Home
Search
Collections
Journals
About
Contact us
My IOPscience
Database of normalised computed tomography dose index for retrospective CT dosimetry
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2014 J. Radiol. Prot. 34 363 (http://iopscience.iop.org/0952-4746/34/2/363) View the table of contents for this issue, or go to the journal homepage for more
Download details: IP Address: 128.6.218.72 This content was downloaded on 08/06/2014 at 06:25
Please note that terms and conditions apply.
Society for Radiological Protection
Journal of Radiological Protection
J. Radiol. Prot. 34 (2014) 363–388
doi:10.1088/0952-4746/34/2/363
Database of normalised computed tomography dose index for retrospective CT dosimetry Eunah Lee, Stephanie Lamart, Mark P Little and Choonsik Lee Radiation Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institute of Health, Bethesda, MD 20852, USA E-mail: [email protected] Received 6 September 2013, revised 16 December 2013 Accepted for publication 9 January 2014 Published 14 April 2014 Abstract
Volumetric computed tomography dose index (CTDIvol ) is an important dose descriptor to reconstruct organ doses for patients combined with the organ dose calculated from computational human phantoms coupled with Monte Carlo transport techniques. CTDIvol can be derived from weighted CTDI (CTDIw ) normalised to the tube current–time product (mGy/100 mAs), using knowledge of tube current–time product (mAs), tube potential (kVp), type of CTDI phantoms (head or body), and pitch. The normalised CTDIw is one of the characteristics of a CT scanner but not readily available from the literature. In the current study, we established a comprehensive database of normalised CTDIw values based on multiple data sources: the ImPACT dose survey from the United Kingdom, the CT-Expo dose calculation program, and surveys performed by the US Food and Drug Administration (FDA) and the National Lung Screening Trial (NLST). From the sources, the CTDIw values for a total of 68, 138, 30, and 13 scanner model groups were collected, respectively. The different scanner groups from the four data sources were sorted and merged into 162 scanner groups for eight manufacturers including General Electric (GE), Siemens, Philips, Toshiba, Elscint, Picker, Shimadzu, and Hitachi. To fill in missing CTDI values, a method based on exponential regression analysis was developed based on the existing data. Once the database was completed, two different analyses of data variability were performed. First, we averaged CTDI values for each scanner in the different data sources and analysed the variability of the average CTDI values across the different scanner models within a given manufacturer. Among the four major manufacturers, Toshiba and Philips showed the greatest coefficient of variation (COV) (=standard deviation/mean) for the head and body normalised CTDIw values, 39% and 54%, respectively. Second, the variation across the different data sources was analysed for CT scanners where more than two data sources were involved. The 0952-4746/14/020363+26$33.00
c 2014 IOP Publishing Ltd
Printed in the UK
363
J. Radiol. Prot. 34 (2014) 363
E Lee et al
CTDI values for the scanners from Siemens showed the greatest variation across the data sources, being about four times greater than the variation of Toshiba scanners. The established CTDI database will be used for the reconstruction of CTDIvol and then the estimation of individualised organ doses for retrospective patient cohorts in epidemiologic studies. Keywords: computed tomography dose index, CT dose survey, dose reconstruction
1. Introduction
Individualised organ dose estimation is required for epidemiologic studies to accurately analyse the risk of late effects from radiation exposure for patients undergoing computed tomography (CT) examinations. Organ dose is usually calculated from a comprehensive Monte Carlo transport simulation where computational human phantoms and CT scanner models are incorporated. Organ dose varies depending on parameters associated with both patient anatomy and CT scanners. Several studies have revealed that there is considerable variation among CT scanners, demonstrated by the variation of volumetric CT dose index (CTDIvol ) which represents x-ray output from a CT scanner (Cody et al 2010, Lee et al 2011, Turner et al 2010). It is reported that CTDIvol can be used as a normalisation factor to produce scanner-independent organ doses which in turn can be used as a multiplying factor for given CT scanners to readily obtain scanner-specific organ dose. Therefore, CTDIvol is one of the key parameters which have been used in organ dose calculation for epidemiologic studies of patients undergoing CT examinations (Berrington de Gonz´alez et al 2008, 2009, Kim et al 2012, 2009, Pearce et al 2012). The normalised CTDIw is derived from the measurements of CTDI100 using a 100 mm-long pencil ionisation chamber inserted in the central and in the four peripheral holes of cylindrical solid phantoms made of Poly Methyl Methacrylate (PMMA) with diameters of 16 cm and 32 cm representing the head and the body of a patient, respectively (McNitt-Gray 2002). Each dose measurement for a given tube potential (kVp) is conducted for a single axial rotation perpendicular to and at mid-length of the cylinder and ion chamber. The normalised CTDIw (mGy/100 mAs) is calculated using the following formula: Normalized CTDIw = 31 CTDI100,center + 23 CTDI100,periphery
(1)
where CTDI100,center is the dose measured per 100 mAs at the centre of the CTDI phantom and CTDI100,periphery is the average dose per 100 mAs from the four measurements conducted at the periphery of the phantoms. The volumetric CTDI, CTDIvol (mGy), is then derived from the normalised CTDIw : CTDIvol (mGy) =
Normalized CTDIw I t Pitch 100
(2)
where I is tube current (mA) and t is a single rotation time (s). Although CTDIvol is readily obtained from the CT dose sheet or the digital imaging and communication in medicine (DICOM) header in modern CT examinations, the value is rarely available for patients who received CT scans decades ago, who are commonly the target patient cohort in retrospective cohort-based epidemiologic studies (Pearce et al 2012). The assessment of CTDIvol is not generally feasible when the CT scanners of interest are no longer available. The CTDIvol must be reconstructed using technical parameters reported in the literature and 364
J. Radiol. Prot. 34 (2014) 363
E Lee et al
survey reports. For that purpose, normalised CTDIw for all CT scanners used in the patient cohort is required together with the values of tube current–time product (mAs), tube potential (kVp), pitch, and CTDI phantom type (head or body) as described in equation (2). Whereas other technical parameters vary on a case by case basis, the normalised CTDIw is one of the scanner-specific characteristics. Although some data is available on scanner manuals and publications, the values or the manuals themselves tend to be scattered and not easy to collect. The current study compiled the normalised CTDIw for different scanner models, tube potentials, and CTDI phantoms from several different resources and to establish a comprehensive data matrix which can form the basis of organ dose reconstruction for retrospective epidemiologic studies. Because the CTDI values were not available for all tube potentials, a method to derive missing values was developed through regression analysis based on existing CTDI values. The variation of the normalised CTDIw across the different scanner models within a given manufacturer as well as across the different data sources was analysed. Average values over the multiple data sources were calculated for different scanner models from multiple scanner manufacturers with four tube potentials and are presented in table format.
2. Materials and methods 2.1. Sources of normalised CTDIw data
The normalised CTDIw values were obtained from four different data sources: the ImPACT dose survey, CT-Expo software, and two CT dose surveys performed independently by the US Food and Drug Administration (FDA) and the National Lung Screening Trial (NLST). The Excel spreadsheet titled ‘ImPACT dose survey CTDI Results, 2000–2004 ImPACT Group’ was downloaded from the website (ImPACT Web site Accessed June 20, 2013). The spreadsheet contains CTDIair , CTDI100,center , and CTDI100,periphery normalised by 100 mAs for different tube potentials ranging from 80 to 140 kVp. The dose survey covers a total of 68 CT scanner models from seven different manufacturers including GE, Philips, Siemens, Toshiba, Elscint, Picker, and Shimadzu. We calculated the normalised CTDIw (mGy/100 mAs) from CTDI100,center and CTDI100,periphery . Another set of the normalised CTDIw was calculated from CT-Expo version 2.1 software package (Stamm and Nagel 2002). CT-Expo provides the data collected in the German survey on CT practice in 1999 and provides, among other parameters, CTDIw when the user selects a category for patient age (baby, child, or adult), gender, scan range, scanner manufacturer and model, and scanner parameters such as mAs, kVp, and pitch. We calculated the normalised CTDIw for a total of 138 CT scanners from eight manufacturers (GE, Philips, Siemens, Toshiba, Elscint, Picker, Shimadzu, and Hitachi) by entering the parameters required by the software. The CTDI phantom (head or body) was automatically selected based on the scan coverage in the adult computational phantoms. The NEXT survey was conducted by the US FDA from 2000 to 2001 across the US (Stern 2007). The report provides the mean normalised CTDI100,central and CTDI100,periphery values (mGy/100 mAs) for a total of 30 different CT scanner models from six manufacturers (GE, Philips, Siemens, Toshiba, Elscint, and Picker). The CTDI values were measured only using the CTDI head phantom in more than 200 CT scanners across the country for tube potentials between 120 and 140 kVp. The NEXT survey grouped scanner models into the categories based on those provided by the ImPACT dose survey. The normalised CTDIw was calculated from the CTDI100,center and CTDI100,periphery . The medical physics working group of the NLST published the normalised CTDIw values (mGy/mAs) based on 247 measurements on 96 CT scanners from hospitals involved in the 365
J. Radiol. Prot. 34 (2014) 363
E Lee et al
trial during 2002–2007 (Cody et al 2010). The normalised CTDIw values measured only from CTDI body phantom and for the tube potential of 120 kVp are available in the publication for 13 CT scanner models from four manufacturers (GE, Philips, Siemens, and Toshiba). The normalised CTDIw data (mGy/mAs) was entered into our database after multiplying 100 to convert to mGy/100 mAs. From the four data sources, the normalised CTDIw for the collimation width of 10 mm were collected for four different tube potentials (80, 100, 120, and 140 kVp) and for the eight manufacturers: GE, Philips, Siemens, Toshiba, Elscint, Picker, Shimadzu, and Hitachi. Each survey presented the data by grouping the scanner models differently. The normalised CTDIw for a total of 68, 138, 30, and 13 scanner model groups were collected from ImPACT dose survey, CT-Expo software, NEXT survey, and NLST survey, respectively.
2.2. Data merging
Once we collected the normalised CTDIw values for head and body CTDI phantoms for tube potentials ranging from 80 to 140 kVp from the four different data sources, we merged the values to establish a single comprehensive database. The simplest case was that each data source provided a CTDI value for a given CT scanner using the same CT scanner name. For example, the same model group, ‘GE 9800 series’, was found in both the ImPACT dose survey and the CT-Expo program. Two sets of CTDI values were collected from each data source and average values were calculated for the merged database. However, in most cases, different data sources do not group CT scanners in the same way. For example, the ImPACT dose survey spreadsheet provides one particular set of normalised CTDIw values for the group of scanner models named ‘GE HighSpeed FX/i, LX/i’ and another set of CTDIw values for the group of models named ‘GE HighSpeed ZX/i, NX/i’. The CT-Expo program grouped scanners differently, assigning identical sets of CTDI values to two different categories, ‘GE HighSpeed DX/i, FX/i, LX/i, ZX/i’ and ‘GE HighSpeed NX/i, -Pro’. The NEXT survey has two different scanner groups, ‘GE FX/i, LX/i’ and ‘DX/i’. If we kept all scanner groups in the final database, then it would be confusing to assign proper CTDI values when a given scanner is ‘GE HighSpeed NX/i’, for example, because the model is listed in different groupings of scanners with different CTDI values. To reduce the problem, the group category in each data source was broken into single scanner model. The CTDI value for each scanner model was collected from different sources. If multiple sources provide different CTDI values, average CTDI was calculated. Then the scanners showing identical CTDI values were regrouped into a new category with a single set of CTDI values in the final database. The merging process is described in table 1.
2.3. Missing values
Since some data sources provided the normalised CTDIw values for limited tube potentials and only for either the head or body CTDI phantom, exponential regression models were developed between the tube potentials and CTDI values for GE, Philips, Siemens, Toshiba, Shimadzu, and Elscint. The database for the manufacturers, Picker and Hitachi, did not contain any missing values. CTDI values for scanner models where all head and body CTDI values are available for the four tube potentials (80, 100, 120, and 140 kVp) were selected and used in the regression analysis, using MATLABTM . An exponential model as defined by y = A × e Bx
(3) 366
J. Radiol. Prot. 34 (2014) 363
E Lee et al
Table 1. An example of the procedure to merge different scanner model groups from
different data sources for CT scanners manufactured by GE. ImPACT Model group CTDIa
Model group
CT-Expo CTDI
NEXT Model group
CTDI
FX/i, LX/i ZX/i, NX/i
DX/i, FX/i, LX/i, ZX/i NX/i, -Pro
15.2 15.2
FX/i, LX/i DX/i
20.9 13.2
18.9 14.7
Breakdown Model
ImPACT
FX/i LX/i ZX/i NX/i DX/i -Pro
18.9 18.9 14.7 14.7
Final database
CTDI CT-Expo NEXT
Average
Model group
CTDI
15.2 15.2 15.2 15.2 15.2 15.2
18.3 18.3 14.9 14.9 14.2 15.2
FX/i, LX/i ZX/i, NX/i DX/i -Pro
18.3 14.9 14.2 15.2
20.9 20.9 13.2
a Normalised CTDI only for head phantom and the tube potential of 120 kVp was included for illustrative w
purposes.
where x is tube potential and y is the normalised CTDIw , was fitted to the data by least squares. Based on the analysis, the missing CTDI were calculated using the available values and the tube potentials for the missing values. 2.4. Analysis of CTDI variability
To understand the variability of the normalised CTDIw values, we analysed the final database where the merging process and the calculation of missing values were completed. First, we analysed the variability among scanner models in a single manufacturer. We picked the normalised CTDIw values for head and body CTDI phantoms for 120 kVp and calculated the mean, standard deviation, and coefficient of variation (COV) across the different scanner models for the four major scanner manufacturers: GE, Philips, Siemens, and Toshiba. Manufacturers such as Picker, Elscint, Shimadzu, and Hitachi, which contributed less than ten CT scanner models to the final merged data matrix, were excluded from the analysis. Second, we analysed the variability across the four different data sources. Scanner models with the normalised CTDIw values for head and body phantoms from more than two sources were included in the analysis to allow the investigation of the inter-source variation. Like the first analysis, only four major manufacturers, GE, Philips, Siemens, and Toshiba, were included in the analysis. The mean, standard deviation, and COV of the normalised CTDIw values for the head and body CTDI phantoms for 120 kVp were calculated and analysed. 3. Results and discussion 3.1. Data collection
The normalised CTDIw values were collected or calculated from the four data sources: ImPACT dose survey, CT-Expo program, and NEXT and NLST surveys. The number of CT scanner groups abstracted from the four data sources are summarised in table 2. Through the merging process described in table 1, a total of 249 original categories were regrouped into 162 categories. CT-Expo program contributed the greatest number of CT scanner groups (n = 138) compared to the NLST survey where 13 scanner groups were available. The NEXT and NLST 367
J. Radiol. Prot. 34 (2014) 363
E Lee et al
Figure 1. Examples of the exponential regression analysis for (A) GE LightSpeed VCT
and (B) GE Discovery CT HD750 scanners.
surveys only provided head and body CTDI values, respectively, whereas the ImPACT dose survey and CT-Expo program provided both head and body CTDI values. 3.2. Missing values
To calculate missing CTDI values, an exponential regression model (as in equation (3)) was used, applied to the CT scanner groups where the CTDI values are available for all tube potentials considered in this study, namely 80, 100, 120, and 140 kVp. The calculation was only performed for manufacturers apart from Picker and Hitachi that did not contain any missing values. Figure 1 shows an example of the exponential regression analysis for (a) GE LightSpeed VCT and (b) GE Discovery CT HD750 for head/body CTDI phantoms and the tube potential of 80–140 kVp. Although the two scanners are from the same manufacturer, the parameter A 368
J. Radiol. Prot. 34 (2014) 363
E Lee et al
Table 2. The number of CT scanner group for eight different manufacturers collected
from the four data sources. The number of CT scanner models after data merging is included in the final column. Manufacturer GE Philips Siemens Toshiba Elscint Picker Shimadzu Hitachi Total
ImPACT 17 13 16 16 2 3 1 68
CT-Expo
NEXT
NLST
27 17 35 34 7 7 6 5 138
10 2 3 8 5 2
5 3 3 2
30
13
Merged 27 25 38 42 10 9 6 5 162
Table 3. Average and standard deviation of the parameter B of the exponential regression model for CTDIw of head and body phantoms and for different manufacturers.
Manufacturer
GE Average SD
Philips Average SD
Siemens Average SD
CTDIw Head CTDIw Body
0.0197 0.0212
0.0216 0.0226
0.0219 0.0230
Manufacturer
Toshiba Average SD
Elscint Average SD
Picker Average SD
CTDIw Head CTDIw Body
0.0206 0.0218
0.0199 0.0216
0.0185 0.0196
0.0013 0.0013
0.0009 0.0007
0.0018 0.0016
0.0003 0.0011
0.0022 0.0015
0.0015 0.0021
shown in equation (3) of the exponential regression curve is significantly different: 1.8 and 1.3 for head phantom in LightSpeed VCT and Discovery CT HD750, respectively. However, the parameter B shown in equation (3) for the head phantom is identical, 0.0217, and the values for body phantom are very close, 0.0225 and 0.0230. Based on the observation, the average and standard deviation of the parameter B for head and body CTDI values were calculated across the different scanner models in six manufacturers to obtain representative values of the parameter B. Since at least one CTDI value is available (mostly for 120 kVp), the representative parameter B was then used to derive other CTDI values by using equation (3) and the available CTDI values for 120 kVp. The mean and standard deviation of the parameter B for head and body CTDI values were calculated across the different scanner models in six manufacturers to obtain the representative parameters B (table 3). To investigate if the parameters B differ by manufacturer, or by phantom body part (head versus body), we fitted linear models to this data on fitted exponential parameters, in which; Bi = α1 Manufactureri + α2 CDTI phantomi + εi .
(4)
Analysis of variance was assessed in the standard way using the F-test (Rao 2002). Table 4 demonstrates that all possible comparisons (of manufacturer within each CTDI phantom and across CTDI phantom types, and CTDI phantoms adjusted for manufacturer) were highly statistically significant ( p < 0.0001). This was also the case when the slope parameters B were log-transformed (results not shown). Based on the results, we decided to use a different parameter B for each manufacturer and CTDI phantom type. 369
J. Radiol. Prot. 34 (2014) 363
E Lee et al
Table 4. Analysis of variance decomposition of fitted slope parameters B from model
described in equation (4). Data (phantom body part)
Test for difference by category
Head only Body only Head and body Head and body Head and body
Manufacturer Manufacturer Manufacturer Manufacturer adjusted for phantom body part Phantom body part adjusted for manufacturer
p-value (via F-test)
Search
Collections
Journals
About
Contact us
My IOPscience
Database of normalised computed tomography dose index for retrospective CT dosimetry
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2014 J. Radiol. Prot. 34 363 (http://iopscience.iop.org/0952-4746/34/2/363) View the table of contents for this issue, or go to the journal homepage for more
Download details: IP Address: 128.6.218.72 This content was downloaded on 08/06/2014 at 06:25
Please note that terms and conditions apply.
Society for Radiological Protection
Journal of Radiological Protection
J. Radiol. Prot. 34 (2014) 363–388
doi:10.1088/0952-4746/34/2/363
Database of normalised computed tomography dose index for retrospective CT dosimetry Eunah Lee, Stephanie Lamart, Mark P Little and Choonsik Lee Radiation Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institute of Health, Bethesda, MD 20852, USA E-mail: [email protected] Received 6 September 2013, revised 16 December 2013 Accepted for publication 9 January 2014 Published 14 April 2014 Abstract
Volumetric computed tomography dose index (CTDIvol ) is an important dose descriptor to reconstruct organ doses for patients combined with the organ dose calculated from computational human phantoms coupled with Monte Carlo transport techniques. CTDIvol can be derived from weighted CTDI (CTDIw ) normalised to the tube current–time product (mGy/100 mAs), using knowledge of tube current–time product (mAs), tube potential (kVp), type of CTDI phantoms (head or body), and pitch. The normalised CTDIw is one of the characteristics of a CT scanner but not readily available from the literature. In the current study, we established a comprehensive database of normalised CTDIw values based on multiple data sources: the ImPACT dose survey from the United Kingdom, the CT-Expo dose calculation program, and surveys performed by the US Food and Drug Administration (FDA) and the National Lung Screening Trial (NLST). From the sources, the CTDIw values for a total of 68, 138, 30, and 13 scanner model groups were collected, respectively. The different scanner groups from the four data sources were sorted and merged into 162 scanner groups for eight manufacturers including General Electric (GE), Siemens, Philips, Toshiba, Elscint, Picker, Shimadzu, and Hitachi. To fill in missing CTDI values, a method based on exponential regression analysis was developed based on the existing data. Once the database was completed, two different analyses of data variability were performed. First, we averaged CTDI values for each scanner in the different data sources and analysed the variability of the average CTDI values across the different scanner models within a given manufacturer. Among the four major manufacturers, Toshiba and Philips showed the greatest coefficient of variation (COV) (=standard deviation/mean) for the head and body normalised CTDIw values, 39% and 54%, respectively. Second, the variation across the different data sources was analysed for CT scanners where more than two data sources were involved. The 0952-4746/14/020363+26$33.00
c 2014 IOP Publishing Ltd
Printed in the UK
363
J. Radiol. Prot. 34 (2014) 363
E Lee et al
CTDI values for the scanners from Siemens showed the greatest variation across the data sources, being about four times greater than the variation of Toshiba scanners. The established CTDI database will be used for the reconstruction of CTDIvol and then the estimation of individualised organ doses for retrospective patient cohorts in epidemiologic studies. Keywords: computed tomography dose index, CT dose survey, dose reconstruction
1. Introduction
Individualised organ dose estimation is required for epidemiologic studies to accurately analyse the risk of late effects from radiation exposure for patients undergoing computed tomography (CT) examinations. Organ dose is usually calculated from a comprehensive Monte Carlo transport simulation where computational human phantoms and CT scanner models are incorporated. Organ dose varies depending on parameters associated with both patient anatomy and CT scanners. Several studies have revealed that there is considerable variation among CT scanners, demonstrated by the variation of volumetric CT dose index (CTDIvol ) which represents x-ray output from a CT scanner (Cody et al 2010, Lee et al 2011, Turner et al 2010). It is reported that CTDIvol can be used as a normalisation factor to produce scanner-independent organ doses which in turn can be used as a multiplying factor for given CT scanners to readily obtain scanner-specific organ dose. Therefore, CTDIvol is one of the key parameters which have been used in organ dose calculation for epidemiologic studies of patients undergoing CT examinations (Berrington de Gonz´alez et al 2008, 2009, Kim et al 2012, 2009, Pearce et al 2012). The normalised CTDIw is derived from the measurements of CTDI100 using a 100 mm-long pencil ionisation chamber inserted in the central and in the four peripheral holes of cylindrical solid phantoms made of Poly Methyl Methacrylate (PMMA) with diameters of 16 cm and 32 cm representing the head and the body of a patient, respectively (McNitt-Gray 2002). Each dose measurement for a given tube potential (kVp) is conducted for a single axial rotation perpendicular to and at mid-length of the cylinder and ion chamber. The normalised CTDIw (mGy/100 mAs) is calculated using the following formula: Normalized CTDIw = 31 CTDI100,center + 23 CTDI100,periphery
(1)
where CTDI100,center is the dose measured per 100 mAs at the centre of the CTDI phantom and CTDI100,periphery is the average dose per 100 mAs from the four measurements conducted at the periphery of the phantoms. The volumetric CTDI, CTDIvol (mGy), is then derived from the normalised CTDIw : CTDIvol (mGy) =
Normalized CTDIw I t Pitch 100
(2)
where I is tube current (mA) and t is a single rotation time (s). Although CTDIvol is readily obtained from the CT dose sheet or the digital imaging and communication in medicine (DICOM) header in modern CT examinations, the value is rarely available for patients who received CT scans decades ago, who are commonly the target patient cohort in retrospective cohort-based epidemiologic studies (Pearce et al 2012). The assessment of CTDIvol is not generally feasible when the CT scanners of interest are no longer available. The CTDIvol must be reconstructed using technical parameters reported in the literature and 364
J. Radiol. Prot. 34 (2014) 363
E Lee et al
survey reports. For that purpose, normalised CTDIw for all CT scanners used in the patient cohort is required together with the values of tube current–time product (mAs), tube potential (kVp), pitch, and CTDI phantom type (head or body) as described in equation (2). Whereas other technical parameters vary on a case by case basis, the normalised CTDIw is one of the scanner-specific characteristics. Although some data is available on scanner manuals and publications, the values or the manuals themselves tend to be scattered and not easy to collect. The current study compiled the normalised CTDIw for different scanner models, tube potentials, and CTDI phantoms from several different resources and to establish a comprehensive data matrix which can form the basis of organ dose reconstruction for retrospective epidemiologic studies. Because the CTDI values were not available for all tube potentials, a method to derive missing values was developed through regression analysis based on existing CTDI values. The variation of the normalised CTDIw across the different scanner models within a given manufacturer as well as across the different data sources was analysed. Average values over the multiple data sources were calculated for different scanner models from multiple scanner manufacturers with four tube potentials and are presented in table format.
2. Materials and methods 2.1. Sources of normalised CTDIw data
The normalised CTDIw values were obtained from four different data sources: the ImPACT dose survey, CT-Expo software, and two CT dose surveys performed independently by the US Food and Drug Administration (FDA) and the National Lung Screening Trial (NLST). The Excel spreadsheet titled ‘ImPACT dose survey CTDI Results, 2000–2004 ImPACT Group’ was downloaded from the website (ImPACT Web site Accessed June 20, 2013). The spreadsheet contains CTDIair , CTDI100,center , and CTDI100,periphery normalised by 100 mAs for different tube potentials ranging from 80 to 140 kVp. The dose survey covers a total of 68 CT scanner models from seven different manufacturers including GE, Philips, Siemens, Toshiba, Elscint, Picker, and Shimadzu. We calculated the normalised CTDIw (mGy/100 mAs) from CTDI100,center and CTDI100,periphery . Another set of the normalised CTDIw was calculated from CT-Expo version 2.1 software package (Stamm and Nagel 2002). CT-Expo provides the data collected in the German survey on CT practice in 1999 and provides, among other parameters, CTDIw when the user selects a category for patient age (baby, child, or adult), gender, scan range, scanner manufacturer and model, and scanner parameters such as mAs, kVp, and pitch. We calculated the normalised CTDIw for a total of 138 CT scanners from eight manufacturers (GE, Philips, Siemens, Toshiba, Elscint, Picker, Shimadzu, and Hitachi) by entering the parameters required by the software. The CTDI phantom (head or body) was automatically selected based on the scan coverage in the adult computational phantoms. The NEXT survey was conducted by the US FDA from 2000 to 2001 across the US (Stern 2007). The report provides the mean normalised CTDI100,central and CTDI100,periphery values (mGy/100 mAs) for a total of 30 different CT scanner models from six manufacturers (GE, Philips, Siemens, Toshiba, Elscint, and Picker). The CTDI values were measured only using the CTDI head phantom in more than 200 CT scanners across the country for tube potentials between 120 and 140 kVp. The NEXT survey grouped scanner models into the categories based on those provided by the ImPACT dose survey. The normalised CTDIw was calculated from the CTDI100,center and CTDI100,periphery . The medical physics working group of the NLST published the normalised CTDIw values (mGy/mAs) based on 247 measurements on 96 CT scanners from hospitals involved in the 365
J. Radiol. Prot. 34 (2014) 363
E Lee et al
trial during 2002–2007 (Cody et al 2010). The normalised CTDIw values measured only from CTDI body phantom and for the tube potential of 120 kVp are available in the publication for 13 CT scanner models from four manufacturers (GE, Philips, Siemens, and Toshiba). The normalised CTDIw data (mGy/mAs) was entered into our database after multiplying 100 to convert to mGy/100 mAs. From the four data sources, the normalised CTDIw for the collimation width of 10 mm were collected for four different tube potentials (80, 100, 120, and 140 kVp) and for the eight manufacturers: GE, Philips, Siemens, Toshiba, Elscint, Picker, Shimadzu, and Hitachi. Each survey presented the data by grouping the scanner models differently. The normalised CTDIw for a total of 68, 138, 30, and 13 scanner model groups were collected from ImPACT dose survey, CT-Expo software, NEXT survey, and NLST survey, respectively.
2.2. Data merging
Once we collected the normalised CTDIw values for head and body CTDI phantoms for tube potentials ranging from 80 to 140 kVp from the four different data sources, we merged the values to establish a single comprehensive database. The simplest case was that each data source provided a CTDI value for a given CT scanner using the same CT scanner name. For example, the same model group, ‘GE 9800 series’, was found in both the ImPACT dose survey and the CT-Expo program. Two sets of CTDI values were collected from each data source and average values were calculated for the merged database. However, in most cases, different data sources do not group CT scanners in the same way. For example, the ImPACT dose survey spreadsheet provides one particular set of normalised CTDIw values for the group of scanner models named ‘GE HighSpeed FX/i, LX/i’ and another set of CTDIw values for the group of models named ‘GE HighSpeed ZX/i, NX/i’. The CT-Expo program grouped scanners differently, assigning identical sets of CTDI values to two different categories, ‘GE HighSpeed DX/i, FX/i, LX/i, ZX/i’ and ‘GE HighSpeed NX/i, -Pro’. The NEXT survey has two different scanner groups, ‘GE FX/i, LX/i’ and ‘DX/i’. If we kept all scanner groups in the final database, then it would be confusing to assign proper CTDI values when a given scanner is ‘GE HighSpeed NX/i’, for example, because the model is listed in different groupings of scanners with different CTDI values. To reduce the problem, the group category in each data source was broken into single scanner model. The CTDI value for each scanner model was collected from different sources. If multiple sources provide different CTDI values, average CTDI was calculated. Then the scanners showing identical CTDI values were regrouped into a new category with a single set of CTDI values in the final database. The merging process is described in table 1.
2.3. Missing values
Since some data sources provided the normalised CTDIw values for limited tube potentials and only for either the head or body CTDI phantom, exponential regression models were developed between the tube potentials and CTDI values for GE, Philips, Siemens, Toshiba, Shimadzu, and Elscint. The database for the manufacturers, Picker and Hitachi, did not contain any missing values. CTDI values for scanner models where all head and body CTDI values are available for the four tube potentials (80, 100, 120, and 140 kVp) were selected and used in the regression analysis, using MATLABTM . An exponential model as defined by y = A × e Bx
(3) 366
J. Radiol. Prot. 34 (2014) 363
E Lee et al
Table 1. An example of the procedure to merge different scanner model groups from
different data sources for CT scanners manufactured by GE. ImPACT Model group CTDIa
Model group
CT-Expo CTDI
NEXT Model group
CTDI
FX/i, LX/i ZX/i, NX/i
DX/i, FX/i, LX/i, ZX/i NX/i, -Pro
15.2 15.2
FX/i, LX/i DX/i
20.9 13.2
18.9 14.7
Breakdown Model
ImPACT
FX/i LX/i ZX/i NX/i DX/i -Pro
18.9 18.9 14.7 14.7
Final database
CTDI CT-Expo NEXT
Average
Model group
CTDI
15.2 15.2 15.2 15.2 15.2 15.2
18.3 18.3 14.9 14.9 14.2 15.2
FX/i, LX/i ZX/i, NX/i DX/i -Pro
18.3 14.9 14.2 15.2
20.9 20.9 13.2
a Normalised CTDI only for head phantom and the tube potential of 120 kVp was included for illustrative w
purposes.
where x is tube potential and y is the normalised CTDIw , was fitted to the data by least squares. Based on the analysis, the missing CTDI were calculated using the available values and the tube potentials for the missing values. 2.4. Analysis of CTDI variability
To understand the variability of the normalised CTDIw values, we analysed the final database where the merging process and the calculation of missing values were completed. First, we analysed the variability among scanner models in a single manufacturer. We picked the normalised CTDIw values for head and body CTDI phantoms for 120 kVp and calculated the mean, standard deviation, and coefficient of variation (COV) across the different scanner models for the four major scanner manufacturers: GE, Philips, Siemens, and Toshiba. Manufacturers such as Picker, Elscint, Shimadzu, and Hitachi, which contributed less than ten CT scanner models to the final merged data matrix, were excluded from the analysis. Second, we analysed the variability across the four different data sources. Scanner models with the normalised CTDIw values for head and body phantoms from more than two sources were included in the analysis to allow the investigation of the inter-source variation. Like the first analysis, only four major manufacturers, GE, Philips, Siemens, and Toshiba, were included in the analysis. The mean, standard deviation, and COV of the normalised CTDIw values for the head and body CTDI phantoms for 120 kVp were calculated and analysed. 3. Results and discussion 3.1. Data collection
The normalised CTDIw values were collected or calculated from the four data sources: ImPACT dose survey, CT-Expo program, and NEXT and NLST surveys. The number of CT scanner groups abstracted from the four data sources are summarised in table 2. Through the merging process described in table 1, a total of 249 original categories were regrouped into 162 categories. CT-Expo program contributed the greatest number of CT scanner groups (n = 138) compared to the NLST survey where 13 scanner groups were available. The NEXT and NLST 367
J. Radiol. Prot. 34 (2014) 363
E Lee et al
Figure 1. Examples of the exponential regression analysis for (A) GE LightSpeed VCT
and (B) GE Discovery CT HD750 scanners.
surveys only provided head and body CTDI values, respectively, whereas the ImPACT dose survey and CT-Expo program provided both head and body CTDI values. 3.2. Missing values
To calculate missing CTDI values, an exponential regression model (as in equation (3)) was used, applied to the CT scanner groups where the CTDI values are available for all tube potentials considered in this study, namely 80, 100, 120, and 140 kVp. The calculation was only performed for manufacturers apart from Picker and Hitachi that did not contain any missing values. Figure 1 shows an example of the exponential regression analysis for (a) GE LightSpeed VCT and (b) GE Discovery CT HD750 for head/body CTDI phantoms and the tube potential of 80–140 kVp. Although the two scanners are from the same manufacturer, the parameter A 368
J. Radiol. Prot. 34 (2014) 363
E Lee et al
Table 2. The number of CT scanner group for eight different manufacturers collected
from the four data sources. The number of CT scanner models after data merging is included in the final column. Manufacturer GE Philips Siemens Toshiba Elscint Picker Shimadzu Hitachi Total
ImPACT 17 13 16 16 2 3 1 68
CT-Expo
NEXT
NLST
27 17 35 34 7 7 6 5 138
10 2 3 8 5 2
5 3 3 2
30
13
Merged 27 25 38 42 10 9 6 5 162
Table 3. Average and standard deviation of the parameter B of the exponential regression model for CTDIw of head and body phantoms and for different manufacturers.
Manufacturer
GE Average SD
Philips Average SD
Siemens Average SD
CTDIw Head CTDIw Body
0.0197 0.0212
0.0216 0.0226
0.0219 0.0230
Manufacturer
Toshiba Average SD
Elscint Average SD
Picker Average SD
CTDIw Head CTDIw Body
0.0206 0.0218
0.0199 0.0216
0.0185 0.0196
0.0013 0.0013
0.0009 0.0007
0.0018 0.0016
0.0003 0.0011
0.0022 0.0015
0.0015 0.0021
shown in equation (3) of the exponential regression curve is significantly different: 1.8 and 1.3 for head phantom in LightSpeed VCT and Discovery CT HD750, respectively. However, the parameter B shown in equation (3) for the head phantom is identical, 0.0217, and the values for body phantom are very close, 0.0225 and 0.0230. Based on the observation, the average and standard deviation of the parameter B for head and body CTDI values were calculated across the different scanner models in six manufacturers to obtain representative values of the parameter B. Since at least one CTDI value is available (mostly for 120 kVp), the representative parameter B was then used to derive other CTDI values by using equation (3) and the available CTDI values for 120 kVp. The mean and standard deviation of the parameter B for head and body CTDI values were calculated across the different scanner models in six manufacturers to obtain the representative parameters B (table 3). To investigate if the parameters B differ by manufacturer, or by phantom body part (head versus body), we fitted linear models to this data on fitted exponential parameters, in which; Bi = α1 Manufactureri + α2 CDTI phantomi + εi .
(4)
Analysis of variance was assessed in the standard way using the F-test (Rao 2002). Table 4 demonstrates that all possible comparisons (of manufacturer within each CTDI phantom and across CTDI phantom types, and CTDI phantoms adjusted for manufacturer) were highly statistically significant ( p < 0.0001). This was also the case when the slope parameters B were log-transformed (results not shown). Based on the results, we decided to use a different parameter B for each manufacturer and CTDI phantom type. 369
J. Radiol. Prot. 34 (2014) 363
E Lee et al
Table 4. Analysis of variance decomposition of fitted slope parameters B from model
described in equation (4). Data (phantom body part)
Test for difference by category
Head only Body only Head and body Head and body Head and body
Manufacturer Manufacturer Manufacturer Manufacturer adjusted for phantom body part Phantom body part adjusted for manufacturer
p-value (via F-test)
