In vitro atherosclerotic plaque and calcium quantitation by intravascular ultrasound and electron-beam computed tomography.

The purpose of this investigation was to compare the accuracy of intravascular ultrasound (IVUS) and electron-beam computed tomography (EBCT) in quantitating human atherosclerotic plaque and calcium. In experiment 1, 12 human atherosclerotic arterial segments were obtained at autopsy and imaged by using IVUS and EBCT. The plaque from each arterial segment was dissected and a volume measurement of the dissected plaque was obtained by water displacement. The plaque from each arterial segment was ashed at 700 degrees F, and the weight of the remaining ashes was used as an estimate of the calcium mass. In experiment II, 11 calcified arterial segments were obtained at autopsy and imaged by using IVUS at one site along the artery. A corresponding histologic cross section stained with Masson's trichrome was prepared. In experiment I, the mean plaque volume measured by water displacement was 165.3 +/- 118.4 microliters. The mean plaque volume calculated by IVUS was 166.1 +/- 114.4 microliters and correlated closely with that by water displacement (r = 0.98, p < 0.0001). The mean calcium mass measured by ashing was 19.4 +/- 15.8 mg. The mean calculated calcium mass by EBCT was 19.9 mg and correlated closely with that by ashing (r=0.98, p<0.001). The mean calculated calcium volume by IVUS was 18.6 +/- 11.2 microliters and correlated linearly with the calcium mass by ashing (r = 0.87, p < 0.0003). In experiment II, the mean cross-sectional area of the calcified matrix was 1.71 +/- 0.66 mm2 by histologic examination compared with 1.44 +/- 0.66 mm2 by IVUS. There was a good correlation between the calcified cross-sectional area by histologic examination and IVUS (r = 0.76, p < 0.007); however, IVUS may underestimate the amount of calcium present depending on the intralesional calcium morphologic characteristics. In conclusion, IVUS accurately quantitates atherosclerotic plaque volume as well as the cross-sectional area and volume of intralesional calcium, especially if the calcium is localized at the base of the plaque. IVUS underestimates the amount of calcium present because of signal drop-off when the calcium is too thick for the ultrasound to completely penetrate. In comparison, EBCT accurately quantitates calcium mass regardless of the intralesional calcium morphologic characteristics; however, EBCT does not accurately quantitate plaque volume and will miss noncalcified atherosclerotic lesions.


In vitro atherosclerotic plaque and calcium quantitation by intravascular ultrasound and electron-beam computed tomography
Dan E. Gutfinger The purpose of this investigation was to compare the accuracy of intravascular ultrasound (IVUS) and electron-beam computed tomography (EBCT) in quantitating human atherosclerotic plaque and calcium. In experiment I, 12 human atheroscleroUc arterial segments were obtained at autopsy and imaged by using IVUS and EBCT. The plaque from each arterial segment was dissected and a volume measurement of the dissected plaque was obtained by water displacement. The plaque from each arterial segment was ashed at 700 ° F, and the weight of the remaining ashes was used as an estimate of the calcium mass. In experiment II, 11 calcified arterial segments were obtained at autopsy and imaged by using IVUS at one site along the artery. A corresponding histologic cross section stained with Masson's trichrome was prepared. In experiment I, the mean plaque volume measured by water displacement was 165.3 ± 118.4 IJI. The mean plaque volume calculated by IVUS was 166.1 ___ 114. 4 pl and correlated closely with that by water displacement (r = 0.98, p < 0.0001). The mean calcium mass measured by ashing was 19.4 ± 15.8 mg. The mean calculated calcium mass by EBCT was 19.9 ± 17.2 mg and correlated closely with that by ashing (r= 0.98, p< 0.0001). The mean calculated calcium volume by IVUS was 18.6 ± 11,2 pl and correlated linearly with the calcium mass by ashing (r= 0.87, p < 0.0003). In experiment II, the mean cross-sectional area of the calcified matrix was 1.71 ± 0.66 mm 2 by histologic examination compared with 1.44 ± 0.66 mm 2 by IVUS. There was a good correlation between the calcified cross-sectional area by histologic examination and IVUS (r = 0.76, p < 0.007); however, IVUS may underestimate the amount of calcium present depending on the intralesional calcium morphologic characteristics. In conclusion, IVUS accurately quantitates atherosclerotic plaque volume as well as the cross-sectional area and volume of intralesional calcium, especially if the calcium is localized at the base of the plaque. IVUS under-estimates the amount of calcium present because of signal drop-off when the calcium is too thick for the ultrasound to completely penetrate. In comparison, EBCT accurately quantitates calcium mass regardless of the intralesional calcium morphologic characteristics; however, EBCT does not accurately quantitate plaque volume and will miss noncalcified atherosclerotic lesions. (AM HEART J 1996;131:899-906.) A method for quantitating atherosclerotic plaque and calcium would be helpful in diagnosing and treating coronary artery disease. Intravascular ultrasound (IVUS) and electron-beam computed tomography (EBCT) are imaging modalities that have been used 11° to measure these components of coronary artery disease. IVUS uses high frequency sound waves to produce a reflectance image that has a spatial resolution of 100 pro. It visualizes the components of the arterial wall in cross section and is therefore able to quantitate the extent of plaque that cannot be obtained by angiography. EBCT uses x rays to produce a densitometric image that has a spatial resolution of 500 ~m. It is most suitable for visualizing calcium. The primary benefit of EBCT is that it is a noninvasive method that can be used to screen large numbers of patients compared with more invasive methods such as angiography and IVUS.
Previous studies 1-~ have shown that EBCT accurarely quantitates atherosclerotic calcium mass; however, no studies have shown that EBCT can quantitate noncalcified atherosclerotic plaque. In contrast, comparison studies with histologic examination have shown that IVUS provides an accurate measure of the cross-sectional area (CSA) of noncalcified atherosclerotic plaque, 4, 5 and there are a few studies that validate IVUS for quantitating atherosclerotic plaque volume. 6, 7 IVUS can also be used to detect intralesional calciumS-l°; however, because of "shadowing" and signal dropout caused by calcium, reliable quantitation of atherosclerotic calcium by IVUS compared with histologic examination has not been demonstrated. Calcium in IVUS images is commonly quantitated by degrees of arc and lesion lengthS-l°; there are no studies evaluating the effectiveness of IVUS in estimating calcium CSA or volume. The objective of this in vitro study was to compare the accuracy of IVUS and EBCT in quantitating h u m a n atherosclerotic plaque and calcium.

Experiment h Volume and mass measurement of plaque and calcium
Intravascular ultrasound imaging. Twelve human atherosclerotic arterial segments (11 coronary and 1 iliac side branch) were obtained at autopsy from 12 cadavers (8 men and 4 women) ranging in age from 41 to 96 years with a mean age of 62.8 _+ 16.9 years. The arteries were stored in normal saline at 4 ° C and were not fixed in formalin until after IVUS imaging was completed. IVUS imaging was performed within 48 hours after autopsy. For each artery two surgical needles were placed through the adventitia of the artery to provide acoustic reference points for the starting and ending points during a pullback of each arterial segment. The artery was connected to an 8F coronary guiding catheter that was pressurized with normal saline within a closed system to permit imaging under physiologic pressures (100 mm Hg). The artery was immersed in a normal saline bath at room temperature during imaging. A 25-MHz IVUS catheter (InterTherapy/CVIS, Sunnyvale, Calif.) with a motorized pullback device was used to capture serial cross-sectional images of the arterial segment between the two surgical needles on super-VHS videotape at 30 frames/sec with a pullback rate of 0.25 mm/sec. At completion of IVUS imaging, each of the needle sites was marked with silk suture and the needles were removed. The arteries were then pressure perfused at 100 mm Hg for 24 hours in 10% formalin. Subsequently the arterial segment localized between the two suture marks was excised, and the remaining arterial specimen was discarded. The excised arterial segments ranged in length from 5 mm to 37 mm with a mean length of 17.0 _+ 11.3 mm. The arterial segments were stored in formalin at room temperature until EBCT imaging was performed.
EBCT imaging. EBCT imaging was performed by using an electron beam scanner (Imatron, South San Francisco, Calif.). This scanner was calibrated so that a CT number of 0 corresponded to the linear mass attenuation coefficient of pure water. Two scans were performed for each arterial segment. In the first scan, a 30-cm field of view was used, and this was immediately followed by a second scan with a 9-cm field of view to provide increased spatial resolution. Both scans were performed with a 3-mm slice thickness. Each image consisted of 512 x 512 x 3 mm volume elements (voxels). The arterial segments during EBCT imaging were placed inside a chest phantom as previously described.2, 3 No changes in the chest phantom orientation or position were made between the two scans. This phantom contained regions corresponding to the chest wall, lungs, vertebral column, and heart. The heart region had 10 cylinders arranged in a circle, as illustrated in Fig. 1). Each cylinder was filled with normal saline, and care was taken to eliminate all air bubbles. Arterial segments were placed in only five of the cylinders, located in the right half of the heart region. The remaining five cylinders in the left half of the heart region were filled with only normal saline so that each cylinder containing an arterial segment had a corresponding contralateral cylinder filled with saline. These saline-filled cylinders provided reference CT numbers, which were used to overcome beam hardening effects. 11 A standard bone mineral phantom (0, 50, 100, 200 mg/ml calcium hydroxyapatite) developed by Arnold and Rowberg12,13 was placed on top of the chest phantom during scanning. This standard bone mineral phantom was visible only in the EBCT images with the 30-cm field of view and provided a correlation between calcium hydroxyapatite density and CT number for each image slice. Because a maximum of only five arterial segments could be imaged during each EBCT scan, six separate scans (three with a 30-cm field of view and three with a 9-cm field of view) were required to image all 12 arterial segments.
Plaque mass and volume measurement. At the completion of EBCT imaging, the arterial segments were removed from the chest phantom. Each arterial segment was "filetopened" by performing a single longitudinal cut through all layers of the arterial wall along the long axis of the artery. The site for the single longitudinal cut was chosen to be in the region with the least amount ofatherosclerotic plaque. The plaque was then separated in its entirety by peeling back the overlying media, adventitia, and perivascular tissue along the natural dissection plane localized between the plaque and media. The dissected plaque, free of media and adventitia, was then weighed, and a volume measure- ment was made by water displacement inside a calibrated 1 ml syringe.
Calcium mass measurement. All the tissue for each arterial segment was saved inside a separate glass flask, whose dried emptied weight was determined. Each of the flasks was placed inside an oven at 700 ° F for 72 hours. The total calcium mass for each arterial segment was then measured as the difference between the weight of the flask containing the remaining ashes and the weight of the dried empty flask. 3, 14 IVUS image analysis. The IVUS images were digitized (RasterOps 24 STV board; Mediagrabber software, Santa Clara, Calif.) from the super-VHS videotape at 0.5 mm intervals along the artery length and stored on a computer (Macintosh IIci, Apple, Cupertino, Calif.). The starting and ending points of each arterial segment were identified on the videotape by the presence of the acoustic needles. The pullback rate (0.25 mm/sec) multiplied by the time interval on the videotape between the appearance of the starting and ending acoustic needles was used as an estimate of the arterial segment length. The plaque CSA was then traced in each image by an operator-guided cursor, and a quantitative measurement of the plaque CSA was obtained by an image processing software package (NIH Image, Internet public domain software). The total volume of plaque for each arterial segment was then calculated by summing the individual plaque CSAs over the length of the artery as follows: where length denotes the arterial segment length, CSAi denotes the i th CSA, and N denotes the total number of images grabbed for the arterial segment.
In estimating the calcium matrix volume from IVUS, the same procedure was used; but instead of tracing the plaque CSA in each image, the calcium CSA was traced. Two criteria had to be simultaneously satisfied to identify a calcifled region. First, the calcified region had to produce a highly echogenic signal that was uniformly saturated in intensity and localized within the region corresponding to the vessel wall. Second, the calcified region had to produce either signal drop-off or some degree of signal attenuation (shadowing) from the tissues peripheral to the calcified region. It was noted that when the ultrasound catheter was advancing through a calcified lesion, there was a gradual increase and decrease in the degree of shadowing and that shadowing was apparent only after the catheter had advanced partly through the calcified lesion. Therefore the second criteria was relaxed if, after reviewing the videotape in real time, it was clear that a highly echogenic region corresponded to the tail end of a calcified lesion. The area traced for each calcified region contained only the highly echogenic region that was uniformly saturated in intensity. No attempt was made to include areas of signal drop-off or shadowing. Fig. 2 illustrates an IVUS image and the corresponding EBCT image.
EBCTimage analysis. The EBCT images were processed online by using the computer system associated with the scanner. Each arterial segment was analyzed separately. Within each 9-cm field of view, the region corresponding to the cylinder containing the arterial segment was first analyzed for evidence of calcium. The criterion for identifying a calcified region was the presence of at least one voxel with a CT number >100 Hounsfield units. 3 If this criterion was not satisfied, then the contribution of this image slice to the calcium mass was considered to be 0. If there was evidence of calcium, the image slice was further analyzed by a mathematical model that takes into account partial volume effects (see Appendix). 3, 15  Statistics. Table I summarizes the measurements made in this experiment. Several comparisons were made: (1) plaque volume from intravascular ultrasound was compared with that from water displacement; (2) calcium matrix volume from IVUS was compared with calcium mass from ashing; and (3) calcium mass from EBCT was compared with that from ashing. For each comparison a paired t test and linear regression analysis were performed. In addition, the Z-transformation was used to compare the correlation coefficients derived for comparisons (2) and (3). To assess the variability of the CSA measurements made for the calcified regions in the IVUS images, a second blinded independent observer traced the calcium cross section in 123 images. Interobserver variability was assessed by linear regression analysis. pl. The correlation coefficient between the measurements of plaque volume by water displacement compared with that by the IVUS images was r = 0.98 (p < 0.0001) (Fig. 3).
The mean calcium mass measured by ashing from the 12 arterial segments was 19.4 _~ 15.8 mg. The mean calcium mass calculated from the EBCT im-   Table II. The correlation coefficients between the calcium mass by ashing compared with EBCT and IVUS are shown in Table III. The correlation between the measurements of calcium mass by EBCT compared with that by ashing (r = 0.98) was with statistical significance (p < 0.05) greater than the correlation between the measurements of calcium matrix volume by IVUS and the calcium mass by ashing (r = 0.87) (Figs. 4 and 5).
Although the correlation between calcium mass by EBCT and that by ashing was excellent (r = 0.98, p < 0.0001), the correlation between calcium mass by EBCT and the measured dissected plaque mass (r = 0.67, p < 0.02) was not as close (Fig. 6). Moreover, the mean calcium mass by EBCT (19.9 _+ 17.2 rag) was one order of magnitude smaller than the mean measured dissected plaque mass (167.8 _+ 115.0 rag).
The mean calcium CSA in 123 IVUS images measured by two independent observers was 1.04 _+ 0.92 mm 2 for observer A compared with 1.14 _+ 1.00 mm 2 for observer B. The correlation coefficient between the measurements made by observers A and B was r = 0.98 (p < 0.0001). Observer B had a mean positive bias of 0.1 _+ 0.2 mm 2 relative to observer A, which clinically is an acceptable variation.
Experi ment II. From the 23 arterial segments processed, only 17 (74%) had evidence of calcified atherosclerosis by histologic examination, but only 11 (48%) had acceptable histologic cross sections in which the vessel wall morphologic features were preserved so that the calcium CSA could be measured. For these 11 histologic cross sections the mean CSA of the calcified region was 1.71 _+ 0.66 mm 2. The mean CSA of the calcified matrix from the 11 corresponding IVUS images was 1.44 _+ 0.66 mm 2. There was not a statistically significant difference between these means by the paired t test. The correlation coefficient between the calcified CSA determined by histologic examination and IVUS was r=0.76 (p < 0.007), Fig. 7.
To further explore how IVUS images represent the geometric deposition of calcium within a plaque, the histologic cross sections of the 11 specimens from experiment II were reexamined, and each was classified into one of three types according to the location of the intralesional calcium within the plaque. Fig. 8 shows a sketch illustrating the three histologic types.
Intralesional calcium localized at the base of the plaque was classified as type I. Intralesional calcium localized more superficially within the plaque so that plaque extended behind the calcium was classified as type II. Intralesional calcium that consisted of two separate pieces of calcium with one behind the other was classified as type III. Of the 11 specimens, 7 (64%) were classified as type I, 3 (27%) were classified as type II, and 1 (9%) was classified as type III. Whereas histologic types I and II were accurately represented by IVUS imaging, the deeper calcified area of histologic type III was obscured by shadowing from the more centrally located calcified matrix and led to underestimation of the calcium CSA by IVUS.

DI SCUSSI ON
This in vitro study demonstrates that IVUS is valuable in quantitating atherosclerotic plaque and calcium, and EBCT accurately predicts only the amount of calcium present in the plaque. EBCT accomplishes this in a noninvasive manner, and it is therefore useful as a screening test for the presence of coronary artery calcium. IVUS is an invasive procedure but provides an accurate image of the plaque cross-sectional morphologic features, which can otherwise only be obtained by histologic examination. In addition, the estimation of plaque volume by IVUS correlates closely with the actual plaque volume measured by water displacement (r = 0.98, p < 0.0001).
Since the majority (76%) 9 of significant lesions have some amount of calcification present, calcium is a good marker for the presence of disease. 16,17 In one recent study in particular, the calcium area noted by EBCT could be used to discriminate between patients with and without significant disease with an overall accuracy of 83%. 16 However, if the atherosclerotic plaque is noncalcified, as demonstrated in experiment I, EBCT will not identify such lesions; it is therefore not as accurate as IVUS in quantitating the extent of noncalcified atherosclerosis. One may argue, however, that if none of the lesions are calcified, then the extent of the disease is most likely not severe enough to be clinically significant. 9, 16, 17 Experiment I also demonstrates that the calcium mass estimated by EBCT correlates with the measured dissected plaque mass (r = 0.67, p < 0.02); however, the mean calcium mass by EBCT (19.9 _+ 17.2 mg) was one order of magnitude smaller than the measured mean dissected plaque mass (167.8 _+ 115.0 mg). This suggests that the calcified portion of the plaque is only a small part of the entire atherosclerotic lesion. Nonetheless, because EBCT estimates calcium mass with great accuracy, it may be reliable in serially tracking the progression of calcium in coronary arteries.
In this study it was demonstrated that IVUS is effective in quantitating the CSA and the total volume of the calcified matrix. Although IVUS cannot quantitate total calcium mass directly, a specific density of 1 mg/pl may be used to convert the volume measurement to a mass estimate. Our data show that the correlation between calcium volume by IVUS and calcium mass by ashing (r = 0.87, p < 0.0003) is very close--but statistically less close than the correlation between calcium mass by EBCT and that by ashing (r = 0.98, p < 0.0001). The ability of IVUS to quantitate calcium CSA or volume has previously been questioned because of the shadowing effect produced by calcium. It was therefore expected that IVUS would tend to underestimate the amount of calcium; however, no statistically significant underestimation was demonstrated in either experiment. The accuracy of IVUS in quantitating the calcium CSA depends on the intralesional calcium location (Fig. 8). Shadowing is less likely to interfere with calcium CSA measurement when the calcium is localized at the base of the plaque (type I histologic pattern) because the thickness of the calcium is narrow and there is no component of the calcium that is shadowed. For intralesional calcium that is more superficially localized (type II histologic pattern), it is unclear how accurate IVUS is in quantitating the calcium CSA because the depth of IVUS penetration into the calcified lesion is limited. Thus it is likely that for thick type II lesions IVUS may penetrate only partially and therefore lead to underestimation of calcium. For intralesional calcium that consists of two separate pieces of calcium with one behind the other (type III histologic pattern), the more peripheral piece of calcium is likely to be shadowed and will also lead to underestimation of calcium by IVUS.
Study limitations. The analysis of the histologic cross sections of experiment II shows the presence of different types of intralesional calcium location. Since only 11 specimens were used, the prevalence of these histologic types is unknown. Type III represents the most challenging morphologic feature for accurate calcium quantitation by IVUS. In experiment I, no histologic pattern was used to validate the plaque and calcium volume measurements because the arteries were sacrificed during ashing. It is unclear what percentage of the 12 specimens from experiment I had calcium with type III histologic pattern. If none of the 12 specimens had calcium with type III histologic pattern, the results would be biased toward a good correlation with ashing.
The criterion for identifying a calcified region in the EBCT images is based on a single-voxel threshold. 3 This criterion was satisfactory in our in vitro setting in which measurements of calcium were restricted only to the cylindrical regions. In an in vivo setting, where motion artifacts are present and the exact location of arteries is not known in advance, using a single voxel to define a calcified region may lead to error, is We therefore agree that by using a series of contiguous voxels is with CT numbers greater than a specified threshold would be a more suitable criterion. Conclusion. IVUS can accurately quantitate atherosclerotic plaque volume and is also useful in quantitating the CSA and volume of intralesional calcium. However, IVUS may underestimate the amount of calcium present depending on the intralesional calcium morphologic features. For intralesional calcium localized at the base of the plaque, or for more superficial calcium that is thin, IVUS accurately quantitates the amount of calcium present. In comparison, EBCT accurately quantitates calcium mass regardless of the intralesional calcium morphologic features. EBCT however, cannot be used to accurately quantitate atherosclerotic plaque mass and will miss noncalcified atherosclerotic lesions.

APPENDIX
To calculate the calcium mass from the EBCT images, a mathematic model that takes into account partial volume effects was used. 3, 15 In this model, the CT number of each voxel, CTi, is represented as a linear superposition of the signals from calcified and noncalcified tissues as follows: CTi = CT20o × WCA + CTo x WNC where CT200 denotes the CT number for a purely calcified voxel with a calcium density of 200 mg/ml, and CTo denotes the CT number for a purely noncalcified voxel with a calcium density of 0 mg/cc. WCA and WNC denote weights corresponding to the fractions of the voxel that contain purely calcified and noncalcified tissues, respectively. These weights satisfy the following relationship: 1 = WCA + WNC. Solving for WcA yields: This equation is illustrated in Fig. 9. CT0 in the numerator of this equation was approximated by the mean CT number within the contralateral cylinder containing only normal saline. Since this reference CT number is computed from a region inside the phantom that is localized at the same depth from the surface of the phantom, beam hardening effects are thereby overcome. 11 If CTo was estimated directly from the standard bone mineral phantom, which was placed on top of the chest phantom, beam hardening effects would not be properly compensated. The expression in the denominator of this equation was approximated by the slope of the linear regression line fitted to the measured CT numbers in the 30-cm field of view and the known physical densities of the calcium hydroxyapatite in the standard bone mineral phantom.12, 13 Analysis has shown that this slope does not significantly change between independent scans performed during the same day.
To obtain the total calcium mass estimate for each arterial segment, the estimated calcium densities for each voxel were integrated over the cylindrical region corresponding to the arterial segment and then summed over all image slices that had evidence of calcium. This double summation is mathematically expressed as follows: CT i -CTo Calcium mass = ~ slices i Slope Because the CT numbers of soft tissues such as collagen, muscle, fat, and water are poorly differentiated, it was not possible to independently estimate plaque volume or mass from the EBCT data.