Is Coronary Stent Assessment Improved with Spectral Analysis of Dual Energy CT? 1

Is Coronary Stent Assessment Improved with Spectral Analysis of Dual Energy CT? 1 Ethan J. Halpern, MD, David J. Halpern, Jeffrey H. Yanof, PhD, Sigal Amin-Spector, PhD, David Fischman, MD, Galit Aviram, MD, Jacob Sosna, MD Rationale and Objectives. The aims of this study were to distinguish stents from iodinated contrast on the basis of spectral characteristics on dual-energy computed tomographic (DECT) imaging and to determine whether DECT imaging might provide a more accurate measurement of true stent lumen. Materials and Methods. Three stainless steel stents and one cobalt chromium stent were scanned using a multidetector, singlesource DECT scanner. Stents 2.5, 3.5, and 4.0 mm in diameter were filled with iodinated contrast, submerged in water, and scanned. Spectral analysis was performed to assess the separation of stents from iodinated contrast. Two independent reviewers measured stent lumen diameter and strut thickness on low-energy (L 0 ), high-energy (L 1 ), and combined-energy (L c ) images. Dual-energy full-width half-maximum edge detection analysis was used to provide an independent assessment of stent luminal diameter and strut thickness. Results. Two-dimensional graphical plots of computed tomographic attenuation for the L 0 and L 1 images did not demonstrate a sharp separation between the absorption characteristics of stents and iodinated contrast material. Stent lumens were underestimated by approximately 50% on L c images. Observer measurements on L 1 images demonstrated a 24% decrease in strut thickness and a 25% increase in stent luminal diameter compared to L 0 images (P <.0001). Full-width half-maximum measurements did not demonstrate significant changes in stent luminal diameters or strut thicknesses between L 0 and L 1 images. Conclusions. Spectral analysis did not clearly distinguish stents from iodinated contrast with the DECT system used in this study. The larger stent lumens visualized by the high-energy components of the x-ray spectrum were not related to improved computed tomographic delineation of stent thickness. Key Words. Coronary artery, coronary stent, CT angiography, dual energy CT, spectral classification. ª AUR, 2009 Acad Radiol 2009; -:1 10 1 From the Department of Radiology (E.J.H., D.J.H.) and the Division of Cardiology, Department of Internal Medicine (D.F.), Thomas Jefferson University, 132 S 10th Street, Philadelphia, PA 19107-5244; Philips Medical Systems, Cleveland, OH (J.H.Y., S.A.-S.); the Department of Radiology, Tel Aviv Sourasky Medical Center, Tel Aviv, Israel (G.A.); and the Department of Radiology, Hadassah Hebrew University Hospital, Jerusalem, Israel (J.S.). Received March 24, 2009; accepted April 21, 2009. Address correspondence to: E.J.H. e-mail: ethan.halpern@jefferson.edu ª AUR, 2009 doi:10.1016/j.acra.2009.04.011 Computed tomographic (CT) evaluation of coronary stent restenosis is limited by blooming artifacts from high x-ray attenuating materials, such as metal struts and mural calcium. Artifacts from high-density stent material lead to the overestimation of stent strut thickness and the underestimation of stent luminal diameter (1). In vitro evaluation on a four-slice scanner revealed that luminal narrowing of a stainless steel stent may be as great as 62% to 94.3% (2). The newer 64 detector row CT scanners provide near isotropic pixels with improved spatial resolution that can reduce blooming artifacts and minimize the importance of stent orientation compared to earlier 16-detector scanners (3,4). Nonetheless, imaging of coronary stents on computed tomography (CT) remains a technically challenging task. Conventional CT scanners display an image on the basis of differences in the x-ray attenuation of different tissues. Dual-energy CT (DECT) imaging improves the differentiation of tissues on the basis of CT data obtained from synchronous CT acquisitions at two different energies (5). On the basis of the preferential absorption of photons by different materials, high-density structures such as calcium and iodine 1

HALPERN ET AL Academic Radiology, Vol -, No-, - 2009 may be separated. Studies have suggested that the CT characterization of the coronary lumen (6), calcified coronary plaque (7), and the coronary stent lumen (8) may be improved with DECT imaging. An additional advantage of DECT imaging may be related to decreases in beam-hardening and blooming artifacts in the high-energy component of DECT imaging. The present study was performed to assess in a phantom model the ability to separate stents from iodinated contrast on the basis of their distinct spectral characteristics and to determine whether DECT imaging might provide a more accurate measurement of the true stent lumen. MATERIALS AND METHODS Experimental Overview Stents were deployed within a phantom model consisting of tubes containing iodinated contrast. After DECT imaging, spectral analysis was performed to assess the separation of the stents from contrast. Quantitative measurements of stent lumen and strut thickness were performed for each energy level by two independent observers. In addition to these observer measurements, full-width half-maximum () measurement were performed to obtain an objective assessment of luminal diameter and stent strut thickness. Detailed description is presented below for each component of our experiment. Data Acquisition Stents Three stainless steel stents the Multi-Link Zeta (Guidant Corporation, Indianapolis, IN), the Zipper MX (Medtronic, Inc, Minneapolis, MN), and the Cypher sirolimus-eluting stent (Cordis Corporation, Miami Lakes, FL) and one cobalt chromium stent, the Vision (Guidant Corporation), were scanned using a multidetector DECT scanner. Stents 2.5, 3.5, and 4.0 mm diameter were deployed within appropriately sized plastic tubing filled with iodinated contrast material (Iopamiro 300 ng iodine/ml, Bracco, Milano, Italy) diluted with normal saline to achieve a density of approximately 300 to 400 Hounsfield units (HU) when imaged in high-resolution mode. This density is similar to the level of enhancement that might be expected on high-quality coronary CT angiography. Phantom Design Each stent was submerged to the center of a water bath and scanned at 0,90, and oblique to the z-axis of the scanner. DECT System The prototype single-source DECT system used in this study is a variation of the Brilliance 64 multislice scanner (Philips Medical Systems, Cleveland, OH). This scanner has a single x-ray source (set to 140 kvp) and a double-layered, 32 detector row panel. The double-decker detector acquires low-energy and high-energy x-ray photons simultaneously, each layer absorbing about 50% of the photons. A preponderance of low-energy photons are processed by the upper detector, which the photons encounter first. Higher energy photons pass through to the deeper detector. Photons absorbed by the upper and lower detector arrays are processed independently, with additional data preprocessing for polychromatic correction to create a low-energy (L 0 ) image from the data of the upper detector and a high-energy (L 1 ) image from the data of the lower detector (9). These images are provided in addition to the conventional (L c ) CT image that combines the data from both detectors and represents the equivalent of a clinically used CT image. The result is three sets of images: one set using the lower portion of the energy spectrum, one set using the higher portion of the energy spectrum, and one set using the entire spectrum. Scanning Technique All images were obtained with 32 0.625 mm collimation. Images were obtained in helical mode, with a pitch of 0.218 and a gantry rotation time of 420 ms. The x-ray tube was set to 140 kvp. Each stent was scanned in both standardresolution and high-resolution modes. A 512 512 image matrix was used for image reconstruction of a 15 15 cm field of view at overlapping intervals of 0.45 mm using the edge-enhancing D reconstruction kernel (Philips Medical Systems). On the basis of the field of view and image matrix size, the in-plane size of each pixel corresponded to approximately 0.3 mm. Data Analysis Spectral Analysis Because each pixel has two separate HU values from the L 0 and L 1 images, high-energy values can be compared to low-energy HU values. A scatterplot of pixel values from the L 0 and L 1 images results in a distribution that is unique for various materials. To determine whether dual-energy images could be used to classify voxels as stent versus lumen, images were displayed with the novel dual-energy Spectral Viewer (Philips Medical Systems). For each slice, the Spectral Viewer allows the user to view a scatterplot of the distribution of paired CT attenuation numbers on the L 0 and L 1 axes. The user can interactively specify a line segment that linearly classifies the image voxels into two groups (stent vs contrast) on the basis of differential CT attenuation on low-energy and high-energy images. Voxels were color coded on the basis of this classification (Fig 1). Quantitative Measurements of Stent Lumen and Strut Thickness For the purposes of measuring stent luminal diameter, strut thickness, and CT density, the Cardiac Viewer (Philips 2

Academic Radiology, Vol -, No-, - 2009 DUAL ENERGY ASSESSMENT OF CORONARY STENTS Medical Systems) was used in multiphase mode to allow simultaneous display and manipulation of multiple images for each stent. Specifically, a single-screen display was created for L c,l 0, and L 1 images acquired in standard-resolution and high-resolution modes. Image volumes were magnified to the maximum possible degree (Z = 10) and rotated to create a 0.5-mm slab maximum-intensity projection image through the center of the stent lumen along the long axis of each stent (Fig 2). Two independent reviewers, blinded to the true stent size and to each other s measurements, measured the stent lumen and strut thickness for each stent. Because the struts often present a beaded, discontinuous appearance along the course of the stent, reviewers were instructed to measure both the lumen and strut thickness at a point of maximum strut thickness (Fig 3). Each stent was measured on L c,l 0, and L 1 images that were acquired in both standard-resolution and high-resolution modes in each of three scan orientations (18 scan measurements per stent). These measurements were performed with standardized window and level settings. The first reviewer used a CT window of 1000 at a level of 800, while the second reviewer used a CT window of 800 at a level of 1000 for all measurements. Fixed widowing was used to eliminate a possible change in image contrast provided by different windowing. Measurement A dual-energy edge detection algorithm included in the qcta package (Philips Medical Systems) was used to provide an independent semiautomatic assessment of stent lumen diameter and strut thickness for each stent at each energy level. For assessment of, the reviewer placed a rectangle of interest perpendicular to the long axis of the stent at the same level that was used for stent lumen and strut measurements (Fig 4). Statistical Analysis Measurements of stent lumen and strut thickness were placed into a single database and analyzed using Stata version 10 (StataCorp LP, College Station, TX). Analysis of variance was performed, with each luminal diameter and strut thickness as the dependent variable. Independent variables included stent type, stent size, plane of imaging (0,90, and oblique), type of measurement (reader 1 vs reader 2 vs ), scan acquisition mode (standard-resolution vs high-resolution mode), and energy level (L c vs L 0 vs L 1 ). An F statistic was calculated for the overall analysis as well as for each independent variable. To further evaluate differences between L 0 and L 1 images, stent luminal diameter and strut thickness data were evaluated with paired t tests to determine which subcomparisons demonstrated statistically significant differences. Finally, to determine the correlation between measurements by our two independent reviewers, Pearson s correlation coefficient was computed for the measurement of strut diameter, internal stent lumen, and external stent diameter. RESULTS Spectral Analysis The two-dimensional graphical plot of CT attenuation values for the L 0 and L 1 images did not demonstrate a sharp separation between the dual-energy absorption characteristics of stents and iodinated contrast material. Nonetheless, using the Spectral Viewer, it was possible to classify stent voxels (from adjacent contrast-enhanced lumen voxels) and to highlight the stent voxels in color on the basis of a userdefined dual-energy classification map (Fig 1). The overlap between the DECT distribution of the stents and contrast material was found for cobalt chromium as well as stainless steel stents and reduced the reliability of the colorized stent images for the assessment of stent luminal diameter (Fig 1A vs 1B and Fig 1C vs 1D). Quantitative Measurements Analysis of variance for measurements of stent luminal diameter and stent strut thickness demonstrated a highly significant F statistic for the overall analysis as well as for the individual independent variables (stent type, stent size, plane of imaging, type of measurement, resolution mode, and energy level). Combined Conventional Imaging The mean measured stent luminal diameter for a 2.5-mm stent was 1.1 to 1.2 mm (standard deviation, 0.44 mm), while the mean measured diameter of a 3.5-mm to 4.0-mm stent was 1.7 to 1.8 mm (standard deviation, 0.44 mm). Stent struts all measured approximately 1 mm in thickness (Fig 3). As demonstrated in Table 1, the true lumens of coronary stents were underestimated by approximately 50% on conventional combined-energy CT images. As demonstrated in Table 2, strut thickness, which measures approximately 0.1 mm in true thickness, was overestimated by a factor of 10 by both readers as well as by estimates. Low-energy Versus High-energy Imaging Overall, mean stent luminal diameter measured 25% (0.27 0.47 mm) larger and mean strut thickness measured 24% (0.26 0.32mm) smaller on L 1 images compared to L 0 images (P <.0001). A similar difference of 0.23 to 0.35 mm was found between mean stent luminal diameter on L c images compared to L 0 images, which always demonstrated a smaller luminal diameter (P <.01). With respect to stent struts, mean strut thickness was 0.16 to 0.25 mm thinner on L c images compared to L 0 images (P <.01). There was a tendency for slightly larger stent luminal diameter 3

HALPERN ET AL Academic Radiology, Vol -, No-, - 2009 Figure 1. Spectral separation. Spectral Viewer displays a color-coded image of the stent (yellow) and contrast-enhanced internal lumen (blue) on the basis of dual-energy classification of the tissues. A graphical scatterplot of the low-energy versus highenergy attenuation values within the computed tomographic (CT) slice is displayed, along with a user-defined line that is used to classify stent material from iodinated contrast. Dual-energy values that fall below the user-defined line are color coded as yellow stent material. (a,b) Dual-energy CT (DECT) classification of a Cypher 3.5-mm stainless steel stent; (c,d) DECT classification of a Vision 3.5-mm cobalt chromium stent. The dual-energy graphical plots do not demonstrate a clear separation between stent and iodine within the stent lumen. Consequently, the positioning of the user-defined classification map is subjective. Slight changes in this user-defined classification map result in obvious difference in stent diameter and strut thickness, as demonstrated in (a) versus (b) and in (c) versus (d). measurements and slightly thinner strut thickness measurements on L 1 images compared to the L c images (Tables 1 and 2), but this finding did not reach statistical significance for either reader. Resolution Mode Use of the high-resolution mode resulted in a 7% to 12% larger measurement of stent luminal diameter (mean difference, 0.1 0.17 mm; P <.001) and a 4% to 5% smaller 4

Academic Radiology, Vol -, No-, - 2009 DUAL ENERGY ASSESSMENT OF CORONARY STENTS measurement of strut thickness (mean difference, 0.04 0.06 mm; P =.01). Imaging Plane There was a definite impact of the stent imaging plane on measurement of stent luminal diameter and strut thickness. Stent luminal diameter appeared smaller and strut thickness was thicker when the stent was oriented perpendicular to the scan plane (Tables 3 and 4). This relationship was most obvious for the smaller 2.5-mm stents (P <.05 for all comparisons). There was a similar tendency toward smaller luminal diameter and greater strut thickness for larger stents oriented perpendicular to the scan plane, but this relationship did not reach statistical significance for comparisons at most stenttype and energy-level combinations. Measurement The difference between mean stent luminal diameter estimates by on L 0 and L 1 images ranged from 0.05 to 0.25 mm but was not statistically significant for any single stent type or size. Similarly, there was no significant difference between stent luminal diameter measurements on L c,l 0, and L 1 images. With respect to stent struts, differences in mean strut thickness by on L 0 and L 1 images ranged from 0.05 to 0.25 mm but were not statistically significant for any single stent type or size. Similarly, there was no significant difference between stent strut thickness on L c images and either L 0 or L 1 images. The use of the high-resolution mode resulted in a slightly larger measurement of stent luminal diameter (mean difference, 0.17 mm; P <.001) and a slightly smaller measurement of strut thickness (mean difference, 0.14 mm; P <.001). Interobserver Correlation Pearson s correlation coefficients for the measurements by our two independent reviewers are reported in Table 5. Excellent interobserver correlation was demonstrated for internal and external stent diameters, with coefficients all $0.90. Interobserver correlation coefficients for measurements of strut thickness were lower (0.50 0.88), especially on the L 1 images (0.50), for which measured strut thickness was <1 mm (Table 2). DISCUSSION The introduction of 64-detector CT scanners with isotropic voxel resolution has resulted in a dramatic improvement in the quality of coronary CT angiography for native coronary arteries. Nonetheless, imaging of calcified coronary vessels and coronary stents continues to be limited by underestimation of the true patent lumen. Studies have suggested that DECT imaging can improve quantification of small calcified plaques and facilitate enhanced coronary stent lumen depiction (7,8). To empirically evaluate DECT imaging for the assessment of stents, we selected four stent types that are commonly used in our clinical practice and evaluated these stents at diameters that are frequently used in our clinical practice. We performed an in vitro evaluation without previously performing a detailed simulation of the absorption patterns of the stents, because their precise composition is not known. Our results suggest that spectral separation of stents from iodinated contrast is limited by the overlap of their respective spectral absorption patterns and that there is a high degree of variability in tissue classification on DECT imaging between metallic stents and iodinated contrast. The lack of a clear linear separation between the attenuation values of the stent and its contrast-enhanced lumen on the dual-energy plot precludes the use of these maps for objective measurement of stent strut thickness and luminal diameter. Measurements by our independent observers with highenergy images indeed confirmed an improvement in the definition of stent lumen compared to low-energy images, though the stent lumen was underestimated even on highenergy images. However, calculations suggest that there is no significant objective improvement in the CT assessment of stent luminal diameter and strut thickness with images reconstructed from the higher energy portion of the x-ray spectrum. is an objective computation of the thickness and diameter of a structure or lumen based on the difference between the two extreme values of the independent variable (position) at which the dependent variable (density) is equal to half of its maximum value. The assessment of the coronary stent lumen may be improved by using a wide window setting, with a window width as large as 1000 to 1500 often used in clinical practice (10). A wider window setting reduces the apparent impact of blooming and allows a closer estimate of the true position of the stent border, though it does not truly reduce the stent thickness in the CT data. Although this change may improve the assessment of stent luminal diameter, it may also negatively affect the assessment of stent patency by reducing the visibility of contrast within the stent lumen. The apparent increase in the measured stent lumen in our high-energy images using DECT imaging is similar to the effect achieved with a wide window setting. The decreased CT density of stent material on high-energy images results in a larger apparent lumen. However, measurements suggest that if the window and level could be adjusted to compensate for this change in CT density, the measured stent lumen and strut thickness would be unchanged by tube voltage. The two major artifacts that limit visualization of the stent lumen are blooming and beam hardening. Blooming artifact is primarily related to limited spatial resolution, which results in the presentation of a very small high-attenuating edge 5

HALPERN ET AL Academic Radiology, Vol -, No-, - 2009 Figure 2. Multiphase display of a 3.5-mm stent that was imaged in an oblique orientation to the scan plane. The display is magnified to the maximum possible degree and rotated to create a long-axis thin slice through the center of the stent lumen. The top row of images demonstrates combined-energy (left), low-energy (middle), and high-energy (right) computed tomographic (CT) images acquired in standard-resolution mode. The bottom row of images demonstrates combined-energy (left), low-energy (middle), and high-energy (right) CT images acquired in high-resolution mode. 6

Academic Radiology, Vol -, No-, - 2009 DUAL ENERGY ASSESSMENT OF CORONARY STENTS Figure 3. Multiphase display of a 3.5-mm stent illustrating the technique used for quantitative measurement. The display is magnified to the maximum possible degree and rotated to create a long-axis thin slice through the center of the stent lumen. The top row of images demonstrates combined-energy (left), low-energy (middle), and high-energy (right) computed tomographic (CT) images acquired in standardresolution mode. The bottom row of images demonstrates combined-energy (left), low-energy (middle), and high-energy (right) CT images acquired in high-resolution mode. The stent lumen is measured in (a), and the strut thickness is measured in (b). within a voxel as filling the entire voxel. Increased sampling density or improved spatial resolution can be used to minimize this artifact (11). Beam-hardening artifact results in distortion of the CT density in the local vicinity of a large change in CT density as a result of the polychromatic nature of the x-ray. Blooming artifact is most important in causing underestimation of the true stent diameter, while beam hardening predominantly results in inaccurate estimation of CT density adjacent to the stent. Both artifacts contribute to difficulty in accurate quantification of the contrast enhanced lumen. Different imaging strategies may be required to address blooming and beam-hardening artifacts in coronary CT angiography. Increased spatial resolution and the use of a sharp reconstruction kernel are known to improve delineation of the stent margin (12). An investigation using a novel flatpanel technology with an isotropic slice thickness of 0.25 mm demonstrated a marked reduction in blooming and resulted in a measured stent luminal narrowing of only 16.1% (13). Because blooming artifact is related to partial volume issues, it seems reasonable that improved CT resolution to <250 mm will reduce blooming artifact. Current CT detectors vary from 0.5 to 0.625 mm in size and are associated with a pixel size that may be half the detector size. Thus, a stent deployed to 2.5 mm may be only 10 pixels in size. The changes measured at different energy levels may be thus <1 pixel and therefore not visualized properly. Future generations of smaller detectors with the use of DECT imaging may be suited better for the assessment of stents. Two recent studies by Boll et al (7,8), using the identical DECT system that was used in the present study, suggested that DECT imaging may be used to improve coronary vascular imaging. In an ex vivo study using atherosclerotic specimens of the common carotid artery, DECT imaging enhanced the accuracy of calcified plaque quantification beyond the scope of single-energy multidetector CT imaging by taking advantage of the difference in dual-energy absorption between calcium and iodinated contrast material 7

HALPERN ET AL Academic Radiology, Vol -, No-, - 2009 Table 1 Stent Lumen by Stent Size and Measurement Method Stent Lumen (mm) Stent Size (mm) L c L 0 L 1 L c L 1 L 1 2.5 1.2 1.1 0.7 0.9 1.2 1.3 1.3 1.3 3.5 1.7 1.8 1.6 1.8 1.9 1.6 1.9 4 1.7 1.4 1.7 1.8 1.9 1.9, full-width half-maximum; L c, combined-energy image; L 0, low-energy image; L 1, high-energy image. Table 2 Strut Thickness by Stent Size and Measurement Method Strut Thickness (mm) Stent Size (mm) L c L 0 L 1 L c L 1 L 1 2.5 0.9 1.0 1.1 1.3 0.8 0.9 1.0 1.0 0.9 3.5 1 1.2 1.2 1.3 0.9 1.1 1.0 1.3 1.1 4 1.0 1.1 1.1 1.4 0.9 1.0 1.2 1.2 1.1, full-width half-maximum; L c, combined-energy image; L 0, low-energy image; L 1, high-energy image. Table 3 Stent Lumen by Stent Size and Stent Orientation Relative to Scan Plane Stent Lumen (mm) Stent Size (mm) Imaging Plane L c L 0 L 1 Figure 4. Full-width half-maximum measurement of a 3.5-mm cobalt chromium stent. To measure the stent lumen (a), the reviewer defined a box perpendicular to the stent lumen. To measure the strut thickness (b), the reviewer defined a box perpendicular to the stent strut. (7). In that study, qualitative analysis demonstrated overestimation of the size of calcified plaque with L 0 images and underestimation of calcified plaque size with L 1 images. A mathematical combination of the L 0 and L 1 images provided a more accurate estimate of true plaque size, as confirmed by optical coherence tomography. Our results differ, as we have not assessed plaques but rather stents. DECT imaging will be most useful when separating two tissues that demonstrate a clear difference in their dual-energy absorption spectra, such as iodine and calcium, rather than iodine and stents. In a second study of coronary stents, a mathematical combination of the L 0 and L 1 images was used to improve the 2.5 Oblique 1.6 1.4 1.6 2.5 Parallel 1.1 0.6 0.9 2.5 Perpendicular 0.8 0.5 1.1 3.5 Oblique 1.8 1.7 1.8 3.5 Parallel 1.8 1.6 1.8 3.5 Perpendicular 1.7 1.5 1.8 4 Oblique 1.7 1.5 1.6 4 Parallel 1.7 1.3 1.7 4 Perpendicular 1.7 1.4 1.8 L c, combined-energy image; L 0, low-energy image; L 1, highenergy image. contrast-to-noise ratio inside a stent and to increase the kurtosis to enhance depiction of the stent lumen (8). By separating different portions of the x-ray absorption spectrum, DECT imaging can be used to compensate for the polychromatic nature of the x-ray, to reduce beam hardening, and to improve the contrast-to-noise ratio within the stent lumen. In our study, spectral analysis resulted in suboptimal separation of iodine and stents from each other on the basis of their physical characteristics. Furthermore, our 8

Academic Radiology, Vol -, No-, - 2009 DUAL ENERGY ASSESSMENT OF CORONARY STENTS Table 4 Strut Thickness by Stent Size and Stent Orientation Relative to Scan Plane Strut Thickness (mm) Stent Size (mm) Imaging Plane L c L 0 L 1 2.5 Oblique 0.9 1.0 0.7 2.5 Parallel 0.9 1.2 0.9 2.5 Perpendicular 1.2 1.4 1.0 3.5 Oblique 1.0 1.2 1.0 3.5 Parallel 1.0 1.1 0.9 3.5 Perpendicular 1.3 1.5 1.2 4 Oblique 1.0 1.2 1.0 4 Parallel 1.0 1.2 0.9 4 Perpendicular 1.1 1.3 1.0 L c, combined-energy image; L 0, low-energy image; L 1, highenergy image. Table 5 Pearson s Correlation Coefficients Between the Two Independent Reviewers Variable L c L 0 L 1 Strut thickness 0.67 0.88 0.50 Stent luminal diameter 0.92 0.95 0.90 Outer stent diameter 0.90 0.93 0.95 L c, combined-energy image; L 0, low-energy image; L 1, highenergy image. Furthermore, the water surrounding our stents provided a less complex environment than the lung and soft tissues that surround the heart. Finally, the lumen of our stents was filled with homogeneous contrast diluted to a CT density of 300 to 400 HU, while in vivo imaging may include more heterogeneous and less optimal enhancement of the coronary arteries. Nonetheless, our in vitro design allowed us to evaluate the impact of dual-energy computed tomography on imaging of the stent lumen and strut thickness. By evaluating both the low-energy and high-energy images from DECT imaging, we were able to assess the impact of CT energy on measurements of stent lumen as well as on the density of stent struts. In conclusion, our study demonstrated suboptimal spectral separation of stents from iodinated contrast with DECT imaging, resulting in a high degree of variability in DECT tissue classification. The true diameter of a coronary stent is underestimated by coronary CT angiography to different extents that depend on stent size, stent orientation, window and level settings, and tube voltage. Although scans obtained with higher tube voltages appear to increase the stent luminal diameter and decrease strut thickness, edge detection using suggests that there is no true improvement in the imaging of stent lumen and strut thickness at higher tube voltages. Future advances in CT spatial resolution may be combined with DECT imaging, along with more advanced classification and image-processing techniques, to reduce both the blooming and beam-hardening artifacts that limit the CT evaluation of coronary stents. results suggest that measurement of stent luminal diameter is not objectively changed in either the L 0 or L 1 component of DECT imaging. The study by Boll et al (8) did not report any quantitative, objective measurements of stent strut thickness or luminal diameter. Because luminal diameter is not significantly changed in either the L 0 or L 1 images, a mathematical combination of these images may sharpen the stent edges but is unlikely to result in a significant improvement in the accuracy of assessment of true stent luminal diameter. Both stent orientation and reconstruction kernel have an important impact on measurement of stent luminal diameter and strut thickness. It is interesting that blooming artifact appeared to be most prominent when the stent was oriented perpendicular to the scan plane. This finding must be related to an asymmetry in the voxels of our DECT scanner. In clinical studies, stents within the more proximal portions of the coronary tree will tend to be aligned more closely in parallel with the scan plane. These stents will also tend to be larger and therefore will be more clearly visualized. Our study was limited by the in vitro design, which used only four different types of stents with three stent diameters and did not attempt to simulate imaging of a beating heart. REFERENCES 1. Nieman K, Cademartiri F, Raaijmakers R, Pattynama P, de Feyter P. Noninvasive angiographic evaluation of coronary stents with multi-slice spiral computed tomography. Herz 2003; 28:136 142. 2. Maintz D, Juergens K, Wichter T, Grude M, Heindel W, Fischbach R. Imaging of coronary artery stents using multislice computed tomography: in vitro evaluation. Eur Radiol 2003; 13:830 835. 3. Mahnken A, Muhlenbruch G, Seyfarth T, et al. 64-slice computed tomography assessment of coronary artery stents: a phantom study. Acta Radiol 2006; 47:36 42. 4. Seifarth H, Ozgun M, Raupach R, et al. 64- versus 16-slice CT angiography for coronary artery stent assessment: in vitro experience. Invest Radiol 2006; 41:22 27. 5. Graser A, Johnson TR, Chandarana H, Macari M. Dual energy CT: preliminary observations and potential clinical applications in the abdomen. Eur Radiol 2009; 19:13 23. 6. Boll DT, Hoffmann MH, Huber N, Bossert AS, Aschoff AJ, Fleiter TR. Spectral coronary multidetector computed tomography angiography: dual benefit by facilitating plaque characterization and enhancing lumen depiction. J Comput Assist Tomogr 2006; 30:804 811. 7. Boll DT, Merkle EM, Paulson EK, Mirza RA, Fleiter TR. Calcified vascular plaque specimens: assessment with cardiac dual-energy multidetector CT in anthropomorphically moving heart phantom. Radiology 2008; 249: 119 126. 8. Boll DT, Merkle EM, Paulson EK, Fleiter TR. Coronary stent patency: dual-energy multidetector CT assessment in a pilot study with anthropomorphic phantom. Radiology 2008; 247:687 695. 9

HALPERN ET AL Academic Radiology, Vol -, No-, - 2009 9. Carmi R, Naveh G, Altman A. Material separation with dual-layer CT. In: Proceedings of IEEE Nuclear Science Symposium 2005. New York: IEEE, 2005; 1876 1878. 10. Halpern EJ. Coronary stents. In: Halpern EJ. Clinical cardiac CT: anatomy and function. New York: Thieme, 2008; 115 129. 11. Joseph PM, Spital RD. The exponential edge-gradient effect in x-ray computed tomography. Phys Med Biol 1981; 26:473 487. 12. Hong C, Chrysant GS, Woodard PK, Bae K. Coronary artery stent patency assessed with in-stent contrast enhancement measured at multi-detector row CT angiography: initial experience. Radiology 2004; 233:286 291. 13. Mahnken A, Seyfarth T, Flohr T, et al. Flat-panel detector computed tomography for the assessment of coronary artery stents: phantom study in comparison with 16-slice spiral computed tomography. Invest Radiol 2005; 40:8 13. 10