| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 All authors: Department of Radiology, Onze Lieve Vrouw Hospital, Moorselbaan 164, Aalst 9300 Belgium.
Received February 10, 2003;
accepted after revision April 3, 2003.
Address correspondence to L. R. Van Hoe
(lievenvanhoe{at}hotmail.com).
Abstract
 Top Abstract IntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
SUBJECTS AND METHODS. In 50 patients, the coronary arteries were investigated twice with ECG-gated MDCT with 500-msec rotation time. Per scan, three sets of images were reconstructed, respectively, at an image-reconstruction window of 40%, 50%, and 60% of the R-R interval. Image quality was assessed, and the optimal image-reconstruction window per scan and per coronary territory was determined. The interscan variability of calcium mass measurements was calculated for different strategies (use of fixed image-reconstruction window [40%, 50%, or 60%] versus individually optimized image-reconstruction window).
RESULTS. A significant improvement in image quality was obtained by selecting the best of three reconstructed data sets (mean image quality score, 4.4 vs 3.7; p < 0.001). Even with individually optimized image-reconstruction window values, we obtained high values for interscan variability (mean ± SD, 27% ± 22% vs 31% ± 35% with a fixed image-reconstruction window).
CONCLUSION. The use of individually optimized image-reconstruction windows leads to a significant improvement in image quality. However, interscan variability of calcium mass measurements remains high.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
The normal progression of coronary artery calcium is approximately 14–27% (average, 24%) increase per year and may be enhanced up to 33–48% in patients with advanced disease [4–6]. For reliable monitoring of the progression of disease, the measurements obtained should be reproducible, and interscan variability should be sufficiently low, preferably below 10% [7].
Although electron beam CT is still considered the gold standard for in vivo detection and quantification of coronary calcium [8–12], encouraging results have been obtained with multidetector CT (MDCT) [7]. However, one recent study reported poor results for interscan variability (between 39% and 60%) using MDCT [13].
Several studies have shown that in ECG-gated helical CT, the choice of the image-reconstruction window (expressed as a percentage of the R-R interval) has an important influence on image quality [14–16]. In clinical practice, two different approaches are possible. The fastest and easiest approach is to reconstruct images at predefined image-reconstruction windows (e.g., 40% for the right coronary artery and 60% for the left anterior descending artery) and to use these image data sets for coronary artery calcium measurement. Alternatively, for each patient, several image data sets may be reconstructed at different image-reconstruction windows, and the data set containing the fewest motion artifacts may be used for further evaluation. The latter procedure requires additional time and effort and, ideally, image transfer to dedicated workstations. Although the additional time and effort are worthwhile in coronary CT arteriography, the precise effect of this procedure on calcium measurement reproducibility remains to be determined.
In this study, interscan variability of coronary artery mass measurements with MDCT was assessed. We compared two strategies: the use of image data sets obtained at fixed (predefined) image-reconstruction windows versus individual selection of the optimal image-reconstruction window (40%, 50%, or 60%). These specific image-reconstruction window values were chosen on the basis of the results of earlier studies on image quality [14, 15, 17]. Also, the correlation of interscan variability with heart rate and absolute calcium mass was assessed to test the hypothesis that the technique could be used to determine progression of disease, at least in specified subgroups of patients (such as those with low heart rate or those with high amounts of calcium).
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
The interval between the two scans ranged between 80 min and 48 hr (mean ± SD, 16 ± 12 hr). Twenty-three patients were ambulatory, and 27 were hospitalized. Informed consent was obtained from each patient after the nature of the procedure and the purpose of the study were fully explained. The study protocol was approved by the institutional review board.
CT Technique
All patients were examined on MDCT (Siemens Volume Zoom, Erlangen,
Germany). The heart rate of the patient was recorded during the study.
Patients were instructed to briefly hyperventilate just before scanning.
Images were obtained during breath-holding, and an ECG trace was obtained
during data acquisition. MDCT was begun at the level of the aortic root above
the coronary ostia and included the entire heart. Imaging parameters were the
following: 120 kV, 80 mA, 4 x 2.5 mm collimation, 0.5-sec rotation time,
pitch of 1.5, and 1.5-mm image-reconstruction interval. A medium sharp
convolution kernel (B35f) was used. Images were reconstructed using the heart
rate–adaptive algorithm in diastole with a 3-mm slice width and a
temporal resolution of 250 msec for heart rates of less than 70 beats/min and
125 msec for heart rates of 70 beats/min or faster
[18]. For each data set, three
sets of images were reconstructed using retrospective ECG-gating beginning at
40%, 50%, and 60% of the R-R interval
[14,
15,
17]. The same protocol was
used for the second scan.
Assessment of Image Quality
Per patient and per scan images obtained at 40%, 50%, and 60% of the R-R
interval were transferred to a dedicated viewing station (Magic View,
Siemens), and simultaneous cine mode evaluation of the three sets was
performed independently by two observers. Images obtained at corresponding
slice positions were compared, and a score for relative image quality (3,
best; 2, intermediate; 1, least adequate) was assigned to each image set with
respect to visualization of the left anterior descending artery, the left
circumflex coronary artery, and the right coronary artery. Thus, each observer
assigned three scores to each image data set, one per coronary territory.
Moreover, the quality of visualization of the left anterior descending artery, the right coronary artery, and the left circumflex coronary artery was scored using a 5-point scale [15]: 5, no motion artifact (clear delineation); 4, minor artifact (slight blurring); 3, moderate artifact (double-imaged structures in more than half of the course of the vessel); 2, severe artifact (doubling and blurring over the whole course of the major vessel with structure discontinuity); or 1, nondiagnostic (vessel structures not differentiable).
Calcium Scoring
For measurement of coronary artery calcium, commercially available software
(HeartView, Siemens) was used. Calcified lesions were identified on the basis
of a threshold of 130 H for all applied scoring methods. Coronary stents were
identified and excluded during the scoring process. Scores for a normalized
Agatston score and calcium mass were obtained
[7]. Separate scores were
calculated for the left main coronary artery, right coronary artery, left
circumflex coronary artery, and left anterior descending artery.
Number of Scores Available for Data Processing
For calculation of measurement reproducibility, only measurements of
calcium mass were used because previous studies suggested that the use of the
traditional or modified Agatston scores provided a lower reproducibility
compared with measurement of calcium mass or volume
[7] and that calcium mass
quantification may be slightly more reproducible than calcium volume
quantification [19].
Per patient and per scan, total calcium mass was calculated separately for images reconstructed at 40%, 50%, and 60% of the R-R interval. Moreover, an individually optimized score for calcium mass was calculated for each scan. To do this, we determined the optimal image data set (i.e., the image data set with the highest image quality) separately for the the right coronary artery, the left circumflex coronary artery, and the left anterior descending artery, and the corresponding scores for calcium mass were added. For the left main coronary artery, the same image data set was used as that used for the left anterior descending artery. Thus, four scores for total calcium mass were available per scan and per patient (three using a fixed image-reconstruction window plus one optimized score). The entire procedure was performed independently by two different observers who were the same as those assigning scores for image quality.
Three months later, all measurements of coronary calcium mass were obtained for the second time by the same observers to allow calculation of intraobserver variability. Sixteen hundred scores were available for total calcium mass: 50 patients, two scans, two observers, two sessions, four scores (three using fixed image-reconstruction intervals and one optimized score).
Calculation of Intraobserver, Interobserver, and Interscan
Variability
Measurement variability was calculated per patient and was expressed as the
percentage of difference and as the coefficient of variation. The percentage
of difference between two measurements can be calculated as 2
(|measurement 1 – measurement 2|) / (measurement 1 +
measurement 2) [7,
20,
21]. The percentage of
interobserver variability was calculated for each pair of measurements
obtained by different observers but with identical parameters: scan, session,
and image-reconstruction interval. Similarly, intraobserver variability was
calculated using each pair of measurements obtained in a different session but
with identical parameters; scan, observer, image-reconstruction interval.
Finally interscan variability was calculated using pairs of measurements
obtained on different scans but with identical parameters: observer, session,
and image-reconstruction interval. Mean values were used for data analysis.
The percentage of variability could not be calculated in three patients
because the most measurements were equal to zero. These patients were excluded
from the study.
The coefficient of variation of a set of measurements corresponds to the SD of the measurements divided by the mean [22]. In this study, the coefficient of variation was calculated separately for variability related to the parameters: scan, session, and observer [23], using a commercially available software package (Statview, version 5.0.1, SAS institute, Cary, NC).
Statistical Analysis
Statistical analysis was performed using a commercially available software
package (6.12, MedCalc Software, Mariakerke, Belgium).
Interobserver agreement in the determination of image quality was assessed by using the kappa statistic. A kappa value greater than 0.6 indicated good interobserver agreement [14]. The McNemar test was used to analyze the different rates of moderate or severe motion artifact (image quality score 1, 2, or 3) in images obtained at different image-reconstruction window values.
The Spearman's rank correlation coefficient was used to assess the correlation between image quality (mean value for both observers) and heart rate. This coefficient can be interpreted as follows: If the the r value is zero, there is no correlation; if the r value is 1, there is a perfect correlation. For r values between 0 and 1, r2 expressed as a percentage indicates the strength of the correlation. For instance, if the r value is 0.4, r2 is 0.16. This means that the correlation is 16%, as strong as it possibly could be. If r has a positive sign, higher values of X tend to be associated with higher values of Y.
Pairwise comparison of values obtained for intra- and interobserver and interscan variability was performed using the paired t test or Wilcoxon's signed rank test, depending on the normality of the distribution. Possible differences in interscan variability between hospitalized and ambulatory patients were assessed using the Mann-Whitney U test. Interscan variability was correlated to the mean heart rate (mean value of mean heart rate recorded at study 1 and 2) and calcium mass by using the Spearman's rank correlation coefficient.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Image Quality
Interobserver agreement in the determination of image quality was good
(
= 0.62). Table 1 shows
the scores for mean image quality in the function of the image-reconstruction
window used (image-reconstruction window fixed at 40%, 50%, or 60% of the R-R
interval versus optimal image-reconstruction window). When comparing the
scores obtained in the same patients, significantly higher scores were
obtained with the use of the optimal image-reconstruction window (p
< 0.001). The mean scores for absolute image quality were highest for the
left anterior descending artery (mean score, 4.2) and lowest for the right
coronary artery (mean score, 3.5). The differences in the mean image quality
among the different coronary arteries were consistent for both observers.
|
The results for the optimal position of the image-reconstruction window in the cardiac cycle are presented in Table 2. For all three vessels, the optimal image-reconstruction window was largely patient-dependent. There was, however, a tendency toward better image quality at an image-reconstruction window of 40% in patients with heart rates of more than 70 beats/min, whereas the optimal cardiac phase in patients with a heart rate of 70 beats/min or less was more frequently at 60%. This pattern was consistently found for all three coronary territories (both observers).
|
The scores for absolute quality were also correlated to the patient's heart rate. Table 3 shows the Spearman's rank correlation coefficients of the image quality mean values for all patients and their heart rates. Spearman's rank correlation coefficients varied between 0.26 and 0.23, indicating a poor and variable correlation. Thus, although the patient's heart rate did influence the optimal phase in the cardiac cycle, it had no major effect on absolute image quality.
|
Intraobserver, Interobserver, and Interscan Variability
The mean percentage of difference for intraobserver variability was 7.4
± 15. The mean coefficient of variation for intraobserver variability
was 8.4 ± 19. The mean percentage of difference for interobserver
variability was 12.8 ± 21. The mean coefficient of variation for
interobserver variability was 26.4 ± 69. Intraobserver variability was
statistically significantly lower than interobserver variability (p
< 0.05). The mean percentage interscan variability was 30.3 ± 31
with the use of an image-reconstruction window of 40%, 33.1 ± 37 with
use of an image-reconstruction window of 50%, 31.7 ± 37 with use of an
image-reconstruction window of 50%, and 27.4 ± 22 with use of the
optimal image-reconstruction window. The mean coefficient of variation for
interscan variability was 37.6 ± 46 with an image-reconstruction window
of 40%, 43.5 ± 54 with an image-reconstruction window of 50%, 38.8
± 49 with an image-reconstruction window of 60%, and 31.3 ± 29
with the optimal image-reconstruction window.
Although there was a clear tendency toward better results with use of the optimized image-reconstruction window, no significant differences were found among the different types of interscan variability (image-reconstruction window fixed at 40%, 50%, or 60% of the R-R interval or optimal image-reconstruction window).
Figure 1A,
1B,
1C,
1D,
1E,
1F is a clinical example
showing large interscan variability. Differences in interscan variability
between hospitalized and ambulatory patients were not statistically
significant. Interscan variability was higher than intra- and interobserver
variability (p < 0.001). As could be expected, interscan
variability showed some inverse correlation to absolute calcium mass. However,
the correlation coefficients were low (
– 0.4), indicating a poor
correlation. Moreover, most correlations were not statistically significant.
The correlation between interscan variability and heart rate was variable and
not statistically significant.
|
|
|
|
|
|
Figure 2 shows the correlation between percentage of interscan variability using an optimized image-reconstruction window and calcium mass in individual patients (r = –0.24). As can be seen in the figure, interscan variability tended to be lower in patients with higher values for calcium mass. However, a large scatter can be observed.
|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Only a few studies have reported on interscan variability of coronary artery calcium measurements using helical CT. Using single-detector helical CT, Goldin et al. [24] obtained mean interscan variability values of 16.5% ± 38% (percentage of difference) and 22.5% (coefficient of variation). Ohnesorge et al. [7] used ECG-gated MDCT and obtained a mean percentage of interscan variability of 8–9%. Although we used the same technique as that of Ohnesorge et al., we found mean interscan variability values that were two to three times higher. No obvious explanation can be given for these striking differences. When comparing the patient-related parameters (age and mean Agatston score) in both studies, we could observe no marked differences. The ability to cope with breath-hold requirements could vary among patient groups and could at least partially explain the differences. Another possible explanation could be the fact that Ohnesorge et al. obtained the second scan within 3–5 min after the first, whereas we used a longer interval between both studies. In another recent study, interscan variability was similar or even higher than that in our study (between 39% and 60%) [13].
In any case, these differences may have important consequences for clinical practice. For instance, if the coefficient of variation of the measurements obtained is 30%, 95% of the measurements are between 40% and 160% of the mean (mean ± 2 SD), and a follow-up of 2–3 years may be required to detect true changes in coronary calcium mass (assuming ± 25% annual increase in coronary calcium mass). If the coefficient of variation was 10%, 95% of the measurements would be between 80% and 120% of the mean, and a follow-up period of approximately 1 year would suffice.
In our study, the strategy to reconstruct and compare three different image data sets at different image-reconstruction windows and to use the image data set showing the least artifact was compared with the (simpler) strategy of using only one predefined image-reconstruction window per coronary vessel. Somewhat surprisingly, we found that the effect of using optimized image data sets on interscan variability was rather marginal and not statistically significant. In other words, the additional time and effort to reconstruct and compare different image data sets did not really solve the problem of poor measurement reproducibility. Like other authors [15, 16], we assessed image quality using simultaneous cine display of axial images, not reformatted or three-dimensional images. The lack of a clear correlation of image quality and interscan reproducibility may be related to the fact that motion artifacts visible on axial reconstructions are mainly related to in-plane motion, whereas the variability of calcium measurements is more related to data gaps caused by motion along the z-axis. This raises the question of whether cine loop display evaluation of axial images alone is sufficient to decide on the optimal image-reconstruction window. A better—but more time-consuming—approach might be to compare reformatted and three-dimensional images at specific workstations [14]. Again, the question of whether this time-consuming approach would significantly reduce interscan variability remains unanswered and requires further study.
The practicing radiologist should be aware of the fact that even with use of apparently optimized image data sets, interscan variability may remain high. This finding can be explained only by rapid multidirectional cardiac motion causing data gaps or overlap, and it raises a fundamental question about the temporal resolution required to successfully and reproducibly investigate coronary arteries.
Inter- and Intraobserver Variability
In this study, mean intra- and interobserver variability was 8.4% and
12.8%, respectively. To our knowledge, no other data are available concerning
intra- and interobserver variability obtained with ECG-gated MDCT. Goldin et
al. [24] used single-detector
helical CT with an 800-msec rotation time and obtained a mean percentage of
interobserver difference of 9.8% and a mean coefficient of variation of 34.6%.
The relatively high values for intra- and interobserver variability obtained
in both studies are not surprising. Coronary artery anatomy is complex, and
particularly for smaller branches, it may be difficult to determine whether a
calcification is really located within the vessel wall or represents an
irrelevant calcification [25].
Also, streak artifacts may be color-marked as calcification by the algorithm
because of their high density; even for an experienced observer, it may be
difficult to determine the exact border between calcification and
artifact.
Image Quality
In this study, the optimal position of the image-reconstruction window for
clear visualization of the coronary arteries varied considerably from patient
to patient. These results confirm those of other recent studies showing that
patient- and coronary artery–specific selection of the
image-reconstruction window is mandatory to achieve optimum image quality for
unenhanced and contrast-enhanced coronary artery CT
[14–16].
In the past, different image-reconstruction windows have been proposed. In
electron-beam CT, the most common trigger time used is 80% of the R-R
interval. However, this trigger occurs on or near the P wave during atrial
systole, and the least motion among all heart rates occurs at 40–60% of
the R-R interval [17].
Therefore, it has been suggested that the protocol of triggering at 80% of the
R-R interval might not be optimal for imaging of the coronary segments near
the right or left atrium. Mao et al.
[26] used a 40% rather than an
80% trigger delay (imaging during early rather than late diastole) and
obtained an interscan variability of 11.5% instead of 17.4%.
Further reduction of coronary motion can theoretically be obtained using different trigger delays for the different coronary vessels. From studies of conventional angiography, we know that each of the three major coronary arteries has a different motion pattern during the cardiac cycle [27]. Initial studies with helical CT seemed to confirm these data, although different results were obtained in two recently performed studies [14, 15]. In a study by Kopp et al. [14], the left anterior descending artery was best visualized in mid diastole at 60–70% of the cardiac cycle, and the left circumflex coronary artery was best visualized at 50%. The optimal reconstruction window for the right coronary artery was significantly different at 40%. In the study by Hong et al. [15], optimal image quality for the left anterior descending artery was obtained equally at 50% and 60% triggering, whereas optimal image quality was achieved with a 60% trigger delay for the left circumflex coronary artery and 50% for the right coronary artery.
In a more recent study, Giesler et al. [16] found that the optimal phase was determined more by the patient's heart rate than by the coronary vessel studied. The best imaging window in patients with a heart rate of 70 beats/min or less was usually during mid to late diastole, whereas the optimal cardiac phase to visualize the coronary arteries in patients with a heart rate of more than 70 beats/min was more frequently during late systole and early diastole. These findings were confirmed in our study and are explained by the shorter duration of diastolic relaxation in patients with high heart rates. On the other hand, we did not find a significant effect of heart rate on the scores for absolute image quality or on interscan variability. This finding may be related to the use of the heart rate–adaptive image-reconstruction algorithm in this study. With use of this recently developed algorithm, effective temporal resolution is 250 msec for heart rates less than 70 beats/min and 125 msec for heart rates of 70 beats/min or faster [18]. The better temporal resolution in patients with high heart rates may partially compensate for the problems related to a shorter diastolic window in these patients. These results may have practical implications because the use of negative inotropic drugs may not be mandatory in patients undergoing coronary artery calcium measurements using helical CT.
Study Limitations
This study has several limitations. First, a relatively small number of
patients (n = 50) were studied. Second, images were reconstructed at
only three reconstruction intervals (40%, 50%, and 60% of the R-R interval).
We selected these time points on the basis of the results of previous studies
[14,
15,
17]. Theoretically, better
results could have been obtained using image-reconstruction intervals that
were higher (70%) or lower (30%). This seems unlikely, however, when the
motion patterns of the different coronary vessels are considered
[14,
15,
17]. Also, the mean image
quality obtained with the optimal image-reconstruction window in our patients
was very high. In other words, no significant further improvement in image
quality could have been obtained, irrespective of the number of different
image-reconstruction window values used. A third limitation is that only axial
images were used to assess image quality. Different results could
theoretically have been obtained if reformatted or three-dimensional images
had been assessed.
In conclusion, our results suggest that coronary motion remains the most important problem in coronary calcium quantification using helical CT, even using MDCT with subsecond rotation time and retrospective ECG-gating. Ritchie et al. [28] and Wang et al. [29] suggested that image-acquisition times should be less than 50–100 msec to obtain artifact-free images of the coronary arteries. With the MDCT technique used in this study, rotation times of 500 msec are available, and effective temporal resolution can be 125–250 msec, depending on the heart rate. The introduction of 16- and 32-detector row MDCT will lead to a further improvement in effective temporal resolution and a decrease in artifact caused by coronary motion. Whereas current results might be optimized by reconstructing multiple data sets obtained at variable image-reconstruction windows and comparing the resulting axial and three-dimensional images, it is likely that the use of a next-generation CT device will make cardiac CT a less time-consuming and more robust technique.
In our study, MDCT was used, and the reproducibility of coronary artery calcium measurements was assessed. Our results suggest that coronary artery motion remains an important problem and that interscan reproducibility is less than optimal, even with individual optimization of image-reconstruction windows. Therefore, this technique may have limited usefulness for monitoring of disease progression unless a long time interval is used for follow-up.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  N. Matsuura, J. Horiguchi, H. Yamamoto, N. Hirai, T. Tonda, N. Kohno, and K. Ito Optimal Cardiac Phase for Coronary Artery Calcium Scoring on Single-Source 64-MDCT Scanner: Least Interscan Variability and Least Motion Artifacts Am. J. Roentgenol., June 1, 2008; 190(6): 1561 - 1568. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. J. Budoff, S. Achenbach, R. S. Blumenthal, J. J. Carr, J. G. Goldin, P. Greenland, A. D. Guerci, J. A.C. Lima, D. J. Rader, G. D. Rubin, et al. Assessment of Coronary Artery Disease by Cardiac Computed Tomography: A Scientific Statement From the American Heart Association Committee on Cardiovascular Imaging and Intervention, Council on Cardiovascular Radiology and Intervention, and Committee on Cardiac Imaging, Council on Clinical Cardiology Circulation, October 17, 2006; 114(16): 1761 - 1791. [Full Text] [PDF]
|
|
|
|
|
|
J. Horiguchi, H. Yamamoto, N. Hirai, Y. Akiyama, C. Fujioka, K. Marukawa, H. Fukuda, and K. Ito Variability of repeated coronary artery calcium measurements on low-dose ECG-gated 16-MDCT. Am. J. Roentgenol., July 1, 2006; 187(1): W1 - W6. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
