| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Cardiac Imaging |
1 From the Ctr for Health Studies, Group Health Cooperative, Metropolitan Park East, Suite 1600, 1730 Minor Ave, Seattle, WA 98101 (J.C.N.); Dept of Biostatistics, Univ of Washington, Seattle, Wash (J.C.N., R.A.K.); Dept of Radiology, Wake Forest Univ School of Medicine, Winston-Salem, NC (J.J.C.); Dept of Radiology, David Geffen School of Medicine at UCLA, Los Angeles, Calif (M.F.M.G., J.G.G.); Heart Disease Prevention Program, Div of Cardiology, College of Medicine, Univ of California, Irvine, Calif (N.D.W.); Div of Epidemiology and Clinical Applications, National Heart, Lung, and Blood Institute, Bethesda, Md (C.M.L.); Dept of Medicine, Div of Preventive Medicine, Univ of Alabama at Birmingham, Birmingham, Ala (O.D.W.); and Harbor-UCLA Research and Education Inst, Div of Cardiology, Los Angeles, Calif (R.D.). Received Mar 18, 2004; revision requested May 27; revision received Jun 25; accepted Jul 27. Supported by NHLBI contracts N01-HC-95159 through N01-HC-95165, N01-HC-95169, N01-HC-95095, N01-HC-48047 through N01-HC-48050, and N01-HC-05187. Address correspondence to J.C.N. (e-mail: nelson.jl@ghc.org).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSION APPENDIX REFERENCES |
|---|
MATERIALS AND METHODS: Institutional review board approval and participant informed consent were obtained at all study sites. An image attenuation adjustment method involving the use of available calibration phantom data to define standard attenuation values was developed. The method was applied to images from two population-based multicenter studies: the Coronary Artery Risk Development in Young Adults study (3041 participants) and the Multi-Ethnic Study of Atherosclerosis (6814 participants). To quantify the variability in attenuation, analysis of variance techniques were used to compare the CT numbers of standardized torso phantom regions across study sites, and multivariate linear regression models of participant-specific calibration phantom attenuation values that included participant age, race, sex, body mass index (BMI), smoking status, and site as covariates were developed. To assess the effect of the calibration method on calcium measurements, Pearson correlation coefficients between unadjusted and attenuation-adjusted calcium measurements were computed. Multivariate models were used to examine the effect of sex, race, BMI, smoking status, unadjusted score, and site on Agatston score adjustments.
RESULTS: Mean attenuation values (CT numbers) of a standard calibration phantom scanned beneath participants varied significantly according to scanner and participant BMI (P < .001 for both). Values were lowest for Siemens multi–detector row CT scanners (110.0 HU), followed by GE-Imatron electron-beam (116.0 HU) and GE LightSpeed multi–detector row scanners (121.5 HU). Values were also lower for morbidly obese (BMI, 
40.0 kg/m2) participants (108.9 HU), followed by obese (BMI, 30.0–39.9 kg/m2) (114.8 HU), overweight (BMI, 25.0–29.9 kg/m2) (118.5 HU), and normal-weight or underweight (BMI, <25.0 kg/m2) (120.1 HU) participants. Agatston score calibration adjustments ranged from –650 to 1071 (mean, –8 ± 50 [standard deviation]) and increased with Agatston score (P < .001). The direction and magnitude of adjustment varied significantly according to scanner and BMI (P < .001 for both) and were consistent with phantom attenuation results in that calibration resulted in score decreases for images with higher phantom attenuation values.
CONCLUSION: Image attenuation values vary by scanner and participant body size, producing calcium score differences that are not due to true calcium burden disparities. Use of calibration phantoms to adjust attenuation values and calibrate calcium measurements in research studies and clinical practice may improve the comparability of such measurements between persons scanned with different scanners and within persons over time.
© RSNA, 2005
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSION APPENDIX REFERENCES |
|---|
Measurement error in CAC measurement is an important issue, and variability in the attenuation of x-rays as they pass through the body is one factor that may cause such error. In this article, we define image attenuation as the CT numbers (measured in Hounsfield units) associated with a standard calibration phantom scanned beneath study participants. Many scanner and participant factors (eg, the anatomy of interest, x-ray beam energy, beam hardening) interact in complex ways to determine attenuation. Thus, image attenuation may differ between CT scanners, scanning protocols, and participants and over time. In turn, CAC measurements can be highly affected by image attenuation because they depend on the detection of aggregates of contiguous image pixels with attenuation values greater than 130 HU and because some CAC measures (eg, the Agatston score) depend on preset attenuation thresholds that determine the multiplier weights to be applied (14). Hence, variability in image attenuation increases measurement error and reduces the accuracy and precision of CAC quantification.
Calibrating images to a phantom containing known amounts of calcium hydroxyapatite to reduce image attenuation variability is essential for accurate measurement of vertebral bone mineral density with CT (so-called quantitative CT) (15). Stanford et al (16) recently concluded that CAC measurements obtained with CT are also significantly affected by attenuation differences and recommended the use of a phantom to calibrate calcium measurements, especially in longitudinal and multicenter studies. In addition, McCollough et al (17) determined that calibrating with a phantom when measuring CAC with CT can reduce scanner variability in image attenuation even when systems are located at the same site and maintained by the same personnel. Despite these data, calibration methods have not been widely implemented for the measurement of CAC with CT in clinical or research studies to date. Rapidly changing scanning technologies and the increasing use of CAC CT scanning in multicenter longitudinal research studies and in clinical practice to monitor subclinical disease progression in patients over time demand that these issues be addressed. Thus, we conducted our study to quantify scanner and participant variability in attenuation values for CT images assessed for coronary calcium and to define a method for standardizing attenuation values and calibrating calcium measurements.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSION APPENDIX REFERENCES |
|---|
Phantoms
A calibration phantom and torso phantom (both from Image Analysis, Columbia, Ky) developed for measurement of bone mineral density with quantitative CT in the evaluation of osteoporosis were used in this research (Fig 1). One pair of phantoms (calibration and torso) was assigned to each site. The calibration phantom contained four rods of known calcium hydroxyapatite concentrations (0, 50, 100, and 200 mg/mL), was positioned beneath each participant so that the length of the phantom covered the expected length of the heart, and was scanned in every section (Fig 2). The torso phantom contained a rod with 100 mg/mL of hydroxyapatite and was scanned (with the calibration phantom underneath it) for quality-control purposes every 2–4 weeks. The CT technologist recorded the mean attenuation value (CT number in Hounsfield units) in the 15-mm-diameter regions of interest placed over the four calibration phantom rods and the torso phantom rod.
|
|
1. Selection of a representative image sample to define standard attenuation levels to which other images are calibrated; the standards equal the average CT numbers (in Hounsfield units) corresponding to the known physical concentrations of calcium in the four calibration phantom rods in this representative sample.
2. Measurement of the CT numbers for the four calibration phantom rods scanned in each image to be calibrated.
3. Calculation of a calibration formula for each image that is based on the linear relationship between the observed CT numbers in the image and the standardized CT numbers.
4. For each participant, definition of an average calibration line through computation of the median slope and intercept among all of their image-specific calibration lines.
5. Use of this participant-specific average calibration line to convert the original attenuation value for each pixel in each image to the standard attenuation scale by using the following equation: NPCTno = (OPCTno – a)/b, where NPCTno is the new CT number for the pixel, OPCTno is the original CT number for the pixel, a is the intercept, and b is the slope.
Algorithm development and rationale.—To develop this calibration algorithm, we needed to study the behavior of calibration phantom data. Hence, two biostatisticians (J.C.N., R.A.K.) explored available image-specific calibration data from the South Bay Heart Watch (SBHW) study of 1312 men and women from an asymptomatic population of adults 45 years of age or older (1044 CT image sections obtained in 27 subjects were available). The principal investigator of the SBHW study gave permission for use of the study data. Details of the CT system and scanning protocol used for the SBHW study are included in the Appendix. We used these data to formulate a statistical model for the calibration phantom data by (a) evaluating the shape of the association between image attenuation values (CT numbers) and the known calcium concentrations in the calibration phantom, (b) comparing image-specific calibration lines to participant-specific calibration lines averaged across images from the same participant, and (c) determining the nature and frequency of calibration data errors to assess whether or not such errors indicated a need for robust regression techniques.
In the SBHW data, (a) a highly linear association between calibration phantom image attenuation and known calcium concentrations was seen for all images, (b) image-specific and participant-specific calibration lines were virtually indistinguishable, and (c) calibration phantom data errors (eg, those due to image noise) that adversely influenced image-specific calibration lines for affected images occurred very occasionally. Given these findings, we chose a linear, participant-specific calibration model that was defined by obtaining a robust average (ie, the median) across image-specific lines to minimize the impact of potential errors. This statistical model and the adjustment algorithm are defined explicitly in the Appendix.
Implementation.—We also used the calibration data from the SBHW sample to estimate standard attenuation values (Appendix). The SBHW examinations were performed in a busy research and clinical scanning center and were chosen both for convenience and to help ensure that our standard image attenuation values were comparable to those of images obtained in current clinical practice and research. The SBHW estimates were similar to those computed for CARDIA and MESA after method development and data collection were completed. Our calibration method may be applied, however, by using any desired sample to define standard attenuation values.
We calibrated CT scans acquired in 3041 participants in the CARDIA study at the year-15 examination (performed between June 2000 and July 2001) and CT scans acquired in 6814 participants in the MESA at the baseline examination (performed between July 2000 and August 2002). The CARDIA study and MESA are two National Heart, Lung, and Blood Institute–initiated multisite population-based cohort studies; the CARDIA study was designed to evaluate early predictors of subclinical cardiovascular disease in subjects aged 33–45 years, while the MESA was designed to evaluate the progression of subclinical to clinical cardiovascular disease in subjects aged 45–85 years. Both study cohorts consist of U.S. men and women from multiple ethnic groups. Details on the design, sampling, recruitment methods, and goals of these studies are described elsewhere (19,20). We applied the calibration methods to a combined CARDIA and MESA data set.
Reading Procedures
Images were assessed for quality and read centrally at the Harbor-UCLA Research and Education Institute by using an interactive computer scoring system similar to that described by Yaghoubi et al (21). Readers were cardiologists or radiologists with experience in reading images of the coronary anatomy and were trained and supervised by R.D., who had 10 years of experience in reading coronary calcium CT scans. The reading software (a custom software developed specifically for these studies and based on software used during the year-10 examination [in 1994–1995] of the CARDIA study) computed the mean CT number within a 15-mm-diameter circular region of interest located over each of the four rods of the calibration phantom. Calcium was considered to be present when a minimum of four adjacent pixels (2.8 mm2) each had a CT number greater than the Agatston score threshold of 130 HU. The Agatston score (22), volume (22), and interpolated volume score (23) were computed.
The images were then adjusted by using the calibration method, and the reader performed a second analysis after image attenuation adjustment. Thus, two sets of calcium measurements were computed for each image: unadjusted and attenuation adjusted. Ten percent of the two scans obtained per study participant were selected at random by statisticians at the CARDIA and MESA coordinating centers and were reread either by a second reader or a second time by the same reader for quality control.
Statistical Analyses
Variability in image attenuation (torso and calibration phantom data).—Analysis of variance techniques were used to compare the mean measured CT numbers of the standardized torso phantom regions across sites, and means were plotted with 95% confidence intervals (CIs) according to site and scanner type. The mean attenuation (CT number) of each calibration phantom was computed across available scans (usually two) for each participant. For consistency with the torso data, calibration phantom results are presented for the 100-mg/mL rod. Results for other rods were similar.
Multivariate linear regression methods were used to model the calibration phantom attenuation values. Model covariates were hypothesis-driven on the basis of previous findings in the literature and included the following: participant age, race, sex, body mass index (BMI) in kilograms per meters squared, smoking status (current, former, or never), and scanning site or type. BMI was categorized by using the National Heart, Lung, and Blood Institute and World Health Organization criteria (24) as indicating normal weight or underweight (<25.0 kg/m2), overweight (25.0–29.9 kg/m2), obesity (30.0–39.9 kg/m2), or morbid obesity (40.0 kg/m2). Covariate-adjusted mean attenuation values with 95% CIs were computed for covariate groups, and F tests were used to assess significance.
Calibration of CAC measurements.—Mean unadjusted and attenuation-adjusted CAC measurements were computed across available scans for each participant. For each participant, we determined the unadjusted and attenuation-adjusted calcium presence (defined as a nonzero mean unadjusted Agatston score and a nonzero mean attenuation-adjusted score, respectively) and the mean Agatston score calibration adjustment (defined as the mean attenuation-adjusted score minus the mean unadjusted score). Summary statistics, including the mean, quantiles, and interquartile ranges, across participants were also computed for the unadjusted score, attenuation-adjusted score, and score adjustment.
Pearson correlation coefficients were computed to examine the association between the unadjusted and the attenuation-adjusted CAC measurements. Observed agreement between measurements of calcium presence defined by using Agatston scores before attenuation adjustment and measurements of calcium presence defined by using Agatston scores after attenuation adjustment was also assessed. Agatston scores were also categorized into clinically meaningful groups (0, 1–10, 11–100, 101–400, 401–1000, and >1000), cross-tabulated, and assessed for discordance between unadjusted and attenuation-adjusted measurements (eg, for the percentage of participants with a zero score before adjustment and a nonzero score after adjustment or vice versa). Mean unadjusted scores, attenuation-adjusted scores, and score adjustments were computed and compared by using analysis of variance methods according to site and scanner type, and mean score adjustments were plotted against mean unadjusted scores for each scanner type.
So that we could further explore the effect of scanner and participant factors on the magnitude of the Agatston score adjustment, multivariate models that included site or scanner type, as well as the participant’s sex, race, BMI, smoking status, and unadjusted score, were developed by using data for participants who had a nonzero score adjustment and nonmissing covariate data. Prespecified interactions between site or scanner type and other covariates were tested for significance by using likelihood ratio tests and retained if they were significant at a .05 
level. Covariate-adjusted mean Agatston score adjustments and 95% CIs were computed for categorical covariate groups, and F tests were performed to determine significance.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSION APPENDIX REFERENCES |
|---|
|
|
Calibration phantom attenuation values also varied significantly according to participant characteristics. The largest participant differences were between BMI groups (mean difference between extreme BMI groups, 11.2 HU; P < .001) and were relatively comparable to the size of the differences between scanners. Morbidly obese participants had the lowest mean attenuation values (108.9 HU; 95% CI: 108.4, 109.4), followed by obese participants (114.8 HU; 95% CI: 114.6, 115.0), overweight participants (118.5 HU; 95% CI: 118.3, 118.7), and normal-weight or underweight participants (120.1 HU; 95% CI: 119.9, 120.3). Mean attenuation was significantly (P < .001 for all) higher for women and Asian-Americans (who were primarily of Chinese descent) and lower for Hispanics, but these differences were clinically negligible (mean difference, <3 HU between extreme groups for each covariate).
Calibration of CAC Measurements
Overall adjustment results.—On the basis of the unadjusted Agatston score, 3654 (38.2%) of the 9553 study participants (340 [11.5%] of the 2962 CARDIA participants and 3314 [50.3%] of the 6591 MESA participants for whom calibration data were available) had some calcium present. The median and the interquartile range for unadjusted Agatston scores among all participants with calcium present were 76 and 195, respectively (median, 17 and interquartile range, 53 for CARDIA participants; median, 88 and interquartile range, 282 for MESA participants), with a maximum unadjusted Agatston score of 6450.
The correlations between unadjusted and attenuation-adjusted Agatston scores, volumes, and volume scores were high (r > 0.99, P < .001 for all CAC measurements). In total, for 84 (0.9%) of 9553 participants, the results regarding calcium presence before attenuation adjustment were discordant with those after attenuation adjustment—that is, these participants had Agatston scores of zero before adjustment and nonzero after adjustment or vice versa. Very few (n = 23 [0.4%]) of the 5899 participants with no calcium on the basis of the unadjusted score had a positive attenuation-adjusted calcium score, and the mean adjustment in this group was small, at 7 ± 13 (standard deviation). Nearly all (3303 [90.4%]) of the 3654 participants with positive unadjusted calcium scores had some adjustment in their scores owing to calibration (mean adjustment, 24 ± 45). The size of the Agatston score adjustment increased with increasing unadjusted score (Fig 4). Some adjustments, especially for participants with high unadjusted scores, were large (range, –650 to 1071).
|
|
|
Other Factors Affecting Attenuation Adjustment
Site, scanner type, BMI, and unadjusted Agatston score were the most significant multivariate predictors (P < .001) of Agatston score adjustments among participants with a nonzero change and nonmissing covariate data (n = 3312) (Table 4). On average, Siemens multi–detector row CT scanners (which yielded images that had the lowest phantom attenuation values) had positive score adjustments (covariate-adjusted mean score adjustment, +20), negative score adjustments were observed for GE-Imatron electron-beam CT scanners (covariate-adjusted mean score adjustment, –13), and GE LightSpeed multi–detector row CT scanners (which yielded images that had the highest phantom attenuation values) had larger negative adjustments (covariate-adjusted mean score adjustment, –26). Normal-weight and underweight participants (whose images had the highest phantom attenuation values) had the largest decreases in Agatston score (–16) owing to attenuation adjustment, and scores for the most obese participants (whose images had the lowest phantom attenuation values) increased (+6). Agatston score adjustments were also significantly larger for participants with higher unadjusted scores (P < .001), with average estimated adjustments of +2.5 (with Siemens multi–detector row CT scanners), –1.9 (with GE-Imatron electron-beam scanners), and –4.4 (with GE LightSpeed multi–detector row scanners) for every 50-point increase in unadjusted Agatston score (regression coefficients are not presented in Table 4). A small difference in Agatston score adjustment was detected between the sexes (P < .001), with scores for women (covariate-adjusted mean score adjustment, –9, 95% CI: –12, –6) decreasing slightly more than scores for men (covariate-adjusted mean score adjustment, –4, 95% CI: –7, –2). No significant differences were found according to race, smoking status, or age.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSION APPENDIX REFERENCES |
|---|
Image attenuation was also affected by the participant’s body size as measured with the BMI, with the lowest phantom attenuation values seen for obese participants and the highest seen for normal-weight or underweight participants (difference between groups, 11.2 HU). This inverse association between BMI and attenuation was also observed by Stanford et al (16) and Raggi et al (25) and may be due to "hardening" of the x-ray beam as it passes through tissues proximal to the anatomy of interest; this hardening can affect attenuation in complex ways. In the present study, significant but clinically negligible attenuation differences were also seen according to sex and race.
These scanner and participant attenuation differences are a matter of concern, because methods for detecting and measuring CAC inherently depend on image pixels exceeding preset attenuation thresholds. Thus, attenuation variability can cause differences in CAC measurements that are not due to true disparities in CAC burden between participants.
Calibration of CAC Measurements
Improved CAC measurements can be obtained by correcting for variability in image attenuation. We developed a method that attempts to equalize image attenuation values across scanners and participants by converting the attenuation of individual pixels in each image to standard attenuation values before CAC scoring. A standardized hydroxyapatite calibration phantom was scanned beneath the participants’ thoraces, and we used the obtained data to estimate a linear, participant-specific calibration line that modeled the association between the calibration phantom CT numbers for the participant’s images and standard CT numbers. The attenuation-adjusted images were then analyzed and scored to obtain calibrated CAC measurements. Equivalent calibrated scores could be obtained by adjusting the Agatston score thresholds rather than each image pixel. However, the latter method was more readily implemented into existing CAC scoring software, and when powerful computers are used, the time saved by using the former approach is insignificant.
Others have recognized that scanner and participant factors affect variability in image attenuation and have proposed the use of calibration phantoms to adjust for image attenuation differences (16,17). McCollough et al (17) used a calibration phantom to convert image attenuation (CT numbers) to physical calcium concentrations and achieved a reduction in attenuation variability. Our method furthers this approach in that we convert the observed CT numbers to standard CT numbers so that calibrated CAC measurements may be computed.
Our calibration method is feasible and simple to perform, as demonstrated by the successful implementation of this method for more than 9500 participants in the CARDIA and MESA trials. Quality calibration data were obtained for nearly all participants, the calibration algorithm is simple and was easily programmed into the computer reading software at the CT reading center so that attenuation adjustments were automatic, and, despite the large volume of scans (close to 20 000) read at the CT reading center, the additional scoring and reading burden required to obtain calibrated CAC measurements was not prohibitive.
Most importantly, implementation of the image attenuation calibration method yielded the desired effect of more comparable CAC measurements between images obtained with different scanners and in different participants. Specifically, the direction and size of the Agatston score adjustments were consistent with the phantom attenuation results in that calibration yielded significant increases in scores at sites and in participants with lower image attenuation values and significant decreases in scores at sites and in participants with higher image attenuation values. For instance, sites that used Siemens multi–detector row CT scanners, which yielded images that had the lowest phantom attenuation values and the highest kilovoltage, showed the greatest increases in scores when attenuation adjustment was applied. It must be made clear, though, that the overall magnitude and direction of the calibration adjustment are simply a reflection of the chosen standard values. Specifically, on average, in CARDIA and MESA we observed a decrease in Agatston scores owing to calibration (–8 ± 50), and this relates directly to the fact that the selected SBHW standards to which images were calibrated were slightly lower than those used in CARDIA and MESA.
Although improved comparability in terms of image attenuation was achieved by performing the calibration, the Agatston score adjustments were small on average, and the correlation between the unadjusted and the adjusted scores was high. This was encouraging in that it indicated that the adjusted scores were measuring the same inherent characteristic (calcium burden) as the unadjusted scores. It implies that in cross-sectional analyses in research studies, the use of adjusted versus the use of unadjusted measurements will not make a material difference. Although one possible interpretation of these statistics is that the effect of attenuation adjustment is minimal and that therefore it is not worth doing, it is important to note that these statistics reflect average effects in the sample. Some participant-specific Agatston score adjustments were large (up to 1071 in our data), and some were important, yielding changes in the clinical categorization of CAC burden. The large differences were primarily seen for participants with higher unadjusted scores and can be expected even for small calibrations in image attenuation. Large differences occur when (a) several calcified lesions have (or one relatively large lesion has) CT numbers close to the thresholds used in the Agatston score formula (130, 200, 300, and 400 HU), and (b) image attenuation adjustment causes those CT numbers to cross over the threshold. The existence of large and important differences in our data suggests that in research and in clinical practice, when one is scanning individuals at a single point in time, and particularly when one is performing repeat examinations to evaluate change over time, it may be important to calibrate CAC measurements to account for image attenuation differences.
Conclusions and Limitations
The use of calibration phantoms to adjust image attenuation values and compute calibrated calcium measurements is potentially important for CT scanning in both research studies and clinical practice. Clearly, there are important differences in attenuation and in Agatston scores between different scanners, particularly when different kilovoltage settings are used. Research studies that involve the use of different scanner technologies would most likely benefit from the utilization of attenuation adjustment. Our data also suggest that adjustment of attenuation due to body size should be performed for all examinations performed both for research and for clinical purposes. Although this may be the case, the physics of x-ray imaging are complex, and our use of a phantom that is placed relatively distant from the coronary arteries may not result in appropriate measurement of attenuation in the coronary tree in the same image section.
We and others (26,27) have noted attenuation gradients across the field of view in images acquired with Imatron scanners. Such gradients may depend on body size and thus may make interpretation of and correction for attenuation differences with the same scanner difficult. However, without more extensive research, we cannot make a strong case for universal application of attenuation adjustment solely on the basis of differences in body size between participants. We have shown, however, that image attenuation, and therefore the meaning of an Agatston score, is not the same from scanner to scanner and from participant to participant. Thus, the lack of calibration in clinical practice means that different CAC score cut points are being used for referrals, a problem discussed in detail by Mitka (28).
In addition, the proposed calibration method could be improved and further refined. For instance, a larger sample of representative images (such as the CARDIA and MESA images) could be used in the future to define the standard attenuation values. Also, there may be some additional reduction in image attenuation variability if calibration is performed on an image-specific, as opposed to participant-specific, basis, although our data suggest that these gains are likely to be minimal.
Finally, additional data are needed to more fully evaluate the potential benefit of attenuation-adjusted CAC measurements. Coronary heart disease morbidity and mortality follow-up data and longitudinal measurements of CAC over time are being collected in CARDIA and MESA, and these data may be used to compare the ability of unadjusted versus attenuation-adjusted CAC measurements to predict coronary heart disease end points and to compare models predicting change in CAC over time. Hopefully, these data can provide additional insight into the extent of the utility of calibrating CAC measurements in specific clinical and research settings.
|
APPENDIX |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSION APPENDIX REFERENCES |
|---|
|
Notation.—Before the collection of CT data in CARDIA and MESA, calibration phantom data from the SBHW study (27 subjects; 28–60 CT sections available per subject) were used to develop the calibration algorithm. The average CT numbers in the SBHW sample corresponding to the four known calcium concentration values (0, 50, 100, and 200 mg/mL) in a calibration phantom were computed, programmed into the reading software at the CT reading center, and used as the standards to which all images were calibrated during the course of CARDIA and MESA. The SBHW study involved use of a GE-Imatron C150 CT scanning system with the following parameters: 130 kVp; 630 mA; and effective exposure time, 0.10 second.
If i indicates the subject (i = 1, 2, ... , n), j denotes the image section acquired in a given subject (j = 1, 2, ... , ni), and k indicates the calibration phantom rod (k = 1, 2, 3, or 4), the following variables can be defined:
RAijk, the observed radiologic attenuation (in Hounsfield units) for calibration rod k on section j for subject i.
PCk, the physical concentration (in milligrams per milliliter) for calibration rod k (constant for all subjects and sections).
RAkstd, the average radiologic attenuation (in Hounsfield units) computed for calibration rod k across all subjects and image sections in the SBHW data set (Table A2).
|
Algorithm Details
To calibrate a new CT image, the observed image attenuation (in Hounsfield units) for each pixel in the image is converted to standard attenuation units (in Hounsfield units) on the basis of that image’s observed radiologic attenuation values for the calibration phantom. This differs from the approach of McCollough et al (17), in which observed attenuation is converted to physical calcium units (in milliliters). So that we could obtain calibrated CAC measurements, we needed to convert observed attenuation into standard radiologic (not physical) attenuation values. Specifically, the calibration method is applied by using the following algorithm:
1. For the jth observed image section for subject i, perform a linear regression with Yijk = RAijk and Xk = RAkstd for k = 1, 2, 3, or 4. Extract the intercept (aij) and slope (bij) estimates obtained from the following regression equation: Yij = aij + (bij · X). Repeat for all sections (j = 1, 2, ... , ni) for subject i.
2. For the ith subject, compute the median intercept and the median slope across all sections: ai = median(aij); bi = median(bij). Use the regression equation based on the median parameters, ai and bi, to adjust the pixels in all sections for subject i. Specifically, for each pixel in image section i, perform the following attenuation conversion: NPCTno = (OPCTno – ai)/bi, where NPCTno is the new CT number for the pixel and OPCTno is the original CT number for the pixel. Repeat for all subjects (i = 1, 2, ... , n).
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Authors stated no financial relationship to disclose.
Author contributions: Guarantors of integrity of entire study, J.C.N., R.D.; study concepts and design, J.C.N., R.A.K., J.J.C., M.F.M.G., R.D.; literature research, J.C.N., N.D.W., R.D.; data acquisition, all authors; data analysis/interpretation, J.C.N., R.A.K., N.D.W., M.F.M.G., R.D.; statistical analysis, J.C.N., R.A.K.; manuscript preparation, definition of intellectual content, editing, revision/review, and final version approval, all authors
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSION APPENDIX REFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  R. L McClelland, D. E Bild, G. L Burke, K. J Mukamal, J. A Lima, and R. A Kronmal Alcohol and coronary artery calcium prevalence, incidence, and progression: results from the Multi-Ethnic Study of Atherosclerosis (MESA) Am. J. Clinical Nutrition, December 1, 2008; 88(6): 1593 - 1601. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. R. Brown, R. A. Kronmal, D. A. Bluemke, A. D. Guerci, J. J. Carr, J. Goldin, and R. Detrano Coronary Calcium Coverage Score: Determination, Correlates, and Predictive Accuracy in the Multi-Ethnic Study of Atherosclerosis Radiology, June 1, 2008; 247(3): 669 - 675. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. Detrano, A. D. Guerci, J. J. Carr, D. E. Bild, G. Burke, A. R. Folsom, K. Liu, S. Shea, M. Szklo, D. A. Bluemke, et al. Coronary Calcium as a Predictor of Coronary Events in Four Racial or Ethnic Groups N. Engl. J. Med., March 27, 2008; 358(13): 1336 - 1345. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H. Chung, R. L. McClelland, R. Katz, J. J. Carr, and M. J. Budoff Repeatability Limits for Measurement of Coronary Artery Calcified Plaque with Cardiac CT in the Multi-Ethnic Study of Atherosclerosis Am. J. Roentgenol., February 1, 2008; 190(2): W87 - W92. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. Y. Wong, N. Cheung, F. M. A. Islam, R. Klein, M. H. Criqui, M. F. Cotch, J. J. Carr, B. E. K. Klein, and A. R. Sharrett Relation of Retinopathy to Coronary Artery Calcification: The Multi-Ethnic Study of Atherosclerosis Am. J. Epidemiol., January 1, 2008; 167(1): 51 - 58. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. A. Kronmal, R. L. McClelland, R. Detrano, S. Shea, J. A. Lima, M. Cushman, D. E. Bild, and G. L. Burke Risk Factors for the Progression of Coronary Artery Calcification in Asymptomatic Subjects: Results From the Multi-Ethnic Study of Atherosclerosis (MESA) Circulation, May 29, 2007; 115(21): 2722 - 2730. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. A. Birnbaum, N. Hindman, J. Lee, and J. S. Babb Multi-Detector Row CT Attenuation Measurements: Assessment of Intra- and Interscanner Variability with an Anthropomorphic Body CT Phantom Radiology, January 1, 2007; 242(1): 109 - 119. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. C. Nelson, X.-C. Jiang, I. Tabas, A. Tall, and S. Shea Plasma Sphingomyelin and Subclinical Atherosclerosis: Findings from the Multi-Ethnic Study of Atherosclerosis Am. J. Epidemiol., May 15, 2006; 163(10): 903 - 912. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. L. McClelland, H. Chung, R. Detrano, W. Post, and R. A. Kronmal Distribution of Coronary Artery Calcium by Race, Gender, and Age: Results from the Multi-Ethnic Study of Atherosclerosis (MESA) Circulation, January 3, 2006; 113(1): 30 - 37. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. C. Detrano, M. Anderson, J. Nelson, N. D. Wong, J. J. Carr, M. McNitt-Gray, and D. E. Bild Coronary Calcium Measurements: Effect of CT Scanner Type and Calcium Measure on Rescan Reproducibility--MESA Study Radiology, August 1, 2005; 236(2): 477 - 484. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
) and 95% CIs (error bars) for quality-control torso phantom attenuation measurements (CT numbers in Hounsfield units) according to scanning site (top) and scanner type (bottom). Sites 1 and 2 used Siemens multi-detector row CT (MDT) scanners. Sites 3-6 used GE-Imatron electron-beam CT (EBT) scanners. Sites 7-8 used GE LightSpeed multi-detector row CT scanners. Means differed significantly across sites (P < .001) and scanner types (P < .001) and were lowest for Siemens multi-detector row CT scanners (at 140 kVp), followed by GE-Imatron electron-beam scanners (at 130 kVp) and GE LightSpeed multi-detector row scanners (at 120 kVp). Means also varied between sites that used the same type of scanner.