Diagnostic Imaging
August 2004

Cover Story

CT slices yield to 'computerized volumetric imaging'

Today is the day your new 16-slice CT scanner goes live. You've spent the last eight months researching, negotiating, and analyzing reports. Your 11-year-old single-slice spiral CT unit has served you well, but it has clearly outstayed its welcome.

Many new clinical applications await, including neurovascular and peripheral CT angiography, coronary artery calcium scoring, coronary artery CTA, CT colonography, advanced 3D imag-ing, high-resolution musculoskeletal imaging, and perfusion imaging for acute stroke. But you're planning to start slowly. The technologists will stick with basic CT studies-brain, chest, abdomen, and pelvis-but will expect much higher resolution than in the past.

When they've completed the first combined chest/abdomen/ pelvis CT study on the new scanner, they send the images to your read station and bring you the requisition. You call up the study to review. And that's when it hits you: 800 slices!

As this begins to register, you wonder how you will handle 30 of these 800-slice studies daily. And worse, you've heard talk of 32-, 40-, and 64-slice scanners that can spit out studies with 4000 slices. Right about now an early retirement package begins to sound attractive.

The issue of data overload is a common subject at multislice CT meetings. The dimensions of this challenging subject include workflow changes, staffing modifications, archival needs, and inherent medicolegal matters. The solution to most of these issues may lie in the mindset change that must accompany this new technology's rollout. Sixteen-slice technology is moving us away from 2D "slices" and into the third dimension: 3D data, analysis, and display. The sooner we in the radiology community embrace this shift to computerized volumetric imaging, the sooner the pieces of the puzzle will fall into place.

HISTORICAL PERSPECTIVE

Three-D imaging was introduced in the late 1980s as an adjunctive tool to cross-sectional CT. Musculoskeletal and craniofacial imaging were among the first general applications. The ability to generate 3D models of the hips, pelvis, spine, and other large joints enabled clinicians to view pathology in a manner more analogous to the way they would address that anatomic part (and the pathology therein) in the operating room. Much of the 3D imaging of the craniofacial area focused on the trauma patient.

Many early studies concluded that 3D imaging added little diagnostic value to the cross-sectional examination. Those studies did conclude, however, that 3D imaging generally aided in surgical planning and patient education and had value in congenital cranio-facial deformities and in craniosynostosis. But the tedious challenge of acquiring thin slices of 1 to 2 mm, combined with time-consuming postprocessing, precluded 3D imaging from playing a more widespread role.

The early days of 3D imaging thus relegated it to "bells-and-whistles" status: a technology introduced only on occasion to play only a complementary role to the primary cross-sectional study.

Multislice scanners introduced in the early 1990s could acquire volumetric data of a higher quality for subsequent postprocessing. Studies composed of thinner transverse sections obtained over larger body parts and in much shorter scan times would lead to the generation of substrates of volumetric data that were more pure and robust than those that could be achieved in the premultislice era. Thinner sections meant more seamless reformations; faster scan times meant less motion and misregistration artifact and better utilization of IV contrast material; larger volumetric coverage meant greater clinical application.

At the same time, significant computer hardware and software advancements were being introduced. More sophisticated and user-friendly postprocessing programs such as volume-rendering and fly-through capabilities complemented earlier 3D imaging options like shaded surface display and maximum intensity projection (MIP). Segmentation tools and preset postprocessing algorithms helped the proliferation of 3D imaging. Everything was coming together at the same time.

With increasing frequency, 3D imaging is moving away from its place as an adjunct to the cross-sectional study and becoming the primary goal. This evolution can be viewed in two ways. On the simpler level, 3D imaging is just part of the growth process from CT's beginnings in the 1970s. In a more global perspective, 3D imaging recalls a time when clinicians combined a posterior-anterior and lateral chest x-ray, for example, to coaxially localize a structure or disease process in 3D space. They did that for years, but with the advent of CT they were forced to abandon the PA and lateral for a set of axial slices. They embraced this challenge because the advantages of CT in spatial and contrast resolution overshadowed the drawbacks.

We now have the paradigm shift: We have the ability to display the pathology in the 3D model that is most analogous to what the surgeon will encounter in the operating room. In the early days of CT and 3D reformations, we had to go through the transverse axial step first, before generating the limited 3D images, because analysis of the transverse images was the more effective method of identifying pathology, and it was faster and easier. But that's no longer the case. Rapid postprocessing of extremely high quality volumetric data, using volume rendering and other techniques, affords us the ability to generate images that will demonstrate pathology to surgeons in the most meaningful way. The 3D postprocessing is becoming the primary goal, not the afterthought (Figure 1).

DATA, ANALYSIS, AND DISPLAY

The new 16-slice scanners have continued the evolutionary process even more dramatically. Data sets are composed of isotropic voxels and are acquired with unprecedented speed. Artifacts associated with motion, breathing, and other causes are eliminated. Viewing an anatomic region in the "native" transverse plane no longer has an advantage; the spatial and contrast resolution is preserved in all planes. The question of how to handle a chest/abdomen/pelvis CT study comprising 800 slices, which has plagued the radiology community for the last several years, is no longer relevant. The answer is in the false premise of the question: There are no slices.

When the data consist of isotropic voxels, and there is no image distortion in any plane, why should anyone want to go through the abdomen and pelvis in the old-fashioned transverse plane along the longitudinal axis of the body, beginning at the diaphragm and proceeding down to the symphysis pubis? The shortest axis is the coronal plane. A series of 50 coronal multiplanar reformation (MPR) images of 5-mm thickness seems a much more plausible way of interrogating the abdomen and pelvis than 2000 axial images (Figure 2 and illustration on page 24).

The basic concept is the collection of strong data that can be manipulated without concern that image distortion may be introduced during that manipulation process. The data themselves are only the raw material that leads to the generation of clinically meaningful images. We have been employing such concepts in everyday radiology practices for almost 15 years; it's called MR angiography. A routine MRA sequence is a collection of axial data that are often of limited diagnostic value as individual images, but which serve as the substrate for processing through MIP algorithms in generating clinically interpretable images. This is what isotropic CT imaging offers us.

Isotropic imaging is the fundamental pursuit of data comprising voxels that are of the same dimension in every plane (x, y, and z). They are essentially perfect cubes. When the data consist of such voxels, multiplanar and 3D reformatted images are free from image distortion. This opens up new vistas of image interpretation. Consider, for example, the geometry of the lumbar spine. The AP and lateral axes of the lumbar spine are much shorter than the longitudinal axis that runs from L1 to S1. Accordingly, high-resolution image review in the axial plane along the long axis from L1 to S1 might require review of hundreds of images, while review of the lumbar spine through a series of sagittal and coronal reformatted images would require inspection of 40 or 50 images, in total (Figure 3).

If the data are truly isotropic, it should be possible to identify a fracture, for example, as easily on the sagittal and coronal reformatted images as on the axial images, but with much less effort. Using the sagittal and coronal reformatted images as the primary diagnostic planes, rather than the axial plane, might take a little getting used to, but it would likely be worth the effort. Additionally, it parallels the AP and lateral radiograph more closely than the transverse axial images of conventional CT.

NEW PROTOCOLS

If the classical way to interrogate the abdomen/pelvis is through a systematic approach-liver/spleen in appropriate windows, then gallbladder/pancreas, followed by kidneys/ureters/bladder, then bowel, then vascular structures, bones, lung bases, and so on-why not have the computer process the data simultaneously in multiple ways, each of which is optimized to the particular organ system? For example, let the volumetric data be processed to produce a series of coronal MPRs through the liver/spleen in appropriate windows and to simultaneously produce a volume rendering of the kidneys, ureters, and bladder, a MIP or predefined volume rendering of the vascular structures, a virtual colonography of the colon, and a set of sagittal MPRs of the bones in a bone algorithm and bone windows. In other words, let the data be processed in many different ways right from the start, in preset algorithms, windows, projections, and postprocessing programs, each customized to the organ system (Figure 4).

The next step is to educate CT manufacturers about how all of this affects the radiologists' and technologists' workflow, and how they can introduce further innovations into protocol design to help streamline these processes. The entire concept of CT protocol design is changing; it used to include basic image parameters such as kV, mAs, pitch, slice thickness, slice reconstruction increment, reconstruction algorithm, and window/level settings. It was all about choosing the parameters for the scan itself. Now, however, as the process has become much more sophisticated, we need to think of the CT protocol as including three subcomponents: scan protocol, display protocol, and archive protocol.

The scan protocol would incorporate most of the parameters mentioned above that determine how to carry out the data acquisition phase of the study. The display protocol indicates how to display the images by choosing the planes, reconstruction algorithms, window/level settings, and postprocessing algorithms. And the archive protocol instructs the computer which portions of the data are to be uploaded to the PACS (Figure 5).

A simple example illustrates these ideas. In the trauma setting, CT imaging of the cervical spine is quickly replacing plain film as the primary diagnostic tool. As in any clinical scenario, defining the CT protocol requires a clear understanding of the imaging goals, which are multifaceted for cervical spine trauma. On one hand, the primary focus is identifying any fractures that might be present and ruling them out quickly and reliably when they are absent. But there is also an interest in interrogating the soft tissues for traumatic disc herniation or epidural hematoma, as well as a need to assess for subluxation, typically employing sagittal and coronal reformatted images. Finally, should a fracture be identified, there may be a role for 3D imaging for surgical planning.

If these are our imaging goals, let us define the display protocols to generate several series of multiple images, each customized to specific goals. Our 16-slice scanner has completed a study that has the potential of generating 300 1-mm images through the cervical spine, but we don't need to review all of them. Those 300 axial images were not generated on the premise that we can identify more fractures from 1-mm images than from 2.5-mm images; in fact, any fracture line that would be seen on only one 1-mm image would likely be dismissed as something other than a fracture. Rather, the 300 images were generated to provide a robust substrate for further processing.

Series one and two, for example, might process the data into sagittal and coronal reformatted images, using bone algorithm and bone windows, as the primary tool for evaluating fractures and alignment. Series three could consist of fused 5-mm sagittal or axial images using a soft-tissue reconstruction algorithm and soft-tissue windows. This series would then be customized for high soft-tissue contrast resolution for the eval-uation of potential disc herniations and epidural hematoma. Series four could be a 3D rendering of the spine in a 360º rotation for additional visualization of potential fractures. All these series are generated automatically and simultaneously by the scanner operator's console because this is what we established as the display protocol for trauma cervical spine CT studies (Figure 6).

NEW GAME

We, our patients, and our clinical colleagues have enjoyed the benefits of CT imaging for the last 30 years, but the last five years have seen a monumental evolutionary development into the world of 3D interrogation and diagnosis. As healthcare providers using this tool, we need to embrace the mindset change and allow our thinking to resonate with the technology. It's no longer computerized tomography, which implies slices. We are no longer dealing with slices. The new name of the game is computerized volumetric imaging.

Dr. Cinnamon is a radiologist with Quantum Radiology Northwest in Marietta, GA. He is a consultant to Vital Images and has been a consultant to Philips Medical Systems.