Three-dimensional (3D) data for displaying the coronary arteries in a noninvasive way can be acquired using contrast-enhanced electron beam CT or multidetector CT. Three-dimensional imaging of this data can be performed in different ways. Most commonly, surface rendering is used, but recent developments in hardware and software have facilitated the use of newer and more computationally expensive techniques such as volume rendering. In this article, we will describe the two rendering techniques, compare them visually, and demonstrate their ability for the visualization of the coronary arteries.

For this study, four data sets of human coronary arteries acquired using noninvasive contrast-enhanced electron beam CT were visualized using both volume rendering and surface rendering. Patients were randomly selected from an ongoing study on noninvasive imaging of the coronary arteries. Both institutional review board approval and informed consent were obtained. Data acquisition was performed on an electron beam CT scanner (Evolution XP; Imatron, San Francisco, CA). The acquisition of the 3D data set began with the injection of 120-180 mL of contrast medium (iopromide, 350 g/L I) at 3-4 mL/sec through an antecubital vein. Scanning commenced just proximal to the left main coronary artery after an ECG trigger at 80% of the R-R interval (diastole). Both slice thickness and table increment were set at 1.5 mm, resulting in contiguous, nonoverlapping slices. In all, 40-60 axial slices with a field of view of 18 cm were acquired during a single breath-hold. Surface renderings were constructed on a Magicview workstation (Siemens Medical Systems, Erlangen, Germany). Volume renderings were constructed using 3D rendering software (VoxelView; Vital Images, Plymouth, MN) running on special purpose graphics hardware (Octane; SGI, Mountain View, CA). A qualitative comparison of the resulting images was performed.

After the required segmentation, a qualitative comparison of surface rendering and volume rendering for visualization of the coronary artery tree was performed using the settings described previously (Figs. 2A,2B and 3A,3B). In one of the cases, the default settings, which were determined by qualitative comparison of several different settings, had to be adjusted to obtain the optimal result. The loss of detail in surface rendering as well as the rendering of only part of the data using a surface representation can be observed in Figure 2A,2B. For coronary artery bypass grafts to be visualized, the same imaging settings can be used (Fig. 3A,3B). Larger vessels (>3 mm) were well visualized using both techniques, whereas smaller vessels (<2 mm) were not adequately visualized by either technique.

When the coronary arteries are evaluated, visualization of calcified plaque depositions plays an important role. Visualization of these calcified plaques can be performed easily using volume rendering by slightly changing the opacity curve. The calcified plaques (that have a much higher voxel value > 250 H) become visible when a lower opacity is assigned to those voxels containing contrast-enhanced blood (100-250 H), and the relationship of the calcified plaques to the vessel and the possible stenotic regions can be evaluated (Fig. 4A,4B).

In the only patient with an intracoronary stent, surface rendering did not allow differentiation between the vessel lumen and the coronary artery wall stent. Usually, the attenuation of stents is comparable to a densely calcified plaque, and thus the same settings can be used for visualization of the calcified plaque. Careful fine-tuning of the opacity settings allows the visualization of the stent in more detail with some stent types (Fig. 5A,5B).

In all four data sets evaluated by visual comparison, volume rendering produced higher quality images compared with surface rendering. In the case of disagreement between surface rendering and volume rendering concerning a possible stenotic region, stenosis was verified on conventional catheter angiography. Volume rendering proved to be correct in all cases of disagreement, and surface rendering showed one false-positive stenosis and one false-negative stenosis (Fig. 2A,2B). No disagreement was found for the coronary artery bypass graft (one patient).

A direct comparison of surface rendering and volume rendering has been conducted for CT of several abnormalities [3,4,5], resulting in a number of advantages and disadvantages for both techniques.

The advantages of surface rendering are fast and interactive manipulation after the determination of the surface representation, good depth perspective and unambiguous 3D images using a virtual light source, accurate clinical measurement of structures because of the distinct surface definitions, simple segmentation using multiple thresholds, and assignment of specific attributes to different thresholds (e.g., different colors).

In addition to the advantages, the following disadvantages are inherent to the surface-rendering algorithm. First, surface rendering uses only a small portion of the available data. Second, determination of the surface representation is time-consuming and has to be performed each time a threshold is changed. Because of this, finding the optimal threshold becomes a difficult and laborious task. Third, surface selection is not adequate in structures without a well-differentiated surface. Fourth, because of the approximation of the surface, possible false-positive and false-negative surface elements can be introduced, and small detail can be lost. Finally, internal structures cannot be seen in combination with the surrounding structure.

Advantages of volume rendering are that all available data can be used, partial rendering is possible (small surface detail is better preserved than when using surface rendering), transparency can be assigned to view internal structures or structures in the background, and three-dimensionally unambiguous images are obtained with good depth cues.

Some disadvantages should also be mentioned. First, volume rendering is computationally expensive, which could result in less interactive rendering. Second, determination of the optimal settings is difficult because of the large number of parameters that can be set, possibly leading to overvisualization. When even slightly wrong settings are used, too much data can be visualized, thus obscuring the structure of interest. Finally, high transparency will decrease the depth information in the resulting image and thus make the 3D relationships difficult to observe.

Previous studies of vessels other than the coronary arteries [3,4] have shown that the percentage of arteries visualized using volume rendering is significantly higher (89-96%) than the percentage of arteries visualized using surface rendering (52-70%). Arteries that have a diameter within the range of 2-3 mm were well seen using volume rendering and incompletely seen using surface rendering. Because of technical limitations, acquisition parameters, and (motion) artifacts, larger vessels were visualized with higher quality than were smaller vessels in both rendering techniques. Increased spatial resolution, as with multidetector CT, could increase in the image quality of smaller vessels. Furthermore, an increase in temporal resolution could decrease motion artifacts and thus facilitate visualization of smaller vessels. Decreased visualization when using surface rendering compared with volume rendering can be explained by the decreased overall quality of the images and the increase of artifacts when using surface rendering.

In conclusion, intrinsic problems associated with surface rendering cause a lower image quality when compared with volume rendering because volume rendering uses all the available data instead of using surface approximation. One main reason to use surface rendering instead of volume rendering is that the computation time for volume rendering is comparatively long. However, with modern workstations, this issue becomes less important because of the availability of faster and better hardware and software. Our experiments corroborate our hypothesis by showing that volume rendering produces images of higher quality than those produced with surface rendering for the visualization of the coronary artery tree.

Address correspondence to P.M.A. van Ooijen.