Until late in the 20th century, the available technologies for medical imaging produced a projection view, such as the standard chest radiograph or abdominal flat plate. With this technology, all information is collapsed along the thickness of the object being imaged. Advantages of this approach are simplicity and low cost; it is widely used to produce conventional chest, abdominal, and musculoskeletal films, for instance. Unfortunately, with projection images, it is difficult to determine what lies in front or back, so that multiple views are needed (eg, the posterior, anterior, and lateral views of a chest examination). Moreover, overlap of various tissues can obscure important pathology. In order to reduce the amount of tissue overlap, motion of the x-ray tube and film cassette during the x-ray exposure permitted the acquisition of thick section tomograms. This technique of conventional tomography was useful, for instance, to improve the conspicuity of small kidney stones.

Paving the way

Figure 1. Two views using a maximum intensity projection reconstruction from a contrast-enhanced MR angiogram of the renal arteries. Courtesy of Robert R. Edelman, MD.

Much more information was made available with the introduction of advanced tomographic imaging modalities such as computed tomography (CT) and magnetic resonance imaging (MRI). These modalities acquire the data as a series of individual slices or as a 3-D volume of tissue that is then subdivided into thin, contiguous slices. The use of thin slices rather than projections eliminates tissue overlap and also improves tissue contrast. For the vast majority of studies, simple inspection of the slice data is all that is required (eg, to interpret a brain CT or a knee MRI). If the data are acquired with sufficiently thin and contiguous or overlapping slices, then the data can be processed to create views along orientations different from the original acquisition. One such technique, known as multi-planar reformation (MPR), is helpful for a variety of clinical applications. For instance, multi-planar reconstructions are commonly used in conjunction with spinal CT in order to provide a long-axis view of the spine. This longitudinal view is particularly helpful for viewing disk protrusions on a CT myelogram or for depicting certain vertebral fractures. In addition to simple MPRs, several other 3-D imaging methods can be used to optimally display the data.

For instance, both MR and CT angiography commonly use a maximum intensity projection (MIP) algorithm to create angiogram-like images of blood vessels (Figure 1). The raw data consist of contiguous or overlapping images in which blood vessels are made to appear brighter than other tissues (because of the administration of a contrast medium, or because of flow-related enhancement). The MIP algorithm extracts the brightest (for MRI) or highest attenuation (for CT) pixel along a specified direction and puts this pixel value into a projection image. With a MIP display, one sees the blood vessels well but stationary background tissue is suppressed. Unlike a conventional x-ray angiogram, MR and CT angiograms can be rotated into multiple views in order to eliminate overlap and to obtain the optimal viewing angle.

A MIP image does not provide a sense of depth because it is essentially a collapsed 3-D structure onto a 2-D surface without perspective. Consequently, one cannot easily distinguish a vessel in front from one that is behind, except by examining multiple views. One alternative is to create two MIPs at slightly different projection angles, and view them as a stereoscopic pair. While feasible, viewing of the stereoscopic pair can be awkward. Alternatively, a surface rendering or surface-shaded display can be done. Apparent surfaces within the imaged volume can be determined based on user-selected thresholds or automatically, and mathematical models (eg, using polygon meshes) of the surfaces are generated. The images are viewed as if they were illuminated by a virtual light source, creating a profound illusion of perspective.

Volume Rendering

Volume rendering generates realistic presentations of the object over all or a portion of its thickness (Figure 2). With volume rendering, all pixels along a line of sight are visible to a greater or lesser degree, depending on preset or operator-selected opacity levels. The translucent nature of the presentation makes it possible to distinguish the anatomic relationships of various tissues. For instance, a neurosurgeon might use a volume rendering derived from a 3-D MRI study so as to depict the relationship of a tumor to vital brain structures and blood vessels. He can then overlay information from a functional MRI study (Figure 3). Finally, he can virtually cut away skull and brain tissue, thereby better visualizing the operative approach to deeper brain structures. One can also virtually remove tumor or infarct volume, a process done automatically when pathological tissue has distinctive signal intensity or signature. Volumetric measurements such as size, distance, and angle, for instance, may provide critical information in surgical planning. The use of volume data also makes it possible to quantitatively monitor changes in anatomy over time since it facilitates spatial registration of volumes acquired at different time points.

Compared with surface rendering, volume rendering manipulates much more data, so that greater computing power is needed. On the other hand, surface rendering may require more editing by the user, and it may be difficult to define the surfaces when the data are noisy or lacking in contrast.

Instead of viewing the object from the outside, one can apply virtual imaging technology to display hollow structures, such as blood vessels, the tracheobronchial tree, and the bowel, from the inside. The resultant views are similar to those that one would obtain with an invasive endoscopic procedure, hence the name virtual endoscopy. With the assistance of a fast-array processor, one can fly through the inside of the structures as if one were actually viewing them in real time through the eyepiece of the endoscope. Virtual endoscopy has proven of particular interest in conjunction with the technique of CT colonoscopy; it can, for instance, be used to distinguish a polyp from a fold, which may be less obvious on the axial 2-D images.

Although CT colonoscopy is still primarily a research tool, the clinical potential is enormous. For instance, a small percentage of invasive colonoscopy studies are unsuccessful because of the presence of a highly redundant colon or due to other technical factors. In addition, some patients have obstructive tumors that preclude passage of the endoscope. Since CT colonoscopy is unencumbered by such physical limitations, it provides a rapid, accurate screening examination of the entire colon.

Practical Considerations

Figure 2. Volume rendering from same data set as Figure 1 provides better sense of depth. Courtesy of Robert R. Edelman, MD.

The practicality of applying three-dimensional image processing methods depends on the available software and hardware, 3-D image processing technique, and on whether trained personnel are available to process the data. The speed and quality of image processing software vary widely depending on the particular vendor. In some cases, the image processing software is designed into the MR or CT console, whereas in others a separate image processing workstation is used. Although a separate workstation entails additional expense, it may be justified if there are high volumes of studies requiring complex image processing (for example, virtual endoscopy, CT angiography), since the clinical work flow may otherwise be impaired.

The MIP is a simple, fast procedure to perform. A series of MIPs from a contrast-enhanced MR angiogram can be generated in just a few seconds with minimal operator intervention. On the other hand, the creation of a virtual endoscopic study (for CT colonoscopy) requires many minutes of semiautomated processing. In a busy clinical practice, it may be difficult to free up a technologist or radiologist for a sufficient time to make this practical or cost-effective. In recognition of the additional effort involved, reimbursement is now available for the performance of 3-D reconstructions with CT or MRI under CPT code 76375.

The use of 3-D data acquisition and image processing is not restricted to CT and MRI data. Two-dimensional ultrasound imaging data can be merged, and then processed into different orientations that would be impossible to acquire with conventional scanning. Orthogonal planes can be created from data acquired in oblique orientations.

Ultrasound Methods

Figure 3. Volume rendering derived from 3-d MRI data, with cutaway of a portion of the scalp and brain. activiation of the motor cortex is shown in red. Courtesy of Robert R. Edelman, MD.

Three-dimensional ultrasound is becoming standard equipment on new machines. More powerful computers allow near real-time production of 3-D images, making the technique easy and convenient. There are two different and distinct methods used to create these images. One approach uses a module that can be added to a conventional ultrasound system. The module tracks the ultrasound probe’s location as each 2-D frame is acquired to create a volume element. The sonographer must maintain a constant scanning sweep speed so that the machine can use the constant beam thickness and an estimate of the distance traveled to create volume elements from the two-dimensional data. Three-dimensional ultrasound modules frequently cost less than $15,000.

Another method of creating three-dimensional ultrasound images is to use a special unit designed specifically for 3-D ultrasound. This machine can electronically steer the ultrasound beam from a stationary position to acquire the data without moving the probe. This eliminates some of the operator dependency in acquiring the data. These machines typically cost more than $100,000.

Three-dimensional ultrasound is proving to have clinical value in several areas. The technique is helpful in early pregnancies to demonstrate the relationship between fibroids and early gestations, and uterine abnormalities such as bicornuate uterus may be easier to identify. Second and third trimester pregnancies can benefit from 3-D ultrasound due to better visualization of the fetal face (eg, to show cleft palate) and the spine. Moreover, parents are usually delighted to be presented with a 3-D picture of their baby’s face. Aortic aneurysms, renal transplants, infant hip dysplasia, and evaluation of the carotid bifurcation may all achieve greater diagnostic certainty when utilizing 3-D ultrasound. Unfortunately, there is no increased reimbursement at the present time for 3-D ultrasound despite the additional equipment costs.

Conclusion

In conclusion, 3-D image processing methods have proved beneficial in several clinical applications, ranging from the improved depiction of spinal anatomy, to detection of vascular disease, to planning of operative interventions in the brain. Some approaches, like MIP, are in widespread use whereas the utility of others, like virtual endoscopy, is still being evaluated. As the software and hardware continue to improve, the use of these methods will become more and more routine. n

?Suggested Readings

?Calhoun PS, Kuszyk BS, Heath DG, Carley JC, Fishman EK. Three-dimensional volume rendering of spiral CT data: theory and method. Radiographics. 1999;19:745-764.

?Davis CP, Ladd ME, Romanowski BJ, Wildermuth S, Knoplioch JF, Debatin JF. Human aorta: preliminary results with virtual endoscopy based on three-dimensional MR imaging data sets. Radiology. 1996; 199:37-40.

?Garjian KV, Pretorius DH, Budorick NE, Cantrell CJ, Johnson DD, Nelson TR. Fetal skeletal dysplasia: three-dimensional US-

initial experience. Radiology. 2000;214:717-23.

?Johnson DD, Pretorius DH, Budorick NE, et al. Fetal lip and primary palate: three-dimensional versus two-dimensional US. Radiology. 2000;217:236-9.

?Johnson PT, Heath DG, Kuszyk BS, Fishman EK. CT angiography: thoracic vascular imaging with interactive volume rendering technique. J Comput Assist Tomogr. 1997;21:110-114.

?Keberle M, Jenett M, Beissert M, Jahns R, Haerten R, Hahn D. Three-dimensional power Doppler sonography in screening for carotid artery disease. J Chin Ultrasound. 2000;28(9):441-51.

?Kuszyk BS, Heath DG, Ney DR, et al. CT angiography with volume rendering: imaging findings. AJR. 1995;165:445-448.

?Luboldt W, Bauerfeind P, Wildermuth S, Marincek B, Fried M, Debatin JF. Colonic masses: detection with MR colonography. Radiology. 2000;216:383-388.

?Maymon R, Herman A, Ariely S, Dreazen E, Buckovsky I, Weinraub Z. Three-dimensional vaginal sonography in obstetrics and gynecology. Hum Reprod Update. 2000; 6:475-84.

?Ney DR, Kuhlman JE, Hruban RH, Ren H, Hutchins GM, Fishman EK. Three-dimensional CT-volumetric reconstruction and display of the bronchial tree. Invest Radiol. 1990;25:736-742.

?Rubin GD, Beaulieu CF, Argiro V, et al. Perspective volume rendering of CT and MR images: applications for endoscopic imaging. Radiology. 1996;199:321-330.

Robert R. Edelman, MD, is chairman