slices
Slice width (mm)
Slice increment (mm)
Reference plane
Start/stop slice no.
Lumbar spine Pelvis Abdomen Kidneys Liver Pancreas
408 326 340 322 332 358
16 10 10 7 10 4
5 10 10 10 10 10
5 10 10 10 10 10
270 135 345 325 350 365
23–31 8–18 29–39 29–36 30–40 34–38
T he British Journal of Radiology, March 1997
275
E J Adams, D S Brettle, A P Jones et al
Figure 3. Alderson phantom dose measurements on the GE CT PACE. The radial compartment from which the reading was taken is indicated by: X, central compartment; +, mid compartment; and $, surface compartment. The error bars shown are the TLD reading errors of ±15%.
Figure 5. Patient skin dose comparison with predicted data for the GE CT PACE. The error bars shown are the TLD reading errors of ±15%.
of examinations in the abdominal and pelvic regions are shown in Figures 6 and 7. Again, a ±20% tolerance is indicated by dotted lines.
Discussion
Figure 4. Alderson phantom dose measurements on the Philips TX. The radial compartment from which the reading was taken is indicated by: X, central compartment; +, mid compartment; and $, surface compartment. The error bars shown are the TLD reading errors of ±15%.
Two predictions are given for the second of the single doses. The initial estimate was based on the standard anatomy, with the uterus position being defined simply by the scaling according to height. However, inspection of this position with reference to the CT images revealed it to be incorrect. The location was therefore altered according to the actual position of the uterus within the patient and a second prediction was obtained. The comparison between the two sets of predicted values of EDE and uterine doses for a range 276
The uncertainty in individual TLD measurements for a known energy, including random and systematic errors, is between 10 and 15% (at the 95% confidence limit). Error analysis involved with the averaging procedures used in the beam data collection result in ~7% uncertainty in the beam data. Combining this with the variation within each compartment means that, were spot doses to be measured in the water phantom, agreement within ±15% would be expected. Since the majority of organs will be spread through at least one compartment, the variation will average out to some extent. So, assuming agreement with the anatomical model, organ equivalent doses will generally have a smaller uncertainty. However, the largest source of uncertainty is in the anatomical modelling. It is difficult to estimate organ positions to better than 1 cm; this was the reason for treating the anatomy as 1 cm sections. A movement of the organ relative to the scanned volume by 1 cm can cause more than a two-fold change in dose. With the facility to match the anatomy to the patient, it was felt that the largest of these errors could be avoided; however, it is difficult to quantify the magnitude of the uncertainty introduced by the anatomical model. The comparison of measured and predicted doses for examinations on the anthropomorphic phantom show agreement within ±20% in approximately 50% of the measurements made, with the greatest deviations from this being for T he British Journal of Radiology, March 1997
Dose estimation in CT Table 3. Comparison of predicted uterine doses (GE CT PACE) with measurements in Selectron patients Measured dose (mGy)
Predicted dose (mGy) NRPB
Authors’ model −1 cm
Cumulative dose for five patients
40.5±6.1
29.4
Single patients Patient A: Standard anatomy Patient B: Standard anatomy Patient B: Patient anatomy
7.8±1.2 6.4±1.0
5.0 3.4
Figure 6. Comparison with NRPB EDEs for a range of examinations (see Table 2 ) on the GE CT PACE.
Figure 7. Comparison with NRPB uterus doses for a range of examinations (see Table 2) on the GE CT PACE.
doses below 10 mGy. The discrepancies may be attributable to the variation between the shape and attenuation coefficients of the anthropomorphic phantom and the water phantom as well T he British Journal of Radiology, March 1997
+1 cm
0 40.3
7.2 2.6 5.7
6.5 2.3 7.1
5.7 2.2 6.3
as inaccuracies in the TLD readings for low exposures near the background level. It should be noted that the anthropomorphic phantom may not necessarily accurately represent genuine anatomy with respect to attenuation coefficients and shape. As such, these discrepancies may not be as great when genuine anatomy is considered. Patient skin dose measurements showed better agreement with the predicted values except for one patient for whom the measured dose was only 50% of the prediction. However, the measured dose was approximately half of that measured for patients undergoing comparable examinations. This suggests that it was the TLD reading, rather than the model prediction, which caused the discrepancy. At present we have been unable to ascertain the cause of this anomaly. Further verification of the model was by comparison of the predicted uterus doses with those acquired in situ from patients undergoing afterloading gynaecological intracavitary treatment. These results were also in good agreement except for one single patient measurement which was lower than predicted. After examining the true patient anatomy from the CT images the model uterus was moved by 5 cm to match the real patient anatomy and better results were achieved. This clearly illustrates the advantage of being able to modify the model anatomy to match the patient reducing the discrepancy from 64% to 11% in this instance. The two sets of predictions of uterine dose for these examinations show some discrepancies. This is most likely to be due to differences in the anatomy between the two models. Because this type of examination is designed to look at the uterus itself, the scan length is short (typically