| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Radiology, Beth Israel Deaconess Medical Center and Harvard
Medical School, 330 Brookline Ave., WCC-302, Boston, MA 02215.
2 Department of Cardiology, Beth Israel Deaconess Medical Center and Harvard
Medical School, Boston, MA.
Received October 24, 2006;
accepted after revision January 2, 2008.
Address correspondence to M. E. Clouse
(mclouse{at}bidmc.harvard.edu).
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. Forty-one normal and eight abnormal arterial cross sections with noncalcified plaque selected from 40 patients undergoing MDCTA were analyzed for percentage of stenosis and plaque volume using a voxel analysis technique.
RESULTS. Using voxel analysis, the normal arterial wall thickness was determined to be 0.8 ± 0.4 mm. Attenuation values (in Hounsfield units) for normal segments ranged between 30 and 175 H and for abnormal (plaque-containing) segments ranged from -49 to 139 H (p < 0.05). Plaque volume measurements varied from 0.90 to 156 mm3 with good interobserver correlation (R2 = 0.9671). Percentage of stenosis correlated with quantitative coronary arteriography measurement (R2 = 0.55). Voxel analysis underestimated the percentage of stenosis (Pearson's correlation coefficient, 1.2; p = 0.03).
CONCLUSION. The study shows that the voxel analysis technique appears to be an accurate and reproducible method to measure arterial wall thickness, noncalcified plaque, and degree of arterial stenosis using density values measured in Hounsfield units. The technique may be useful on further correlative studies.
Keywords: cardiac imaging • coronary artery wall • coronary imaging • CT angiography • MDCT • micro-CT • voxel analysis
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The latest MDCT scanners have made possible the in vivo detection of noncalcified plaque [14-20]. However, accurate and reproducible measurement of atherosclerotic plaque using MDCT angiography (MDCTA) is difficult and time-consuming because of technical and physiologic factors such as arterial size and cardiac motion. By dividing minimal detectable motion by peak physiologic velocity, Ritchie et al. [21] determined that a scanning time of 19.1 milliseconds is necessary to avoid motion artifacts; and Achenbach et al. [22] determined the lowest velocity of all coronary arteries to be 27.9 mm/s at 48% of the cardiac cycle in late systole and early diastole before atrial contraction. For the foreseeable future, technologic limitations will impose a certain degree of edge unsharpness of the arteries due to motion.
Historically, accurate and reproducible measurement of plaque has been limited by the need to manually draw lines separating epicardial fat from the outer arterial wall, plaque, and lumen interfaces because even a slight pixel shift causes marked change in density as measured in Hounsfield units. This may be partially overcome by computer automation. Other contributing factors include partial volume averaging, detector size, slice thickness, and difference in display settings. One study reported significant variability of lumen diameter and plaque measurements using different display settings for window and gray-scale threshold levels when measuring coronary arteriograms [23]. Another empirically selected a 500-H window and 150-H threshold level using MDCT and slightly over estimated the plaque area compared with intravascular sonography [24]. Funabashi et al. [25] reported settings related to peak attenuation in vessel lumen as superior to fixed settings, and Leber et al. [26] advanced the technique significantly by manipulating the window settings so that the MDCT image equals the intra vascular sonography image sized to the external elastic membrane to separate the surrounding tissue from the lumen and plaque, although there was no discussion of the adventitial component to the wall. However, they noted the difficulty of separating the outer wall boundary from the surrounding tissue.
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
The optimal single-plane cine projection was selected that identified the stenosis in its greatest severity with minimal foreshortening or overlapping of branches, and end-diastolic frames were chosen for quantitative angiographic analysis using a previously described and validated automated edge detection algorithm [28].
This study was approved by our institutional review board and all subjects provided written informed consent. Patients with unstable angina, contraindications to contrast material, serum creatinine level > 1.7 mg/dL, liver disease, prior coronary stents, bypass grafts, pacemakers, or arrhythmias were excluded. Caffeine was not allowed after 12:00 midnight, and nasal oxygen was administered to help in the breath-hold. IV Lopressor (metoprolol, Novartis Pharma) in 5-mg titrated doses (not to exceed 15 mg total) was given for heart rates > 70 beats per minute (bpm) to reduce the heart rate below 65 bpm for two- or three-segment reconstruction.
MDCT Protocol
MDCTA data were obtained using an Aquilion 16 scanner (Toshiba America
Medical Systems) with an X-ray tube focal spot size, 1.4 x 1.6 mm; slice
thickness, 0.5 mm with 0.1-mm overlap; gantry rotation, 400 milliseconds; 135
kVp; 350 mA; retrospective ECG gating; and a pitch factor of 0.2-0.3
autoselected to heart rate, for a total scanning time of 16-24 seconds. Field
of view was 32 cm; image matrix, 512 x 512; and pixel size, 0.39
mm2. Image acquisition was triggered at 180 H above the left main
coronary artery orifice using automated bolus tracking; 90 mL of ioversol
(Optiray 350, Tyco Healthcare) followed by 30 mL of a saline chaser was
injected at 4 mL/s. The radiation dose was 11.4 mSv, as calculated using
IMPACT software (IMPACT Scan).
Image Reconstruction
Image reconstruction (retrospective gating) was half (single segment) for
rates < 50 bpm with temporal resolution of 230 milliseconds (180° of
projection data plus 49.5° of fan angle) and multisegment (2-3 segments)
reconstruction for rates between 55 and 64 bpm, autoselected by the scanner
and varying between 163 milliseconds for rates of 55 bpm and 142 milliseconds
for rates of 64 bpm. The images were reconstructed with a slice thickness of
0.4 mm (resolution of scanner) using a medium sharp convolution kernel at 10%
intervals from 10% to 90% of the R wave and transferred to a 3D workstation
(Vitrea 2, version 3.5, Vital Images). DICOM images from the late diastolic
phase of the cardiac cycle at 65% to 75% or 80% of the R-R interval and in the
range of 680 to 700 or 720 milliseconds to select images with least motion
were reconstructed for plaque measurement. These images were reviewed, and
only noncalcified segments without artifacts (grade 1, excellent, no motion)
were selected by two observers independent of one another. Segments rated
grade 2, good, with minor motion artifacts; grade 3, fair, with moderate
motion artifacts; and grades 4 and 5, major blurring, luminal assessment of
questionable accuracy and heavily calcified, were not considered
[29,
30]. Plaque with any calcium
was not considered because of the possible error from blooming artifacts. The
MDCT images were evaluated using axial, multiplanar reformatted, and
maximum-intensity-projection (MIP) reconstructions using a combination of
Hounsfield unit window and threshold levels. Percentage of stenosis was
calculated by selecting the lumen at greatest narrowing compared with the
diameter of the normal vessel, as with QCA
[31]. The images were then
transferred to an Analyze 6 workstation (Analyze Direct) for wall, plaque
volume, and percentage of stenosis measurements.
|
|
|
|
Two reference lines were then drawn across the long axis in the normal lumen above and below the plaque to establish a reference to measure luminal narrowing and intraluminal and plaque volumes. A third line was then drawn across the segment with the largest amount of plaque to determine the percentage of stenosis (Figs. 1A, 1B, 2A, 2B, 2C and 2D). Voxel analysis was then performed by plotting eight radial lines at 45° intervals across the arterial wall from the epicardial fat to the center of the lumen (Figs. 1A and 1B). Seven voxels (A-G) were usually sufficient to be certain that that line had crossed the wall from epicardial fat to the lumen; however, in larger (left main and left anterior descending arteries) or eccentric plaques the voxel line was extended to H, I, and J. Attenuation values (in Hounsfield units) of these seven consecutive isotropic (400-µ) voxels (A-G) were measured along each one of these eight radial lines from outside the arterial wall into the lumen (attenuation of each voxel sampled eight times) to normalize the attenuation densities and to determine the density and thickness of the normal arterial wall; 400-µ was chosen because this is the resolution of the scanner. Two voxels, A and B (attenuation, -30 to -50 H ± 15-20 H [SD]), were always placed in epicardial fat to establish a baseline reading. Voxel C (density near 0 H), was always placed at the interface of the epicardial fat and arterial wall. Voxels D and E spanned the arterial wall, and F and G were within the lumen. A total of 2,296 voxels across 328 lines in the normal and 448 voxels across 64 lines in the abnormal (noncalcified plaque) segment were used in this analysis to normalize the values for thickness of normal wall and plaque density. Volume averages for the two arterial wall interfaces were calculated for the outer and inner walls, resulting in the value for wall thickness (Fig. 3). Based on the normalized voxel values, an attenuation plot was made by selecting a section across the normal arterial wall above and below the plaque area that establishes the length of the segment. Voxel C was placed at the interface of normal arterial wall and epicardial fat, and a second voxel (F) was placed at the inner wall-lumen interface in each location above and below the plaque of the 3D segment. These thresholds (attenuation values in Hounsfield units) established the arterial wall composed of two voxels (D and E). The computer then generated a 3D volume, and a function was used to remove the wall containing two voxels, leaving the luminal volume. The computer then added and subtract ed the volume of the attenuation values representing lumen, leaving the densities (in Hounsfield units) of the plaque volume. The percentage of stenosis is derived from luminal voxels in the line representing the largest diameter of plaque.
|
Statistical Analysis
All continuous variables were expressed as their means and SDs. The
Student's t test and analysis of variance were used to evaluate
differences in the mean attenuation value of voxels. Pearson's correlation was
used for differences in plaque volumes calculated by both observers, with
p < 0.05 considered significant. All statistical analyses were
performed on commercially available statistical software (SPSS Inc).
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Abnormal arterial wall cross sections—In sections with noncalcified plaque, corresponding attenuation values were -49 ± -16 H, -30 ± 14 H, 2 ± 12 H, 43 ± 23 H, 83 ± 41 H, 115 ± 54 H, and 139 ± 61 H, for voxels A-G, respectively. The mean attenuation values of voxels E-G were significantly lower (p < 0.05) than their counter parts in the normal sections, indicating the presence of lower-density material (noncalcified plaque) compared with higher-density material (contrast medium and blood) in the normal cross sections (Fig. 3). In arterial segments with a large eccentric plaque, the voxel profile was extended beyond voxel G to reach the center of the contrast-filled lumen (Fig. 2D).
Plaque Volume and Luminal Diameter Calculation
Plaque volume and luminal diameter were calculated using the voxel method
by two experienced observers working independently of one another. In normal
segments, 30 ± 25 H was the average of voxels C and D and therefore
represented the interface between epicardial fat and outer wall; 175 ±
55 H was the average of voxels E and F and represented the interface between
the inner arterial wall and lumen. These attenuation values were based on the
average calculations for the interfaces of the outer and inner wall boundaries
to compensate for partial volume effect. Using these values (30 and 175 H), we
measured average wall thickness for normal vessels to be 2 ± 1 voxels,
corresponding to 0.8 ± 0.4 mm. The wall and luminal volume were then
subtracted from the arterial segment, leaving the plaque volume. The plaque
volume calculation is automatically performed by the computer summing the
attenuation densities of the plaque (voxels) in the cross section and down the
length of the plaque for volume. The two observers measured the plaque volume
independently with excellent correlation (R2 = 0.9671).
Plaque volume from each of the eight segments varied from 0.90 to 156
mm3 (Fig. 4). The
percentage of stenosis as measured by voxel analysis correlated with that
measured by QCA (R2 = 0.55, p = 0.04), although
the voxel analysis method was found to underestimate the percentage of
stenosis (Pearson's correlation coefficient, 1.2; p = 0.03)
(Fig. 5). The stenosis values
by QCA ranged between 40% and 70%, and those by voxel analysis were 39-56%. By
comparison, the percentage of stenosis measurements on MDCT did not correlate
as strongly with those done by QCA (R2 = 0.19, p
= 0.3).
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The importance of quantifying plaque is well known because total plaque burden is considered the most important predictor of coronary events [33], and the rupture of noncalcified plaque has been implicated as the cause of acute coronary syndromes [1-4]—hence, the intense interest in using MDCTA to quantify plaque volume [14-17]. Inoue et al. [4] have suggested MDCTA may be useful in differentiating stable versus unstable plaque because of its ability to differentiate plaque density. Estes et al. [34] first showed the ability of CT to differentiate lipid from fibrotic plaque: 39 ± 12 H and 90 ± 24 H, respectively (p < 0.001). Others using histology and intravascular sonography have validated this original observation, and meta-analysis shows the range for lipid plaque to be 14-50 ± 26 H, fibrous plaque to be 80-104 ± 28 H, and calcified plaque to be > 391 H [14-16, 35]. Leber et al. [19] compared intravascular sonography with MDCTA and detected 78% of noncalcified (hypoechoic) plaque, 78% of plaque containing hyperechoic areas, and 95% of calcified plaque. The undetectable plaque had a maximum thickness of 0.9 ± 0.3 mm versus 1.5 ± 0.3 mm for detectable plaque and was located in coronary sections with a smaller external elastic membrane diameter (3.6 ± 1.1 mm vs 4.5 ± 1.2 mm; p < 0.02). In addition, the percentage of the cross-sectional area of the plaque was smaller, 22% ± 5% versus 42% ± 16% [17]. By manipulating the display setting, Leber et al. [26] were able to compare the results between MDCTA and intravascular sonography, and detected 83% of noncalcified plaque, 70% of plaque with a lipid pool, and 90% of plaque with spotty calcification. Although interobserver agreement for identifying atherosclerotic sections was good, interobserver variability for measuring plaque volume was 37%. Achenbach et al. [18], comparing intravascular sonography with MDCT, also detected plaque in the large coronary segments, with a reported sensitivity of 78% and specificity of 87%. False-negatives resulted when the plaque volume and area were small (47 ± 11 mm3 and 8 ± 3 mm2) compared with those in true-positive segments (76 ± 10 mm3 and 11 ± 4 mm2) (p = 0.2 and p = 0.08, respectively).
Another reason for discrepancy in comparing MDCTA with intravascular
sonography for plaque measurement may be that the two methods are not always
measuring the same arterial components, especially in advanced disease.
Intravascular sonography measures intraluminal plaque and media, whereas MDCT,
as with other body techniques, measures all components of the wall, including
the adventitia. Intravascular sonography accurately estimates the amount of
intraluminal plaque but may underestimate the total amount of disease when a
large adventitial component is present; MDCT seems to overestimate the amount
of plaque compared with intravascular sonography by accurately estimating the
entire atherosclerotic process, yet MDCT cannot separate each component of the
process. Using high-frequency epicardial echocardiography, Gradus-Pizlo et al.
[12] showed the importance of
including the adventitia in measurements of plaque and arterial wall thickness
by body imaging methods. Their work showed that adventitia represented
approximately 54% of the wall thickness in normal and 51% in abnormal vessels,
with normal wall thickness being 1.0 ± 0.2 mm and abnormal being 1.8
± 0.2 mm. Their findings of the normal arterial wall thickness are in
agreement with the intravascular sonography work of others
[36-38].
Fayad et al. [13], using
high-resolution MRI, reported similar findings of coronary wall thickness
(0.75 ± 0.17 mm in normal and 4.38 ± 0.71 mm in atherosclerotic
vessels) with 
40% stenosis by arteriography but could not differentiate
separate components [13]. This
is not unexpected because the resolution of high-frequency epicardial
echocardiography is much greater than that of MDCT and MRI.
To overcome many of the problems relating to accuracy and reproducibility of measuring atherosclerotic plaque and to standardize the process, we used the voxel analysis technique of plotting the density (in Hounsfield units) of 0.4-mm voxels across the entire isolated arterial segments in normal and atherosclerotic vessels. Plotting the line from outside the arterial wall rather than from the lumen center outward allowed us to always place voxel C, the interface of the outer wall and epicardial fat, at or near 0 H to define the boundary of the outer wall and to separate it from the surrounding tissue. With volume averaging, density of the outer wall (voxel D) was established at 30 ± 25 H and of the inner wall (voxel E), at 175 ± 55 H. By placing the voxels at the interface of the outer wall and plotting inward, we were able to determine the normal wall thickness to be two 0.4-mm voxels (0.8 mm). Initially, the measurement of density of seven voxels (A-G) arranged in a radial fashion around the arterial wall (each measured eight times) was sufficient to show that the wall was free of disease. The number of measurements was adequate to normalize the values across the arterial wall, plaque, and lumen. In larger arteries and those with eccentric plaque, the voxels extended to the lumen center, depicted as voxels H, I, and J, when necessary (Figs. 1A, 1B and 2A, 2B, 2C, 2D). Our focus was on larger arteries, which have the greatest locus of disease, known wall thickness, and diameter, to perform a pilot study [36-39]. However, wall thickness is less in smaller-caliber arteries; there have been no reports of measurements in these arteries using body imaging techniques [36-39].
The voxel analysis method of measuring and removing voxels was verified by creating a stationary phantom whose wall and lumen diameter were first measured using micrometry and micro-CT (100-µ voxels) [32]. Using these data, we then used MDCT (400-µ voxels) for comparison with the micrometry measurements.
In the phantom study, three intraluminal polyethylene particles of different sizes were measured using micro-CT and MDCT, and both techniques were accurate relative to true volume measurements. The overall accuracy error was 2.05% (range, -10.44% to 4.00%) less than true volume for the three averaged polyethylene particles for both scanners. Excellent correlation of volume measurements was also seen between intra- and interreader reliabilities for both micro-CT (concordance correlation coefficient = 0.999) and MDCT (concordance correlation coefficient = 0.972) [32]. Although individual accuracy errors vary, micro-CT did reach significance, and MDCT did not. We believe this is related to using large (i.e., 400-µ) voxels. The 400-µ voxels can create significant errors when small plaque volumes are being measured—that is, smaller plaque may be missed or even (more likely) overestimated. Hence, we believe that using 100-µ vessels will reduce accuracy errors.
The measurement of intraluminal plaque volume using MDCTA in the clinical studies by two readers independently of one another also showed excellent correlation, indicating reproducibility. Plaque volume from each of the eight segments varied from 0.90 to 156 mm3 (Fig. 4). In our clinical studies the voxel measurements also showed significant correlation with QCA and tended to underestimate the percentage of stenosis (Fig. 5). Insignificant correlation of MDCTA percentage of stenosis with that on QCA was related to improper window and threshold settings in two of the eight arterial segment measurements.
Our study establishes that the arterial wall as assessed by MDCTA is approximately two voxels (0.8 mm), which is supported by the work of others using body imaging techniques [12, 13]. However, the densities of the inner wall and lumen interface may change with increasing or decreasing concentrations of contrast material in the blood because they depend on the iodine content of the contrast material, the injection rate, cardiac output, vessel size, and blood volume. Each investigator would need to normalize the attenuation values at this interface. However, regardless of the luminal density, there should always be a significant density change at this interface. Our injection rate was 4 mL/s. The ability of MDCT to separate the boundaries of the adventitia, arterial wall, and fibrous plaque is currently limited because of the overlap of density values [14-16, 34]. Future improvements in MDCT technology of shorter exposure times and higher spatial resolution, and comparison with intravascular sonography, may help to overcome these limitations and to separate intraluminal plaque volume and media from adventitia. A limitation of our study is the lack of interobserver variability data in plaque measurements, which can be corrected in further studies by obtaining kappa value measurements.
We have since learned that we no longer need to draw eight lines in a radial fashion across the normal or abnormal segment, but can establish equally accurate normalized values by drawing two reference lines perpendicular to the long axis above and below the plaque-containing segment to establish wall thickness and luminal diameter; and only eight measurements of voxels in each line across the wall are needed to establish outer and inner wall boundaries. A third line of voxels continues to be drawn across the largest plaque for measuring the percentage of stenosis (Figs. 1B and 2A, 2B, 2C, 2D).
Weaknesses of our study are the small size of the sample and that arteries smaller than 3 mm were not included (interreader variability only was assessed). For accuracy in future studies, the wall lumen interface attenuation values should be normalized for each patient because the intraluminal contrast attenuation values vary, especially with different injection rates and contrast concentrations. Further studies comparing histopathology with micro-CT and MDCT and incorporating intravascular sonography are needed. This knowledge could enable development of automatic postprocessing algorithms for more rapid and accurate calculation of plaque volume in order to make the voxel analysis method clinically useful.
In conclusion, this pilot study suggests that the voxel analysis technique is a reproducible method and may be used to quantify the atherosclerotic process (plaque and arterial wall, including adventitial thickening) in the diseased segment. If proven accurate with histopathologic studies and intravascular sonography, it may be helpful in following the effects of medical treatment to determine whether statins, lifestyle change, and exercise reduce plaque burden. However, software must be automated because the current manipulations and calculations are time-consuming.
Acknowledgments
We thank John V. Frangioni, director of the molecular imaging laboratory,
for providing access to the micro-CT scanner for the phantom studies.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
10 years
after successful percutaneous transluminal coronary angioplasty. Am
J Cardiol 1997; 79:1005
-1011[CrossRef][Medline]
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  H. Brodoefel, C. Burgstahler, A. Sabir, C.-S. Yam, F. Khosa, C. D. Claussen, and M. E. Clouse Coronary Plaque Quantification by Voxel Analysis: Dual-Source MDCT Angiography Versus Intravascular Sonography Am. J. Roentgenol., March 1, 2009; 192(3): W84 - W89. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
