| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Vascular and Interventional Radiology |
1 From the Department of Radiology, Thomas Jefferson University Hospital, 132 S 10th St, 7th Floor Main Building, Philadelphia, PA 19107 (P.T.J., E.J.H., R.J.W., L.N.N., G.A.G., D.C.L.), and the Russell H. Morgan Department of Radiology, Johns Hopkins Hospital, Baltimore, Md (B.S.K., D.G.H., E.K.F.). Received February 25, 1998; revision requested April 29; final revision received July 24; accepted October 16. Address reprint requests to P.T.J.
 |
Abstract |
|---|
 TOP Abstract IntroductionMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
MATERIALS AND METHODS: Twenty-five patients underwent both conventional and CT angiography of the renal arteries. Images were blindly reviewed after rendering with MIP and VR algorithms. MIP images were viewed in conjunction with axial CT images; VR models were evaluated in real time at the workstation without CT images. Findings in 50 main and 11 accessory renal arteries were categorized as normal or by degree of stenosis.
RESULTS: All arteries depicted on conventional angiograms were visualized on MIP and VR images. Receiver operating characteristic (ROC) analysis for MIP and VIR images demonstrated excellent discrimination for the diagnosis of stenosis of at least 50% (area under the ROC curve, 0.96–0.99). Although sensitivity was not significantly different for VR and MIP (89% vs 94%, P > .1), specificity was greater with VR (99% vs 87%, P = .008 to .08). Stenosis of at least 50% was overestimated with CT angiography in four accessory renal arteries, but three accessory renal arteries not depicted at conventional angiography were depicted at CT angiography.
CONCLUSION: In the evaluation of renal artery stenosis, CT angiography with VR is faster and more accurate than CT angiography with MIP. Accessory arteries not depicted with conventional angiography were depicted with both CT angiographic algorithms.
Index terms: Computed tomography (CT), angiography, 961.12915, 961.12916, 961.12917 • Computed tomography (CT), comparative studies, 961.12915, 961.12916, 961.12917 • Computed tomography (CT), maximum intensity projection, 961.12912, 961.1299 • Computed tomography (CT), volume rendering, 961.12916, 961.12917 • Renal arteries, CT, 961.12915, 961.12916, 961.12917 • Renal arteries, stenosis or obstruction, 961.721
|
Introduction |
|---|
|
TOPAbstractIntroductionMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
Volume rendering (VR) has been implemented as an interactive 3D rendering algorithm. The VR algorithm computes a volumetric image that includes all voxels within the imaging volume and simultaneously demonstrates vascular anatomy, vascular calcifications, and surrounding tissues. Until recently, this algorithm has not been used widely owing to computational complexity (8). However, affordable workstations are now capable of real-time VR. With VR, 3D vascular relationships are maintained and calcifications are displayed as distinct from vascular enhancement, overcoming several of the limitations of MIP and shaded surface display, respectively (8,9).
The accuracy of the VR algorithm for the grading of stenoses has been demonstrated in vitro (10). A comparison of algorithms by Szeimies et al (11) demonstrated that VR was superior to MIP and shaded surface display in vivo in the detection of vascular stenoses, mural calcium, and thrombus. However, the rendering process involves subjective optimization of the display, and the accuracy for grading vascular stenoses in vivo has not been adequately demonstrated. The purpose of this study was to compare CT angiography with VR to CT angiography with MIP in the evaluation of main and accessory renal artery stenosis.
|
MATERIALS AND METHODS |
|---|
|
TOPAbstractIntroductionMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
All studies were performed on the same CT scanner (HiSpeed Advantage; GE Medical Systems, Milwaukee, Wis). A test scan series was obtained to localize the level of the imaging volume and to determine optimal timing of contrast material infusion. Sequential 10-mm-thick axial images were obtained from T12 through L5 with the start of imaging 8–15 seconds after 20 mL of nonionic contrast material (Omnipaque 300; Nycomed, Princeton, NJ) was infused at 4 mL/sec. Scans were obtained every other second (with a 1-second interscan delay between sections). CT angiography was subsequently performed at peak enhancement from the celiac axis to the aortic bifurcation during a single breath hold with 120 kVp and 260–280 mAs, with 1-second rotation. For each study, 120 mL of contrast material was infused at 4 mL/sec, followed by a 20-mL saline solution flush. Data were acquired with 1-mm collimation and pitch of 2.0 (n = 10) or with 3-mm collimation and pitch of 1.0–2.0 (n = 15). Targeted images were reconstructed at 1 mm (1-mm collimation) or 1.5 mm (3-mm collimation) with a 22-cm field of view.
Data sets were transferred to a workstation (Advantage Windows; GE Medical Systems). The original 512 x 512-matrix CT images were rendered with an MIP algorithm after manual editing to remove osseous structures and other surrounding tissues. MIP images were obtained in projections selected to depict the renal arteries by a radiologist experienced with CT angiography (E.J.H.). The MIP images were interpreted independently and blindly in conjunction with the axial images by two different CT radiologists (R.J.W., L.N.N.).
The same 25 data sets were reduced to 256 x 256-matrix CT images and rendered on a workstation (Onyx with Infinite Reality Graphics; Silicon Graphics, Mountain View, Calif) with customized software (VOLREN 6; Silicon Graphics). The reduction in matrix size was necessary to allow rendering at real-time rates (4–10 frames per second) with available hardware. Display parameters including width, level, opacity, and brightness were chosen subjectively by the individual radiologist. The 3D display images were evaluated independently and blindly in real time without the axial images by two radiologists (B.S.K., E.K.F.) who did not interpret the MIP images and who were familiar with the VR software.
Renal arteries were categorized as normal; with mild (1%–49%), moderate (50%–69%), or severe (70%–99%) stenosis; or occluded. Stenosis was judged subjectively by the two readers for each modality. Weighted 
values were computed as a measure of interobserver agreement for the diagnosis of stenosis greater than 50% for each MIP and VR reader (SYSTAT 7.0; SPSS, Chicago, Ill).
Conventional angiograms were also evaluated independently and blindly by two angiographers (G.A.G., D.C.L.), who graded each vessel on the basis of percentage diameter reduction (0%–100%) at the point of maximum narrowing. The angiographers could use subjective assessment or calipers at their discretion. Since agreement between the two angiographers was excellent ( = 0.89), the average of their two readings was used as the standard of reference for grade of stenosis. Sensitivity and specificity were computed for each MIP and VR image for the detection of stenosis of at least 50% diameter reduction. Empirical receiver operating characteristic (ROC) curves were generated for each reader to demonstrate the relative accuracy of MIP and VR interpretations. To compare the areas under the ROC curves, parametric ROC models were constructed with the CORROC2 program (12). Comparison of sensitivity and specificity between MIP and VR was performed for all comparisons with the McNemar test for symmetry. A Bonfferoni correction was used to adjust for multiple comparisons (two modalities with two readers yields four comparisons), requiring a P value of .05/4 = .0125 for statistical significance (13).
|
RESULTS |
|---|
|
TOPAbstractIntroductionMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
Main Renal Arteries
Comparisons of VR and MIP images with conventional angiograms are summarized in Figures 1 and 2 and in the ROC curves in Figure 3. With the parametric ROC models, the area under the ROC curve for MIP reader 1 was 0.96, for MIP reader 2 was 0.98, and for VR reader 1 was 0.99. No statistically significant difference was found among these areas. A parametric model and area could not be calculated for VR reader 2, because the maximum likelihood estimates did not converge for this reader.
|
|
|
|
|
|
|
|
|
|
|
The main renal vessels that were under- or overestimated as having 50% stenosis with MIP and VR were categorized according to the collimation used during the helical acquisition. With VR, the one overestimated vessel was studied with 3-mm collimation; the two underestimated vessels were evaluated with 1- and 3-mm collimation, respectively. With stenosis greater than 70% as the cutoff point, four of the six vessels with moderate stenosis overestimated as severe were evaluated with 1-mm collimation. Three of the five vessels with underestimated stenosis were evaluated with 3-mm collimation.
Similarly with MIP, the vessels were also categorized according to collimation. Among the vessels with stenosis of at least 50%, seven of the 11 vessels with overestimated stenosis were studied with 3-mm collimation. Among the vessels miscategorized as having a stenosis of greater than 70%, seven were studied with 3-mm collimation and 10 were studied with 1-mm collimation.
The weighted value for the two MIP readers was 0.83 and for the two VR readers was 0.79. The McNemar test for symmetry in the detection of stenosis of at least 50% demonstrated no statistically significant differences in sensitivity between MIP and VR (P > .1 for all comparisons). Specificity was better with VR, however, and the differences for all comparisons between the VR readers and MIP reader 1 were significant (P = .008) and MIP reader 2 approached significance (P = .046 and .08).
Accessory Renal Arteries
At CT angiography, 14 accessory renal arteries were detected. Both algorithms had 100% sensitivity for the 11 accessory arteries depicted at conventional angiography. Three of the accessory renal arteries visualized with CT angiography were not visualized at conventional angiography (Fig 7). At conventional angiography, no accessory renal artery was characterized as having a stenosis of at least 50%; however, both MIP readers identified such stenosis in four accessory arteries (Fig 8). Only one of the VR readers identified such stenosis, also in four vessels. Two of the accessory renal arteries depicted with stenosis at CT angiography were not visualized at conventional angiography.
|
|
|
|
|
DISCUSSION |
|---|
|
TOPAbstractIntroductionMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
CT angiography provides direct visualization of the vasculature during the arterial phase of contrast material infusion. CT volume data can be viewed from an infinite number of projections, providing a clear advantage over conventional angiography for the characterization of eccentric stenoses and vascular calcifications. A number of authors have reported the accuracy of CT angiography for renal artery stenosis (1–4), and several studies have addressed optimization of data acquisition protocols and rendering algorithms (4,6,7).
Review of these studies reveals a number of shortcomings with MIP, the most widely used 3D rendering algorithm. Galanski et al (2) demonstrated that multiplanar reconstruction images and axial images were more accurate than MIP images and shaded surface display images in the characterization of renal artery stenosis. Pitfalls of MIP images include obscuration of the renal arteries by opacified renal veins and nonvisualization of accessory renal arteries. The quality of MIP images is degraded by suboptimal vascular contrast enhancement. Limitations of shaded surface display include the inability to distinguish calcification from vessel lumen and the misgrading of stenosis based on threshold selection (15).
Rubin et al (3) reported that MIP images and shaded surface display images may not depict renal venous variants, and preliminary editing, which can require 5–50 minutes, carries the risk of excluding accessory renal arteries. Despite 92% sensitivity for depicting severe stenoses, the MIP algorithm overestimates stenoses, particularly in the setting of hyperattenuating eccentric calcification, unless multiple MIP images are viewed in a cine mode (4). In vitro phantom studies have also shown that coronal MIP images overestimate critical (>85%) stenosis (6).
The 3D algorithm, VR, has not been widely used owing to computational complexity. Unlike MIP images, the final VR display is a weighted sum of data from all voxels along lines projected through the data set, and it conveys 3D information. With VR, each voxel is classified according to constituent tissue types on the basis of Hounsfield units; this classification and the position of the voxel in the 3D volume are part of the "weighting" process. The anatomy is displayed with a gray scale based on Hounsfield units that enables distinction of calcification from vascular enhancement. Because the 3D relationships are maintained, the algorithm allows real-time editing with a clip plane. Vessels can be viewed from any orientation (Fig 4) without the need for preliminary editing (8–10). Overlapping vessels can be evaluated more easily with VR than with MIP. With the former, 3D models are rendered in less than 1 minute, and real-time interactive rendering is possible. With the latter and use of an older workstation in this study, approximately 20–30 minutes of postprocessing time were required to generate MIP images for each subject. At present, smaller, less expensive desktop workstations are capable of performing VR or MIP in real time, overcoming one of the greatest limitations of these techniques.
With VR, the rendering process still involves subjective decisions. To display the vasculature optimally, the transfer function is interactively modified in real time at the workstation. To our knowledge, clinical validation is lacking for the accuracy of this interactive rendering process for the grading of vascular stenoses, although a number of studies have described the technique (8–10). Findings in our study confirm that real-time interactive VR is accurate for the grading of renal artery stenosis. The improved specificity of the VR algorithm over that of the MIP algorithm stems from the percentage classifier. For the final display, the MIP algorithm selects only the voxel with the highest attenuation along a ray projected through the data set. As a result, voxels representing vasculature may be erroneously excluded, yielding an overestimation of stenosis. For the 3D image, the VR algorithm calculates a weighted sum of data from all the voxels along a ray projected through the data set. This enables volume-averaged voxels to be included in the final image and decreases the potential for overestimation of stenosis.
Because the VR model is rendered interactively, however, individual readers can vary in their display and interpretation. The sensitivity for stenosis of at least 50% was 78% for one VR reader and was 100% for the other. Hyperattenuating calcification at the origin of a vessel limits visualization of the ostial region (Fig 4). The sensitivity, specificity, and accuracy for renal artery stenosis of at least 50% were high for VR, but the performance of both MIP and VR deteriorated for stenosis greater than 70%.
In the absence of surgical correlation, we cannot confirm whether the accessory renal arteries depicted at CT angiography but not at conventional angiography were true-positive cases. In retrospect, one of the cases appears to be a true-positive case that was misinterpreted at conventional angiography owing to overlapping visceral vasculature (Fig 6). Other authors have reported a similar discrepancy between findings at conventional and CT angiography (16). Explanations include the better low-contrast discrimination with CT angiography, which facilitates the distinction of small vessels, and the ability to view the vasculature from any orientation, which enables visualization of accessory renal arteries situated directly anterior to main renal arteries. In addition, anteriorly positioned accessory renal arteries may not opacify reliably at conventional angiography since intraarterial contrast material may layer into the dependent portion of the aorta. Similarly, unless oblique or lateral images are obtained at conventional angiography, a stenosis at the origin of an anterior accessory renal artery may not be depicted (Fig 7). In fact, the stenoses in accessory renal arteries that were overestimated at CT angiography in this series may not be false-positive cases.
At this time, decisions regarding angioplasty are based on appearance at conventional angiography, which remains the standard of reference. As acquisition and display technologies improve, the ability to evaluate vessels from any orientation and the improved low-contrast discrimination may result in more accurate evaluation with CT angiography than with conventional angiography. A large series with surgical correlation would be required to validate this. Alternatively, an assessment of patient outcomes after CT angiography to evaluate renal artery stenosis would be useful. A future study might compare the rate of cure for renovascular hypertension in patients screened with CT angiography versus those screened with conventional angiography.
When vessels were categorized according to collimation, two trends were noted. With the VR algorithm, most overestimations of stenosis of at least 70% were made with 1-mm collimation. With the MIP algorithm, most overestimations of stenosis of at least 70% were made with 1-mm collimation. This suggests that with both MIP and VR, use of 1-mm collimation results in more overestimations of severe stenosis. Findings in studies have shown that the accuracy of MIP correlates with the collimation (6,7); however, the small number of patients in this series provides insufficient power for a formal statistical analysis of this trend.
This study has some limitations. The number of patients is small. Conventional angiography, used as the standard of reference, was not performed specifically to evaluate the renal arteries in all cases. The collimation used for the CT angiographic renderings varied between 1 and 3 mm.
CT angiography is an accurate technique for the diagnosis of renal artery stenosis. In this series, CT angiography with VR enabled higher specificity than did CT angiography with MIP in the evaluation of renal artery stenosis. The interpretation is performed more rapidly due to the ability to obtain the VR model in real time without preliminary editing and to interpret the model without comparison to axial images. Our study represents a pilot investigation. Larger series are needed to demonstrate clinical applications of VR.
|
Footnotes |
|---|
Author contributions: Guarantor of integrity of entire study, P.T.J.; study concepts and design, P.T.J., E.J.H., E.K.F.; definition of intellectual content, P.T.J., E.J.H.; literature research, P.T.J.; clinical studies, E.J.H.; data acquisition, all authors; data analysis, P.T.J., E.J.H.; statistical analysis, P.T.J., E.J.H.; manuscript preparation, P.T.J., E.J.H.; manuscript editing, P.T.J., E.J.H., B.S.K., R.J.W., L.N.N., E.K.F.
|
References |
|---|
|
TOPAbstractIntroductionMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  F. J. Parrish Volume CT: State-of-the-Art Reporting Am. J. Roentgenol., September 1, 2007; 189(3): 528 - 534. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D.Y. Yoon, K.J. Lim, C.S. Choi, B.M. Cho, S.M. Oh, and S.K. Chang Detection and Characterization of Intracranial Aneurysms with 16-Channel Multidetector Row CT Angiography: A Prospective Comparison of Volume-Rendered Images and Digital Subtraction Angiography AJNR Am. J. Neuroradiol., January 1, 2007; 28(1): 60 - 67. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. K. Fishman, D. R. Ney, D. G. Heath, F. M. Corl, K. M. Horton, and P. T. Johnson Volume rendering versus maximum intensity projection in CT angiography: what works best, when, and why. RadioGraphics, May 1, 2006; 26(3): 905 - 922. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D.Y. Yoon, C.S. Choi, K.H. Kim, and B.-M. Cho Multidetector-Row CT Angiography of Cerebral Vasospasm after Aneurysmal Subarachnoid Hemorrhage: Comparison of Volume-Rendered Images and Digital Subtraction Angiography. AJNR Am. J. Neuroradiol., February 1, 2006; 27(2): 370 - 377. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. M. Cheung, J. Hegarty, and P. A. Kalra Dilemmas in the management of renal artery stenosis Br. Med. Bull., September 7, 2005; 73-74(1): 35 - 55. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. K. Kim, J. H. Kim, S.-J. Bae, and K.-S. Cho CT Angiography for Evaluation of Living Renal Donors: Comparison of Four Reconstruction Methods Am. J. Roentgenol., August 1, 2004; 183(2): 471 - 477. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. W. Olin, J. A. Kaufman, D. A. Bluemke, R. O. Bonow, M. D. Gerhard, M. R. Jaff, G. D. Rubin, and W. Hall Atherosclerotic Vascular Disease Conference: Writing Group IV: Imaging Circulation, June 1, 2004; 109(21): 2626 - 2633. [Full Text] [PDF]
|
|
|
|
|
|
S. Sheth and E. K. Fishman Multi-Detector Row CT of the Kidneys and Urinary Tract: Techniques and Applications in the Diagnosis of Benign Diseases RadioGraphics, March 1, 2004; 24(2): e20 - e20. [Abstract] [Full Text]
|
|
|
|
 |
|
 J. K. Willmann, S. Wildermuth, T. Pfammatter, J. E. Roos, B. Seifert, P. R. Hilfiker, B. Marincek, and D. Weishaupt Aortoiliac and Renal Arteries: Prospective Intraindividual Comparison of Contrast-enhanced Three-dimensional MR Angiography and Multi-Detector Row CT Angiography Radiology, March 1, 2003; 226(3): 798 - 811. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Mallouhi, S. Felber, A. Chemelli, A. Dessl, A. Auer, M. Schocke, W. R. Jaschke, and P. Waldenberger Detection and Characterization of Intracranial Aneurysms with MR Angiography: Comparison of Volume-Rendering and Maximum-Intensity-Projection Algorithms Am. J. Roentgenol., January 1, 2003; 180(1): 55 - 64. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Mallouhi, M. Rieger, B. Czermak, M. C. Freund, P. Waldenberger, and W. R. Jaschke Volume-Rendered Multidetector CT Angiography: Noninvasive Follow-Up of Patients Treated with Renal Artery Stents Am. J. Roentgenol., January 1, 2003; 180(1): 233 - 239. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. H. Rundback, D. Sacks, K. C. Kent, C. Cooper, D. Jones, T. Murphy, K. Rosenfield, C. White, M. Bettmann, S. Cortell, et al. Guidelines for the Reporting of Renal Artery Revascularization in Clinical Trials Circulation, September 17, 2002; 106(12): 1572 - 1585. [Full Text] [PDF]
|
|
|
|
|
|
D. D. Cody AAPM/RSNA Physics Tutorial for Residents: Topics in CT: Image Processing in CT RadioGraphics, September 1, 2002; 22(5): 1255 - 1268. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. M. Horton and E. K. Fishman Volume-rendered 3D CT of the Mesenteric Vasculature: Normal Anatomy, Anatomic Variants, and Pathologic Conditions RadioGraphics, January 1, 2002; 22(1): 161 - 172. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 U J Schoepf, C R Becker, R D Bruening, B M Ohnesorge, A Huber, L-G Haw, H Hildebrandt, and M F Reiser Multislice CT angiography Imaging, December 15, 2001; 13(5): 357 - 365. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. F. Tomandl, P. Hastreiter, C. Rezk-Salama, K. Engel, T. Ertl, W. J. Huk, R. Naraghi, O. Ganslandt, C. Nimsky, and K. E. W. Eberhardt Local and Remote Visualization Techniques for Interactive Direct Volume Rendering in Neuroradiology RadioGraphics, November 1, 2001; 21(6): 1561 - 1572. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. P. Lawler and E. K. Fishman Multi-Detector Row CT of Thoracic Disease with Emphasis on 3D Volume Rendering and CT Angiography RadioGraphics, September 1, 2001; 21(5): 1257 - 1273. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Macari, G. M. Israel, P. Berman, M. Lisi, A. J. Tolia, M. Adelman, and A. J. Megibow Infrarenal Abdominal Aortic Aneurysms at Multi-Detector Row CT Angiography: Intravascular Enhancement without a Timing Acquisition Radiology, August 1, 2001; 220(2): 519 - 523. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. A. Urban, L. E. Ratner, and E. K. Fishman Three-dimensional Volume-rendered CT Angiography of the Renal Arteries and Veins: Normal Anatomy, Variants, and Clinical Applications RadioGraphics, March 1, 2001; 21(2): 373 - 386. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Kawashima, C. M. Sandler, R. D. Ernst, E. P. Tamm, S. M. Goldman, and E. K. Fishman CT Evaluation of Renovascular Disease RadioGraphics, September 1, 2000; 20(5): 1321 - 1340. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. H. Frager Invited Commentary RadioGraphics, September 1, 2000; 20(5): 1353 - 1354. [Full Text] [PDF]
|
|
|
|
|
|
E. J. Halpern, D. G. Mitchell, R. J. Wechsler, E. K. Outwater, M. J. Moritz, and G. A. Wilson Preoperative Evaluation of Living Renal Donors: Comparison of CT Angiography and MR Angiography Radiology, August 1, 2000; 216(2): 434 - 439. [Abstract] [Full Text]
|
|
|
|
), 0.98 for MIP reader 2 (
), and 0.99 for VR reader 1 (
) (differences not significant). A parametric model and area under the ROC curve could not be calculated for VR reader 2 (
).
