| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Imaging, Cedars-Sinai Medical Center, 8700 Beverly Blvd., Rm.
1258, Los Angeles, CA 90048.
2 Tennessee Heart and Vascular Institute and EBT Research Foundation, Nashville,
TN.
3 Department of Medicine, Division of Cardiology, University of California, Los
Angeles, CA.
4 Heart Disease Prevention Program, University of California, Irvine, CA.
Received February 3, 2005;
accepted after revision June 7, 2005.
Address correspondence to D. S. Berman
(bermand{at}cshs.org).
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. A computer algorithm was developed to automatically find contiguous lipid-rich lesions with voxel intensities below a calculated patient-specific lipid threshold. Lipid density and lipid inhomogeneity in Hounsfield units were calculated in the proximal left coronaries of three populations: 34 low-risk patients (low-risk group < 6% Framingham risk score, no calcium), 31 high-risk patients (high-risk group > 20% Framingham risk score, no calcium), and 37 patients with calcified plaque (calcium group).
RESULTS. The mean lipid density was -19.6 ± 3.0 (SD) H in the low-risk group, -25.3 ± 8.2 H in the high-risk group, and -34.3 ± 13.0 H in the calcium group (p < 0.05). The mean lipid inhomogeneity was 17.7 ± 3.6 H in the low-risk group, 21.5 ± 5.5 H in the high-risk group, and 29.0 ± 7.6 H in the calcium group (p < 0.05). The mean interscan variability in lipid density and lipid inhomogeneity were 2.0 ± 3.3 H and 2.1 ± 3.6 H, respectively. In five patients, the locations of lipid-rich plaque correlated well with available intravascular sonography findings.
CONCLUSION. Our method may be able to identify lipid-rich plaque on noncontrast cardiac CT.
Keywords: cardiac CT • cardiac imaging • cardiovascular disease • CT • electron beam tomography
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Cardiac CT with either MDCT or electron beam tomography (EBT) has emerged as noninvasive imaging techniques that are highly sensitive for the assessment of coronary calcium [7-9], which has proven to be of diagnostic and prognostic importance. Although the extent of coronary artery calcification is closely related to the extent of coronary atherosclerosis [10], the presence of calcium has not been found to be useful in identifying unstable plaques [5, 11, 12]. The presence of a large amount of lipid in the plaque, however, appears to increase its biomechanical stresses and the risk of rupture [6, 11]. Extending the use of EBT to evaluate lipid-laden plaque would enhance its diagnostic utility through the identification of atherosclerotic lesions that may be reversible [13, 14] and could potentially help identify vulnerable patients. Because lipid is less dense than water, blood, or coronary calcifications [15], identification of lipid-laden plaque is feasible using EBT or MDCT methods. There is growing evidence that low-attenuation regions in the coronary arteries may indicate the presence of lipid-rich plaque on both contrast [16-19] and noncontrast [20, 21] cardiac CT.
Our objective in this study was to develop computer software to derive added information regarding lipid-rich plaque from noncontrast cardiac CT scans acquired for routine calcium scoring and to evaluate its feasibility in patients undergoing CT for the assessment of coronary calcium. Our automated software (Plaquant) identifies lipid-rich regions on noncontrast cardiac CT or EBT images using a patient-specific biologic lipid attenuation threshold that is tailored to and derived from the individual CT or EBT scan.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Imaging Protocol
Each complete scan contained 50-60 contiguous (nonoverlapping) 512 x
512 matrix slices over a 35-cm field of view. The pixel size was 0.68 x
0.68 mm. For each patient, the scan was obtained in a single breath-hold and
extended from the aortic arch to the level of the diaphragm. Depending on the
patient's heart rate, ECG triggering was set to 45-60% of the R-R interval.
For EBT, images were obtained with a 100-msec exposure time and 3-mm-thick
slices. For MDCT, 120 kVp was used
[22] and the slice thickness
was 2.5 mm. MDCT scans were acquired with prospective ECG-gating. Each scanner
was calibrated daily using both air and water phantoms.
Coronary Calcium Scoring
Cardiac CT and EBT images were reviewed by an imaging cardiologist. Each
scan was analyzed using semiautomated commercially available calcium scoring
software (ScImage, ScImage). The total coronary calcium score using the
Agatston method [7] for each
scan was measured as the sum of the plaque scores of each coronary artery.
Patients
Patients were selected retrospectively. First, patients were distinguished
on the basis of their coronary calcium score
[7] being zero or greater.
Patients with a nonzero coronary calcium score were defined as the calcium
group. Patients with a zero coronary calcium score were further sorted into
low-risk and high-risk groups on the basis of their 10-year Framingham risk
score [23]. We analyzed three
patient groups: first, the low-risk group, which was composed of 34
consecutive patients with a < 6% Framingham risk and no calcium
[24]; second, the high-risk
group, which was composed of 31 consecutive patients with > 20% Framingham
risk and no calcium [24]; and,
third, the calcium group, which was composed of 37 consecutive patients with a
nonzero coronary calcium score. Table
1 shows the characteristics of our patient population.
|
Coronary Catheterization
Eight patients from the calcium group underwent coronary catheterization
within 1 month of the cardiac CT examination. Of these patients, five patients
underwent intravascular sonography. Two expert observers visually reviewed the
coronary angiography and intravascular sonography studies.
Computer-Aided Analysis
Lipid is less dense than water and typically has attenuation values ranging
from -30 to -190 H [15,
25]. Blood has an attenuation
value of approximately 40 H
[15], and coronary
calcifications are usually quantified from a fixed threshold value of 130 H
[7,
9,
15]. The attenuation values of
the normal blood pool are typically peaked symmetrically at about 40-50 H. For
the coronary arteries, if lipid-laden voxels exist, then the attenuation
values extend to negative values corresponding to lipid. In a previous study
by Teichholz et al. [20], a
fixed lipid threshold of 0 H was used
[20]. However, large
interindividual variations in attenuation values have been reported for soft
tissue and fat [26,
27]. Because of such
variations, some researchers have suggested that the use a fixed-attenuation
threshold for all scans is not appropriate and that a threshold tailored to
the individual scan should be used
[26,
27]. In our algorithm, a
patient-specific biologic lipid threshold is calculated from each patient
scan.
All studies were exported to DICOM format and transferred to a stand-alone Windows (Microsoft) workstation. Using our software (Plaquant), a trained observer reviewed the images to select the transverse slices in which the proximal left main (LM) and left anterior descending (LAD) arteries were best represented. Regions of interest (ROIs) were drawn by the trained observer to identify a normal blood pool region in the ascending aorta (a uniform region in the center of the aorta away from any calcifications), LM artery, and proximal one third of the LAD artery. The normal blood pool regions were similar in size for all patient scans. The ROIs in the coronary arteries were placed as follows: in every transverse slice, one ROI was drawn around the entire identifiable portion of the LM and proximal LAD arteries seen in that slice. Finally, each ROI was visually verified to be in the coronary artery by displaying the area in a zoomed fashion in transverse, coronal, and sagittal views.
We automatically identified lipid-rich areas in the coronary arteries by analyzing the attenuation on image histograms corresponding to the drawn ROIs. Figure 1 shows a schematic diagram of attenuation on image histograms corresponding to a normal blood pool ROI and an ROI drawn on a coronary artery with lipid-rich content. In our algorithm, each image histogram was first smoothed by a 5-point convolution kernel. A gaussian function was iteratively fitted to the smoothed normal blood pool histogram using the Levenberg-Marquardt minimization algorithm [28]. The intersection of the image histogram for each drawn coronary artery ROI with the normal blood pool-fitted curve was calculated. The patient-specific lipid threshold was defined as the mean intersection of all the coronary artery image histograms with the normal blood pool-fitted curve (Fig. 1). This calculated lipid threshold is specific to each cardiac CT scan.
|
For all lipid-rich lesions, we compared three parameters: lipid density, lipid inhomogeneity, and lipid minimum. We defined lipid density as the average voxel value in Hounsfield units below the lipid threshold for all voxels in all identified lipid-rich lesions. We defined lipid inhomogeneity as the SD in Hounsfield units for all voxels in all identified lipid-rich lesions. Lipid minimum was defined as the minimum value in Hounsfield units for all voxels in all identified lipid-rich lesions.
Statistical Analysis
We compared the lipid density, lipid inhomogeneity, and lipid minimum
values for the low-risk, high-risk, and calcium patient groups using the
one-way analysis of variance. Pairwise group comparisons were done using the
Student's t test. A p value of less than 0.05 was considered
to be statistically significant.
Reproducibility of Measurements
To investigate the interscan reproducibility of our method, we rescanned 25
patients from the low-risk, high-risk, and calcium patient groups using the
same scanner and the exact same scanning parameters within 24 hours of the
initial scan. For each patient, both sets of acquired data were analyzed using
our program, with the same observer drawing the ROIs. For each patient, we
compared the lipid density, lipid inhomogeneity, and lipid minimum values from
the two data sets.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
|
|
|
Lipid density and lipid inhomogeneity values for the coronary arteries in the three patient groups are shown in Figures 6 and 7. The mean lipid minimum values for the low-risk, high-risk, and calcium groups were -54.2 ± 11.2 H, -70.4 ± 24.9 H, and -106.4 ± 38.9 H, respectively (p < 0.01 across groups). Lower lipid density indicates lower average density, and higher lipid inhomogeneity indicates more inhomogeneity within the lesions found. The differences in lipid density, lipid inhomogeneity, and lipid minimum values were significant for all three patient groups (p < 0.05). There was a progressive decrease in lipid density and a progressive increase in lipid inhomogeneity across the low-risk, high-risk, and calcium groups. Our results therefore may indicate progressively increasing lipid-rich inhomogeneous content across the low-risk, high-risk, and calcium groups.
|
|
Reproducibility of Measurements
A summary of our interscan measurement reproducibility results in the
coronary arteries for 25 patients is shown in
Table 2. The mean absolute
difference for lipid density, lipid inhomogeneity, and lipid minimum was less
than 3 H. In comparison, the mean absolute differences for lipid density,
lipid inhomogeneity, and lipid minimum between our patients with abnormal
findings and those with normal findings were 27.4 ± 11.7 H, 25.2
± 8.8 H, and 83.9 ± 37.1 H, respectively. Our measured
variations in lipid density, lipid inhomogeneity, and lipid minimum between
scan 1 and scan 2 are, therefore, small (10%) compared with variations between
"normal" and "abnormal" coronary arteries.
|
We also evaluated a lipid threshold given by the full-width tenth maximum of the normal blood pool gaussian curve. The results were very similar to those presented above.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CT evidence of the greatest amount of lipid-rich plaque was seen in the calcium group. Although this may represent greater lipid-rich plaque in these patients, phantom studies indicate that there can be beam-hardening artifacts that introduce low-attenuation areas around areas with dense calcium [15] that may mimic lipid-rich lesions. Further study of patients with a nonzero coronary calcium score would be helpful in determining whether our method can distinguish true lipid-rich plaque from beam-hardening artifact. Teichholz et al. [20] analyzed areas between calcified lesions (skip areas) on EBT scans that showed coronary calcifications, which might have been particularly prone to calcium-related artifacts. In our study, we analyzed the entire coronary artery as seen in the transverse slice, not just the skip areas or areas immediately adjacent to calcific regions (Figs. 2, 3, 4, and 5), thereby lessening the possibility that our findings are related only to beam-hardening artifact.
We developed semiautomated software to identify lipid-rich lesions in the coronary arteries on noncontrast cardiac EBT and MDCT images. The preliminary comparison shows that our method is reproducible. In our study, it was necessary to draw separate ROIs on each of the two serial scans, which can cause additional variation in the results. Budoff et al. [29] found that the interscan variation of blood pool values as measured on EBT was 3.47 H, similar to our reported absolute difference values in Table 2. The mean interscan variability in the Agatston score [7] and the volumetric score [9] was reported to be 23% and 21% by Ohnesorge et al. [30] and 21.6% and 17.8% by Lu et al. [31], respectively. With our method, each cardiac CT has its own reference normal region. Hence, cardiac CT scans from three separate scanners could be analyzed using the same method requiring no normal limits.
Although further validation of our method using gold standard intravascular sonography is necessary, our results suggest that it may be possible to identify lipid-rich plaque on noncontrast cardiac CT, which could be important for the early detection of lipid-laden plaque. Noncontrast CT for assessment of coronary calcium is now widely applied in clinical practice [8] and the potential to further identify lipid-laden plaque by computer-aided analysis would enhance the clinical utility of this approach.
Other groups have investigated lipid-rich plaque detection with cardiac CT. Baumgart et al. [21] compared visual assessment of patients with noncontrast EBT, coronary angiography, and intravascular sonography scans. Using intravascular sonography as the gold standard to classify plaques, they reported that noncontrast EBT had a high sensitivity (97%) and high specificity (80%) to detect calcified plaques and a lower sensitivity (47%) but equivalent specificity (75%) to detect noncalcified plaques [21]. Schroeder et al. [16] evaluated the accuracy in detecting plaque configuration using contrast MDCT by correlating attenuation values from coronary lesions shown on CT with those shown on intravascular sonography images; they found that contrast CT can distinguish lipid-rich, fibrous, and calcified plaque. Achenbach et al. [18] recently assessed the accuracy of contrast MDCT to detect atherosclerotic plaque in nonstenotic coronary arteries, using intravascular sonography as the gold standard. They reported that overall sensitivity and specificity were high (92% and 88%, respectively) for proximal artery segments [18]. Teichholz et al. [20] analyzed noncontrast EBT scans of patients with and those without coronary calcium. Their study lacked gold standard intravascular sonography validation. By simple ROI analysis, they showed that the mean voxel values for the patient group with coronary calcium were significantly lower than those in the group with no coronary calcium. In their study, a fixed lipid threshold of 0 H was used for all patients.
Because there are large inter- and intraindividual variations in EBT attenuation values, it has been recommended that a biologic threshold tailored to the individual scan be used [26]. In all these studies, either visual or simple ROI mean and SDs were used, which can be time-consuming and are subject to high intraobserver and intrascan variations. In this work, which differs from all previous studies, we implemented a fast computer-aided analysis that uses a patient-specific biologic lipid threshold, thus providing a more standardized way for assessing lipid in the coronary arteries from noncontrast cardiac CT.
Study Limitations
Partial volume effect—The coronary arteries are narrow
curvilinear structures surrounded by epicardial fat. Because the slice
thickness is 2.5-3.0 mm, partial volume effects tend to reduce voxel
intensities, mimicking lipid-rich lesions. However, it is unlikely that the
partial volume effects would be different between the three groups and would
influence the differences between the groups. To reduce the effects of partial
volume on our results, each ROI was visually verified to be within the
coronary artery by displaying the area on zoomed transverse, coronal, and
sagittal views. Importantly, the objective of this study was to develop the
software program and then to evaluate only its feasibility in the clinical
setting with EBT and 4-MDCT. As the field of cardiac CT evolves to isotropic
resolution with voxel dimensions of 0.4 mm, the potential of this method,
which would not need significant modification to handle thin-slice image data,
may be greatly enhanced.
Identification of coronary arteries—It is often difficult to distinguish coronary arteries from the epicardial fat that surrounds them. In an attempt to assess the impact of coronary artery fat, we also studied the aorta, which has little surrounding fat. The fact that our findings were similar in the aorta and in the coronary arteries suggests that our observations are related to lipid-rich plaque and not to epicardial fat.
Lack of a gold standard—Our study lacks validation with a standard with which our findings could be compared, such as intravascular sonography or MRI [32]. In our study, only five patients had both cardiac CT scans and intravascular sonography.
Better noise estimation—In our method, we excluded lipid-rich lesions that were less than 3 voxels. If other standards for plaque classification were available, a patient-specific limiting cluster size could be validated.
Plaque assessment—From our preliminary study, we found that our method can be used to identify lipid on EBT images, but intermediate fibrous plaques cannot be distinguished from blood and tissue. However, because lipid-rich plaques can potentially pose more risk to the patient [4, 6], early noninvasive detection and evaluation of such plaques are important clinical goals [13, 14].
In conclusion, we developed a computer-aided method with which to identify lipid-rich lesions on noncontrast cardiac CT using a patient-specific biologic lipid threshold. Despite several limitations, our findings suggest that the method may have the potential to identify lipid-rich plaque on noncontrast cardiac CT images.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  P. M. Colletti Cardiac Imaging 2006 Am. J. Roentgenol., June 1, 2006; 186(6_Supplement_2): S337 - S340. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
