Size-specific Dose Estimates for Chest, Abdominal, and Pelvic CT: Effect of Intrapatient Variability in Water-equivalent Diameter

Although water-equivalent diameter (D_w) can vary substantially along the z-axis within any given patient, using the mean volume CT dose index over the whole scan range and the D_w from the image located at the center of the scan range provides an accurate and easily obtained estimation of the size-specific dose estimate for the whole scan range, with a root mean square error of less than 1.4 mGy or 9% for a scan range of the complete chest, abdomen, and pelvis.

Abstract

Purpose

To develop software to automatically calculate size-specific dose estimates (SSDEs) and to assess the effect of variations in water-equivalent diameter (D_w) along the z-axis on SSDE for computed tomographic (CT) examinations of the torso.

Materials and Methods

In this institutional review board–approved, HIPAA-compliant, retrospective study, a software program was used to calculate D_w at each image position in 102 consecutive CT examinations of the combined chest, abdomen, and pelvis. SSDE was calculated by multiplying the size-dependent conversion factor and volume CT dose index (CTDI_vol) at each image position. The variations in D_w along the z-axis were determined for six hypothetical scanning ranges: chest alone; abdomen alone; pelvis alone; chest and abdomen; abdomen and pelvis; and chest, abdomen, and pelvis. Mean SSDE was calculated in two ways: (a) from the SSDE at each position and (b) from the mean CTDI_vol over each scan range and the conversion factor corresponding to D_w at the middle of the scan range. Linear regression analysis was performed to determine the correlation between SSDE values calculated in these two ways.

Results

Across patients, for scan ranges 1–6, the mean of the difference between maximal and minimal D_w within a given patient was 5.2, 4.9, 2.5, 6.0, 5.6, and 6.5 cm, respectively. The mean SSDE values calculated by using the two methods were in close agreement, with root mean square differences of 0.9, 0.5, 0.5, 1.4, 1.0, and 1.1 mGy or 6%, 3%, 2%, 9%, 4%, and 6%, for the scan ranges of chest; abdomen; pelvis; chest and abdomen; abdomen and pelvis; and chest, abdomen, and pelvis, respectively.

Conclusion

Using the mean CTDI_vol from the whole scan range and D_w from the image at the center of the scan range provided an easily obtained estimate of SSDE for the whole scan range that agreed well with values from an image-by-image approach, with a root mean square difference less than 1.4 mGy (9%).

^© RSNA, 2015

Online supplemental material is available for this article.

Introduction

Computed tomographic (CT) dose index and its derivatives, such as volume CT dose index (CTDI_vol), are used to quantify radiation output from CT scanners and are widely used in CT quality control programs (1,2). However, CTDI_vol does not represent patient dose, in part because the measurement is performed on a polymethyl methacrylate cylindrical phantom of a specified size (either 16 cm or 32 cm in diameter) (3). Actual patient dose depends on patient size and attenuation, as well as scanner output (CTDI_vol); patients with different body shapes and sizes can absorb very different doses for the same CTDI_vol. Other studies demonstrated that scanner-independent organ dose estimates can be obtained by using CTDI_vol as a normalization factor and then correcting for patient size (4,5). On the basis of data from these studies and from three other groups, the American Association of Physicists in Medicine (AAPM) Report 204 (6) introduced the concept of a size-specific dose estimate (SSDE), which is calculated as the product of a size-dependent conversion factor and CTDI_vol. This provides a simple estimate of mean patient dose from a CT scan at the center of the scan range, which takes into account patient size and attenuation and can be computed from a readily available measure of scanner output, CTDI_vol. The method has been evaluated for adult and pediatric CT scans of the torso (7,8), and Radiology recommends that authors report SSDE whenever possible, together with CTDI_vol and dose-length product (9,10).

Several descriptors of patient size were proposed in the AAPM Report 204, including anterior-posterior dimension, lateral dimension, anterior-posterior and lateral dimensions, circumference, and effective diameter (6). These descriptors, however, are based only on patient geometry and do not take into account the different attenuations of various substances, such as bone, tissue, and air. The water-equivalent area and water-equivalent diameter (D_w), which represent the total attenuation of the patient normalized to that of water, have been proposed to account for these differences in attenuation. Wang et al (11,12) demonstrated that the use of D_w, which takes into account tissue attenuation in addition to patient geometric size, is more accurate in calculating SSDE in thoracic CT; taking attenuation into account is less critical for the abdomen and pelvis. AAPM Report 220 addresses this topic in detail, and the sole use of D_w for calculations of SSDE is recommended (13–16).

During a CT scan of a torso, patient size and attenuation can change considerably along the z-axis (cranial-caudal direction). Since automatic exposure control is widely used in most torso CT scans, tube current and consequently CTDI_vol also change with patient size. Therefore, both components of SSDE, the conversion factor and CTDI_vol, change with varying size and attenuation along the z-axis within any given patient. The purpose of this study was therefore to develop software to automatically calculate SSDEs and to assess the effect of variations in D_w along the z-axis on SSDE for CT examinations of the torso.

Materials and Methods

Patient Enrollment and CT Scanning

This Health Insurance Portability and Accountability Act–compliant, retrospective study was approved by our institutional review board, with waiver of informed consent. Only data from adult patients were included. Data from consecutive CT examinations of the combined chest, abdomen, and pelvis performed with four 128-section CT scanners were collected. For each patient, the chest, abdominal, and pelvic examination was conducted in one acquisition for the whole chest, abdominal, and pelvic areas. All scanners were from the same manufacturer (Definition AS+ and Definition Flash; Siemens Healthcare, Forchheim, Germany), which are essentially the same when operated in single-source mode, as was done in this study. Scans were performed by using our routine chest, abdominal, and pelvic protocol, in which the scanner was operated in the single-source and single-energy scan mode. The tube potential was automatically selected on the basis of patient size and imaging task (CareKV; Siemens Healthcare) (17), with a reference tube potential of 120 kV. Automatic exposure control (CareDose 4D; Siemens Healthcare) was turned on, and a quality reference effective tube current–time product of 240 mAs was used. Other key parameters included a 0.5-second rotation time, 128 × 0.6-mm collimation, and helical pitch of 0.8. For a standard-sized adult (with attenuation equivalent to that of a 33-cm–diameter water phantom, approximately 80 kg), this would correspond to a CTDI_vol of approximately 16 mGy (18,19). Images were reconstructed by using a medium-sharp kernel (B40) with 5-mm image thickness and 5-mm increment. The reconstruction field of view was large enough to ensure that all anatomy was included—that is, there was no truncation of anatomy.

Six scan ranges were retrospectively identified within each chest, abdominal, and pelvic scan: (a) chest alone, (b) abdomen alone, (c) pelvis alone, (d) chest and abdomen, (e) abdomen and pelvis, and (f) entire chest, abdomen, and pelvis. For each patient, the beginning and end of each of the scan ranges were determined on the basis of anatomic markers; chest scan ranges were from the top to the bottom of the lung, abdominal scan ranges were from the top of the liver to the top of the pelvic crest, and pelvic scan ranges were from the top of the pelvic crest to the pubic symphysis. Combined scans (chest and abdomen; abdomen and pelvis; and chest, abdomen, and pelvis) were formed by addition of the individual scan ranges.

Software to Calculate D_w and SSDE by Using Reconstructed CT Images

We developed software by using Matlab (Mathworks, Natick, Mass) to automatically calculate D_w and SSDE from reconstructed CT images. For a given CT image, the software first used morphology operations to exclude the patient table. Threshold-based segmentation was then performed with a threshold of −900 HU. This ensured inclusion of all patient anatomy but excluded potential artifacts with relatively low CT numbers, such as streaks in the air outside of the body contour. The water-equivalent attenuation (A_w) and D_w of an object were calculated by using the following equations (12):

and

where CT(x,y) is the CT number (in Hounsfield units) at the image location of (x,y) and A_pixel is the area of a pixel, and the summation was performed over the segmented region of each image (ie, for all pixels with a CT number higher than −900 HU).

For each image, the effective tube current–time product was obtained from the Digital Imaging and Communications in Medicine header. A CTDI_vol that corresponded to each image location was calculated by multiplying the effective tube current–time product by the scanner output normalized to the tube-current product (CTDI_vol/tube current–time product), which was obtained from direct measurement by using standard methods (20,21). For CT scans of the torso, as in this study, CTDI_vol was reported by using the 32-cm–diameter CT dose index phantom. A size-dependent conversion factor to convert CTDI_vol to SSDE was obtained by using the following equations (6):

and

where represents the SSDE conversion factor as a function of patient size (D_w).

Mean SSDE for Each Scan Range

Image-by-image approach.—With D_w and SSDE calculated at each image location, the mean SSDE over the whole scan range was calculated as

where z represents the location of each of the N images inside the scan range. This method required image data and CTDI_vol at all locations in the scan range, which might not be easily obtained in certain scenarios and would require software to calculate these values in clinical practice.

D_w at the center of the scan range and mean CTDI_vol over the scan range approach.—An approximation method to calculate the SSDE over the scan range is proposed in this work. A single estimate of patient size and attenuation at the middle of the scan range (D_w,mid) and the mean CTDI_vol over the scan range were used to calculate mean SSDE. This was motivated by the absence of software tools at the scanner by which to perform the image-by-image calculations. This latter approach (use of D_w from the middle of the scan range and the mean CTDI_vol over the scan range) could be manually performed with minimal effort:

where z_mid represents the location of the central image in the scan range and is the mean CTDI_vol over the whole scan range, which is reported by the scanner.

Statistical Analysis

D_w and SSDE were calculated at each image location along the z-axis for scan ranges 1–6 by using Equations (1)–(4). For each patient and scan range, the mean D_w value, standard deviation of D_w values, D_w coefficient of variation, and range of D_w values along the z-axis, as well as the differences between the maximum and minimum D_w values, were calculated. The mean value, standard deviation, and range values of these parameters were then calculated across all patients for each scan range.

In this study, both and SSDE based on D_w in the mdidle of the scan range and mean CTDI_vol over the scan range (SSDE_mid) were calculated for each of the cases. Linear regression analysis was performed to determine the correlation between and SSDE_mid, and the root mean square difference between and SSDE_mid was calculated for each of the six scan ranges. The least-squares linear fit was calculated with the fitting equation . Additionally, we performed cross-validation for the chest, abdominal, and pelvic scan range. This was conducted by fitting the model on the first 51 patients and then using the model to predict the values for the remaining 51 patients. The predicted values were then compared with the original values, and the root mean square difference was calculated.

Results

A total of 102 patients (59 men and 43 women) were included in this study, with a mean age of 59 years (range, 18–88 years). The mean age of women was 59 years (range, 18–88 years), and the mean age of men was 59 years (range, 24–86 years). There was no statistical difference in age between men and women (P = .92 with a nonpaired t test). These patients represented a wide spectrum of body habitus, with a mean weight of 84.8 kg (range, 36.9–183.1 kg) and a mean body mass index of 28.5 kg/m² (range, 14.7–56.6 kg/m²).

As seen on the CT localization radiograph of a sample chest, abdominal, and pelvic scan and the corresponding D_w versus z-axis position plot (Fig 1), patient size can change substantially along the z-axis. At the top of the lung, a large D_w value was observed because of the attenuation of the shoulder. Farther down in the lung area, D_w decreased because patient size decreased and the attenuation of lung is much lower than that of soft tissue. D_w reached a minimal value above the top of the liver and increased again as the scan went through the pelvis. The tube current modulation prescribed by the automatic exposure control system produced a highly variable z-specific value of CTDI_vol (Fig 1), which can be observed from the CTDI_vol plot, which in general followed the trend of patient attenuation (D_w). Higher CTDI_vol values were observed at z locations with larger D_w values because tube current was boosted by the automatic exposure control system. Conversely, lower CTDI_vol values were observed when D_w values were smaller. SSDE was higher than CTDI_vol in this patient because the patient size and attenuation were smaller than those of the 32-cm–diameter phantom. Hence, the conversion factor used in the SSDE calculation was larger than 1.0 (6).

Figure 1:

In a sample case, A, a CT localizer radiograph and B–E, corresponding D_w versus z-axis position plots demonstrate the effect of patient size variation on B, D_w, C, conversion factor (f_Dw), D, CTDI_vol, and E, SSDE.

Considerable variation in D_w was observed inside the scan range (Table 1); the mean difference between maximal and minimal D_w ranged from 2.5 cm (or 8%, in the pelvis) to 6.5 cm (or 20%, for chest, abdominal, and pelvic scans). The largest difference in D_w observed among all scan ranges was 10.8 cm (in a chest, abdominal, and pelvic scan), which represented a 32% variation in D_w across the scan range (as a function of each patient’s mean D_w value).

Table 1.

D_w and Variations in D_w within a Given Patient and Scan Range, Averaged over All Patients

Note.—Values are means ± standard deviations, with ranges in parentheses.

SSDE_mid and were almost perfectly correlated (R² ≥ 0.9713) in all six scan ranges (Fig 2, Table 2). The root mean square difference was 0.9, 0.5, 0.5, 1.4, 1.0, and 1.1 mGy or 6%, 3%, 2%, 9%, 4%, and 6% for the scan ranges of chest; abdomen; pelvis; chest and abdomen; abdomen and pelvis; and chest, abdomen, and pelvis, respectively. For the cross-validation study, SSDE_mid was modeled as by using the first 51 patients. By using this model, SSDE_mid values were predicted for the remainding 51 patients and compared with the original SSDE_mid directly calculated by using the middle image. The root mean square difference was 0.7 mGy or 4%.

Figure 2:

A–F, Scatterplots are shown for SSDE_mid relative to for A, chest; B, abdominal; C, pelvic; D, chest and abdominal; E, abdominal and pelvic; and F, chest, abdominal, and pelvic CT scans. The least squares linear fit was calculated with the fitting equation . The correlation coefficient (R²) as determined from the linear regression is also shown.

Table 2.

Values of a, b, and R² for Linear Fit

Note.—Values were obtained by using the equation for indicated scans.

Discussion

Patient size varies along the z-axis of a CT scan of the torso because of changes in body shape and attenuation at different areas (eg, shoulder, lung, abdomen, and pelvis). We observed considerable variation of D_w in each of the six scan ranges. This size variation is expected to affect SSDE values because it changes both factors in the SSDE calculation, the size-dependent conversion factor and CTDI_vol, which vary because of the use of automatic exposure control in most body CT scans. Our results supported this expectation.

For most clinical users, a single number for SSDE is likely preferred, as it might represent the mean dose to the patient in the center of the scan range. Because SSDE was defined as the mean dose to the central image of the scan range in a uniform phantom, the mean SSDE may be a reasonably desirable dose metric for clinical users. The mean SSDE, however, needs to take into account the variation of patient size (D_w) along the z-axis. Therefore, we performed intermediate calculations at each image location—CTDI_vol(z) and SSDE(z)—even though we recognize that for image locations not in the center of scan range and for scans where tube current varies at locations, this is an approximation to the original definition of CTDI_vol and SSDE. The mean SSDE was calculated by averaging SSDE at each image location over the whole scan range. However, it is not practical to do this manually. In this study, we demonstrated that a simpler method can be used to calculate the mean SSDE in the scan range, in which the central image is used to calculate D_w,mid, and the conversion factor from CTDI_vol to SSDE is chosen on the basis of D_w,mid. The mean SSDE is obtained by multiplying the mean CTDI_vol over the scan range by this conversion factor. This simpler method can be readily performed at the scanner console by using region of interest measurement tools provided by the scanner to measure the patient attenuation at the center of the scan range. The operator performing this calculation would, however, require a calculator and a copy of the data table from AAPM Report 204. Detailed instructions are included in Appendix E1 (online) to guide the operator through the needed steps by using an example case of a combined chest, abdominal, and pelvic CT scan. Because only one image is needed, the mean SSDE could be calculated manually with minimal effort.

In a previous study, Cheng proposed a method for automated estimation of patient effective diameter in abdominal CT (22). It was found that effective diameter in the middle section of an abdominal CT examination was close to the mean effective diameter, and the difference between the SSDE calculated by using the effective diameter of the middle image in the scan range and the mean of the per-image SSDE was relatively small (22). These findings are similar to our results for the abdominal scan range. One limitation of using effective diameter instead of D_w to quantify patient size is that effective diameter does not take into account the different attenuation of various substances, such as bone, tissue, and air. For body regions that include large portions of air, such as lung, the effective diameter leads to overestimation of patient size and therefore underestimation of SSDE. In our study, we used D_w to quantify patient attenuation, which was calculated on the basis of the CT number of each pixel to take into account the attenuation of each voxel. (Although the term “effective diameter” was used in AAPM Report 204, the size-specific conversion factors were based on the effective diameter of water-equivalent materials, which is the same as calculating D_w).

In this study, we further investigated patient size variation and its effect on SSDE for six different scan ranges. Our data indicate that scans involving the chest area are associated with greater variation of patient size along the z-axis than those only involving abdomen and/or pelvis regions. Regardless, in all body regions, the SSDE calculated by using D_w from the central image of the scan range and mean CTDI_vol of the whole scan range were highly correlated with the SSDE calculated on an image-by-image basis across the scan region.

There are limitations to this study. First, it is a retrospective study. However, since the patients were included consecutively, there should be no difference in D_w and SSDE values between prospective and retrospective studies. Also, our patient cohort represented a wide spectrum of patient sizes. Thus, we believe our results are generalizable to any adult population. Second, all patients included in this study were adults. We chose to do this because greater size variation along the z-axis was expected in adult patients compared with pediatric patients (prior to adolescence, body habitus is more cylindrical than it is postadolescence). However, further study with pediatric patients is warranted to obtain a definite conclusion on the effect of pediatric patient size variations on SSDE calculations, particularly because the SSDE conversion factor as a function of patient size (D_w) changes more rapidly at low D_w values (ie, children) than at large D_w values (ie, adults).

In conclusion, we have developed software that can automatically calculate D_w and SSDE at each image location within a patient. D_w was found to vary considerably over the scan range. Although D_w can vary substantially along the z-axis within any given patient, using the mean CTDI_vol over the whole scan range and the D_w from the image located at the center of the scan range provides an accurate and easily obtained estimation of the SSDE for the whole scan range, with a root mean square error of less than 1.4 mGy or 9% for the scan range within the chest, abdomen, and pelvis. Therefore, we conclude that in the absence of automated tools to calculate the mean SSDE on an image-by-image basis, use of one image in the center of the scan range and the mean CTDI_vol from the entire scan provided a sufficiently accurate method for calculating the mean SSDE for CT examinations of the torso in adults.

Advances in Knowledge

• ■ Automatic calculation of water-equivalent diameter (D_w) and size-specific dose estimates (SSDEs) can be performed rapidly by using simple software tools.

• ■ The maximal difference in D_w values along the z-axis for any given patient varied between 0.5 and 10.8 cm in an adult patient cohort.

• ■ The SSDE for a torso CT scan was calculated by using two methods: (a) using the mean volume CT dose equivalent (CTDI_vol) over the whole scan range and the D_w from the central image and (b) using an automated image-by-image calculation; the two methods agreed well with each other, with a root mean square difference of 1.4 mGy.

Implication for Patient Care

• ■ SSDE can be calculated easily by using the central image of a CT scan of the torso, providing a method for reporting patient-specific dose that takes into account both scanner output (CTDI_vol) and patient size and attenuation (D_w).

APPENDIX

Appendix E1

Supplementary PDF provided by authors