| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Cardiac Imaging |
1 From the Colleges of Medicine (W.S., T.L.B., B.H.T., R.M.L., L.T.M.) and Public Health (T.L.B., J.D.W., R.M.L., L.T.M.), University of Iowa, 200 Hawkins Dr, 3896 JPP, Iowa City, IA 52242. Received July 1, 2002; revision requested August 28; final revision received March 14, 2003; accepted May 19. L.T.M. supported by National Heart, Lung, and Blood Institute grant R01 HL48050 "CT Vascular Calcium—An Epidemiologic Study in the Young." Address correspondence to W.S. (e-mail: william-stanford@uiowa.edu).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: Mean attenuation values for three calcium phantoms (each with a different calcium hydroxyapatite concentration), as measured at each of four different image section levels, were obtained for 691 participants in the Muscatine CT Vascular Calcium Study. The subjects were grouped according to sex-specific BMI quartiles, and the degree of attenuation in each phantom was investigated as a function of image section level and BMI quartile. Spearman rank order correlation coefficients and one-, two-, and three-factor repeated-measures analysis of variance were used to examine the association between section level and BMI and the mean phantom attenuations.
RESULTS: Attenuation was, for the most part, significantly associated with both section level (P < .005) and BMI quartile (P < .0025–.05). The degree of attenuation tended to decrease in images obtained at the more distal cardiac levels and to increase with increasing BMI quartile.
CONCLUSION: Differences in attenuation related to BMI and image section level appear to have a significant effect on current calcium scoring methods.
© RSNA, 2004
Index terms: Computed tomography (CT), experimental studies, 54.12118 • Coronary vessels, calcification, 54.75, 54.81 • Coronary vessels, CT, 54.12118 • Phantoms • Test objects
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
To determine the baseline attenuation value for each of the calcium hydroxyapatite mattress phantoms, uniform tissue-equivalent polyurethane cylindrical body phantoms of 22, 30, and 36 cm in diameter were placed on the mattress and scanned. These body phantoms (Image Analysis) had an atomic number ± 2 of muscle and a physical density of 1.02 g/cm3. The phantoms simulated different-sized individuals and were similar to the phantoms used at Imatron to standardize their scanners. The CT attenuation of each of the hydroxyapatite phantoms was determined, and the value was used as the standard of reference against which each subject’s measurements were compared (Fig 1).
|
The analyses focused on attenuation measurements for each of the three calcium phantom cylinders obtained at four different image section levels for each subject. The initial set of phantom measurements was obtained from the first section in which the left main and/or left anterior descending coronary artery was visualized. This was referred to as level 0. The second set of measurements was obtained from the section 10 sections below level 0, referred to as level 10. The third set of measurements was obtained from the section 20 sections below level 0, referred to as level 20, and the fourth set of measurements was obtained from the section 30 sections below level 0, referred to as level 30.
Sex-specific BMI quartiles (with BMI calculated as weight in kilograms divided by height in meters squared) were determined and used to classify the subjects into four groups (where group 1 included subjects with the lowest BMIs and group 4 included subjects with the highest BMIs) so that measured mean phantom attenuation values could be examined according to BMI quartile.
The dependent variable was the measured mean attenuation value in Hounsfield units for each of the three phantom cylinders at each of the four section levels—that is, levels 0, 10, 20, and 30. The measurement at section level 10 represented an average of the measurements at sections 9, 10, and 11; the measurement at section level 20, an average of the measurements at sections 19, 20, and 21; and the measurement at section level 30, an average of the measurements at sections 29, 30, and 31. These measurements were obtained by placing the regions of interest to include as much of the circumference of each phantom as possible. Each region of interest measured 215.89 mm2 and was placed by a single senior CT technologist. Only phantom data (ie, no coronary calcium data) were included in this analysis.
Parameters Evaluated
The following parameters were evaluated: (a) measured mean phantom attenuation by section level, (b) the association between BMI and measured mean attenuation, (c) results of subject- and section level–specific linear regression, and (d) results of estimation of the attenuation value that would be read as 130 HU on a calcium scan. All of the data analyses were performed by two of the authors (T.L.B., J.D.W.).
Data and Statistical Analyses
Sex-specific means, SDs, and minimum and maximum attenuation values were used to describe the phantom data. Spearman rank order correlation coefficients were used to examine the association between BMI and measured mean phantom attenuation. The body attenuation phantoms simulated different-sized individuals, and these values were used to determine a standard of reference. Subject- and section level–specific linear regression models were fit by using the measured mean attenuation for phantoms 1, 2, and 3 as the dependent variable and the three standard-of-reference attenuation values as the independent variable.
One-, two-, and three-factor repeated-measures analysis of variance (ANOVA) models were used to analyze the phantom data and the slopes and intercepts from the fitted regression models. The following factors were considered in various combinations: (a) section level (levels 0, 10, 20, and 30), (b) phantom (phantoms 1, 2, and 3), and (c) BMI quartile (quartiles 1, 2, 3, and 4). Least-squares means were estimated on the basis of the repeated-measures ANOVA models, and a Bonferroni correction was used to determine the appropriate P value when multiple pairwise comparisons among section level–specific and/or phantom-specific and/or BMI quartile–specific means were performed. Procedures from the Statistical Analysis System software (SAS version 8; SAS Institute, Cary, NC) were used in all of the analyses, and a P value less than .05 was considered to indicate a statistically significant difference.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
Phantom Cylinder Attenuation
The mean attenuation values (as averaged across 40 3-mm sections) for each body attenuation phantom cylinder were as follows: -11.9 HU for phantom 1 (water, 0 mg Ca2+/mL), 89.9 HU for phantom 2 (75 mg Ca2+/mL), and 162.9 HU for phantom 3 (150 mg Ca2+/mL). These reference standard attenuation measurements were obtained by averaging the attenuation measurements for the 22-, 26-, and 30-cm tissue-equivalent body phantoms for each of the calcium mattress phantoms.
Measured Mean Phantom Attenuation by Section Level
For the most part, mean phantom attenuation in women and in men was significantly different from the reference standard values. In women, the measured mean attenuation value for phantom 1 was significantly higher (P < .005), that for phantom 2 was significantly lower (P < .005), and that for phantom 3 was significantly higher (P < .005) at each section level than the reference standard value (Table 1). In men, the measured mean attenuation value for phantom 1 was significantly higher at section levels 0 and 30 (P < .005), that for phantom 2 was significantly lower at each section level (P < .005), and that for phantom 3 was significantly lower at section levels 0 and 10 (P < .005) and significantly higher at section level 30 (P < .005) than the reference standard value. In general, there appeared to be less attenuation at the more distal sections than at the more proximal sections—that is, the observed means were closer to the reference standard values in the more distal sections. However, there was considerable variability in mean attenuation among subjects, as indicated by the phantom level–specific minimum and maximum values.
|
In women, there were no significant differences between any of the section level–specific pairs of mean attenuation values for phantom 1; however, there were significant pairwise differences among the section level–specific means for phantoms 2 and 3. For phantom 2, each pair of section level–specific means was significantly different (P < .0025 after correction for multiple comparisons). For phantom 3, each pair of means (except for section level 10 vs section level 30) was also significantly different (P < .0025) (Table 1).
There were also significant pairwise differences among section level–specific means for phantoms 1, 2, and 3 for men. For phantom 1, there was a significant difference for section level 0 versus section level 10, level 0 versus level 20, level 10 versus level 20, and level 20 versus level 30 (P < .0025 for all) and for level 0 versus level 30 (P < .01). For phantom 2, each pair of means was significantly different (P < .0025). For phantom 3, differences for all pairs were significant (P < .0025), with the exception of the difference between section level 20 and section level 30, which was not significant (Table 1).
These results indicate significant overall measured mean attenuation differences for phantoms 2 and 3 among the four different image section levels with two exceptions (section level 10 vs section level 30 for phantom 3 for women and section level 20 vs section level 30 for phantom 3 for men). They also indicate a considerable amount of variation among individuals within the data for each phantom according to section level category.
Reproducibility of Mean Phantom Attenuation Values
To verify the reproducibility of mean phantom attenuation values, duplicate measurements were examined. Repeat-scan phantom measurements at all four section levels were available for 279 of the 691 subjects. The similarity between the two measurements by level and by phantom was examined by estimating intraclass correlation coefficients. For phantom 1, the correlation coefficients across the four section levels ranged from 0.73 to 0.85 (P < .0001 for all); for phantom 2, they ranged from 0.85 to 0.91 (P < .0001 for all); and for phantom 3, they ranged from 0.76 to 0.86 (P < .0001 for all). This suggests that the phantom measurements from two scans obtained on the same day are reproducible.
Association between BMI and Measured Mean Attenuation
The association between BMI and measured mean attenuation was examined for each phantom at each level by computing sex-specific Spearman rank order correlation coefficients. The magnitude of the correlation coefficients for phantom 1 ranged from -0.11 to 0.02 in women and from -0.13 to -0.01 in men across the four section levels. Many of these correlation coefficients were not significant. The magnitude of the association increased dramatically for phantoms 2 and 3 (-0.61 to -0.49 in women and -0.58 to -0.34 in men; P < .0001 for all). The negative values for these correlation coefficients indicate that there tended to be a greater degree of attenuation in subjects with larger BMIs.
The measured mean phantom attenuation was also examined by BMI quartile, section level, and phantom. The sex-specific quartiles for BMI were as follows: 22.3 (25th percentile), 25.2 (50th percentile), and 30.0 (75th percentile) for women; and 24.2 (25th percentile), 26.9 (50th percentile), and 30.5 (75th percentile) for men. There was a reasonably consistent mean attenuation pattern for phantoms 2 and 3, indicating an increase in attenuation with increasing body size (Fig 2).
|
|
|
|
|
|
Results of Subject- and Section Level–Specific Linear Regression Models
For each subject and at each of the four section levels, a linear regression analysis of the three measured mean phantom attenuation values versus the three reference standard attenuation values was performed, and intercepts and slopes were estimated. Sex-specific one-factor (ie, section level) repeated-measures ANOVA models were used to analyze the mean intercepts and slopes from these subject- and section level–specific fitted linear regression models. For both men and women, there was a significant difference (P < .0001) among the mean slopes and intercepts for section levels 0, 10, 20, and 30. For women, the mean slopes for section levels 10 (0.97), 20 (1.00), and 30 (0.98) were significantly higher than that for level 0 (0.95), reflecting the fact that there was less attenuation for phantoms 2 and 3 in the more distal sections. For men, the mean slopes for section levels 10 (0.96), 20 (0.99), and 30 (0.98) were also significantly higher than that for level 0 (0.93). For women, each pairwise comparison of mean slopes yielded a significant difference (P < .001) except the comparison for section level 10 versus section level 30 (P < .025). For men, each pairwise comparison of mean slopes revealed a significant difference (P < .001).
Estimation of the Attenuation Value That Would Be Read as 130 HU on a Calcium Scan
Parameter estimates from the subject- and section level–specific fitted regression lines were used to estimate the attenuation value in Hounsfield units that actually corresponded to a voxel that would be read as 130 HU (the accepted threshold value for identifying coronary artery calcium [1]). This attenuation value is important because the change in attenuation due to BMI implies that a voxel read as 130 HU actually represents a voxel with higher attenuation. As one example (Fig 4), the intercept of the fitted regression line for one subject at one section level was -1.9641, and the slope was 0.9404. The value of x that satisfied the equation -1.9641 + 0.9404x = 130 HU was calculated to be 140.3 HU. This indicates that if the change in attenuation is taken into account, a voxel identified as having an attenuation value of 130 HU actually represents a voxel with an attenuation value of 140.3 HU. Table 4 displays the means of these estimated attenuation values by section level and BMI quartile for women and men. The results of pairwise comparisons among section level– and BMI quartile–specific estimated mean attenuation values are summarized in the footnote to Table 4.
|
|
Because of the association between phantom measurements and body size, subthreshold lesions (ie, those < 130 HU in attenuation) may reach a threshold level if phantom data are incorporated into the measurement process, and, hence, total calcium scores, areas, and volumes may change. To investigate the possible use of a lower threshold, we estimated subject- and section level–specific thresholds that corresponded to a measured attenuation of 130 HU and found an actual attenuation of 120.3 HU (Fig 4). This is in contrast to the method we used to create Table 4, in which we estimated the attenuation that corresponded to a 130-HU threshold. These estimated threshold levels ranged from 101 to 145 HU in women, with mean values of 123, 127, 131, and 130 HU for section levels 0, 10, 20, and 30, respectively. In men, they ranged from 90 to 142 HU, with means of 114, 116, 120, and 123 HU, respectively.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
Watson et al (8) used a post hoc phantom data adjustment method to quantitate calcium. The electron-beam CT scan was scored, and the mean attenuation was then adjusted for the degree of attenuation. The slope of the phantom calibration line calculated for each scan but from only one scoring of the phantom was used to normalize the mean CT number (attenuation) for a lesion. Before division by the slope, the mean CT number for the adjacent cardiac blood pool was subtracted from the mean CT number of the lesion. On the basis of the results of our investigation of the phantom used in the Muscatine CT Vascular Calcium Study, this adjustment would not appear to be sufficient because of the differences in attenuation from the proximal to the distal portions of the heart. Also, it is our contention that if a subject- and section-specific threshold is used during the scoring process, a lesion with an attenuation value that did not reach the 130-HU minimum standard threshold might be identified rather than missed. Thus, to apply a post hoc adjustment to an observed calcium measure may not be adequate.
Raggi et al (9) measured the soft-tissue attenuation of the regions surrounding the coronary arteries at the level of the left main coronary artery ostium (high) and at the bottom of the heart (low) with the aim of assessing the inter- and intraindividual variability of the attenuation threshold used to identify coronary calcification at electron-beam CT. The investigators found a strong association with body weight for values adjacent to the distal left anterior descending coronary artery and a weak but still significant association at the higher level. Their conclusion was that a coronary calcium threshold of 120 HU seemed more appropriate than a threshold of 130 HU.
Unlike the very strong association between body size measurements and phantom attenuation values that we observed, no association was found for 250 consecutive patients who underwent electron-beam CT in the study of Greaser et al (10). They reported that when they applied a phantom-data correction (by using the averages of the attenuation values in three superior, three middle, and three inferior phantom images) to images obtained in 28 patients with dual examinations and low but positive coronary artery calcium scores, they found only a slight reduction in interscan variation. Their conclusion was that the use of phantom data does not reduce interscan variation to an extent sufficient to warrant regular clinical use. This conclusion was based on results of the examination of a relatively small number of subjects, and their approach to adjustment was extremely labor intensive in that the voxels in each section were adjusted by a correction factor based on the phantom data.
Carr et al (11) evaluated a calcium phantom with an electron-beam CT scanner and with a subsecond-gated helical CT scanner. At each of four concentrations in the phantom, the measurement with electron-beam CT averaged from 25 to 30 HU greater than the measurement with helical CT. For electron-beam CT, the standard threshold of 130 HU corresponded to a calcium concentration of 106 mg/mL. They suggest the need for a standard based on the concentration of calcium for the evaluation of progression and regression of coronary calcium over time.
To evaluate phantom attenuation in the present study, different-sized tissue-equivalent polyurethane body phantoms were used to simulate individuals of different body habitus. This method is used by the manufacturer and was believed to more likely replicate the actual scanning protocol than would the method of comparing results with a fixed calcium value.
In calcium scoring it must be remembered that the initial Agatston threshold score of 130 HU was developed on the basis of where the CT number of calcium was positioned in relation to that of blood. Therefore, a comparison of the 130-HU threshold with an absolute concentration of calcium does not replicate the original calculation.
Still, the need for phantom-data adjustment is not universally accepted. The importance of the data presented herein is that there appears to be a significant attenuation effect on calcium measurements that is associated with body size. The extent of the attenuation effect also varies with the section level of the image. Whether these section level attenuation changes are due to the presence of liver tissue, increased tissue thickness at the diaphragmatic level, or increased scatter is unclear; all these factors may play a role. As a result, it is our recommendation that a standard phantom data analysis be performed with each calcium CT examination. Because a single phantom measurement is not adequate for all section levels, it is important that the length of the phantom extend from the base of the heart to the apex.
The uncorrected threshold of 130 HU was selected because it is much higher than values ascribed to myocardium, and values above this threshold relate to the pathologic finding of calcium and advanced atherosclerosis (2). Thus, 130 HU is still an acceptable threshold for defining the presence of coronary artery calcium. Our data raise several issues that need to receive further consideration. With the lower attenuation observed at the more distal distributions of the coronary arteries, a lower-attenuation distal lesion can appear to be of similar magnitude as a higher-attenuation proximal lesion. If a lower threshold were to be used, scatter or noise may be observed in distal sections, and a lower-attenuation voxel could be mislabeled as a lesion. The actual Agatston method scores (1) may need to be adjusted for phantom data in studies in which multiple electron-beam CT scanners are used (each scanner will require its own phantom data– and body size–adjusted reference standard) or in which comparisons between and among subjects undergoing serial examinations over several years (during which body size may change) are performed.
In summary, differences in attenuation related to BMI and image section level appear to have a significant effect on current calcium scoring methods, and there appears to be a need for use of a phantom for value adjustments in longitudinal and multicenter investigations.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Author contributions: Guarantors of integrity of entire study, W.S., T.L.B.; study concepts, W.S., T.L.B., L.T.M., R.M.L.; study design, W.S., T.L.B., L.T.M., R.M.L., B.H.T.; literature research, W.S., B.H.T.; clinical and experimental studies, W.S., B.H.T.; data acquisition, W.S., B.H.T.; data analysis/interpretation, T.L.B., J.D.W.; statistical analysis, T.L.B., J.D.W.; manuscript preparation, W.S., T.L.B., L.T.M.; manuscript definition of intellectual content, W.S.; manuscript editing and final version approval, W.S., T.L.B., L.T.M., R.M.L., B.H.T.; manuscript revision/review, W.S.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  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]
|
|
|
|
 |
|
 C. H. McCollough, S. Ulzheimer, S. S. Halliburton, K. Shanneik, R. D. White, and W. A. Kalender Coronary Artery Calcium: A Multi-institutional, Multimanufacturer International Standard for Quantification at Cardiac CT Radiology, May 1, 2007; 243(2): 527 - 538. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. C. Nelson, R. A. Kronmal, J. J. Carr, M. F. McNitt-Gray, N. D. Wong, C. M. Loria, J. G. Goldin, O. D. Williams, and R. Detrano Measuring Coronary Calcium on CT Images Adjusted for Attenuation Differences Radiology, May 1, 2005; 235(2): 403 - 414. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
= actual data point.