Innovations in multislice CT have created a clinical dilemma for radiologists. The quantity of data produced during a single imaging exam has soared since 1990, and today a typical multidetector CT scan generates hundreds of reconstructed images. The data overload can be overwhelming to interpret, and it pushes clinical workstations to the limit.

Although radiologists could reconstruct fewer images, studies with single-slice CT scanners show that disease detection improves with at least a 50% overlap of cross sections. Thus, the challenge for radiologists is to maximize CT’s technical and clinical capabilities without taxing their workstations or their workload.

The problem is two-fold: how to effectively interpret the volumes of data that are now available, and how to store the deluge of image bytes for easy retrieval and archiving. Unfortunately, the shift to PACS has not solved this problem.

“When these scanners came out, people were still filming slices, and they realized that their film budgets were being blown out the door,” said Sandy Napel, Ph.D., an associate professor of radiology at Stanford University. “These scans have to be read electronically, because you can’t afford to print the film. But you are still going to page through thousands of slices.”

Addressing the issue requires better workstations, capable of providing efficient access to data, and new tools and techniques to assist radiologists in interpreting those data.

“Our greatest challenge with multidetector CT is how to display, interpret, record, and archive the data,” said Dr. Geoffrey Rubin, an associate professor of radiology at Stanford who collaborates with Napel on CT applications in cardiovascular imaging and the design of algorithms to interpret the modality’s data-rich images.

Software for creating automated curved planar reformatted images is one such effort at Stanford that is undergoing experimental review. Curved planar reformatting addresses the problem posed by cross-sectional modalities like CT in evaluating blood vessels, but to assess the status of a blood vessel on those images, one must view hundreds of them, either side by side or in sequence.

“Curved planar reformatting is a way of creating from cross-sectional images an image that looks more like a conventional angiogram, but without the problems of regular angiograms,” Napel said. “If you can find a way to do it efficiently and accurately, then the radiologist would look at that curved planar reformat and, if suspicious about a particular region, be able to retrieve the cross-sectional image corresponding to that spot.”

The challenge is to find a way to create curved planar reformatted images quickly and easily. The procedure can be executed on many conventional, commercially available workstations, but almost without exception the reformatted images must be created manually.

At Stanford, three full-time technologists review CT angiography images and mark the cross-sections of the blood vessels of interest. The marked scans are then read by the computer that generates the reformatted image, but the prep work on some cases can take up to 45 minutes.

Rubin and Napel have developed software and algorithms that automate this entire process. With such software, a radiologist would simply click on the two ends of a blood vessel to generate a curved planar reformat.

In an article submitted for publication in Radiology, Napel assesses automated versus manual curved planar reformatted images, using the algorithms and software developed by the Stanford team. He created images from six different arteries in three patients with aortic aneurysm. Four radiologists graded each image on its quality and guessed whether it was automated or manually reconstructed. The automated images not only read well, but they required only one-seventh of the time. Radiologists were unable to tell the difference between the two types of generation.

Their software, however, addresses just one clinical application among many, Napel said.

“Curved planar reformat is a solution to one specific niche problem in CT angiography,” Napel said. “In general, we have to solve the problem for every application of CT and other cross-sectional imaging devices.”

A host of other applications in CT could benefit from similar interpretive aids, he said. The liver, lung, and pancreas are regions of particular interest. Specifically, in the lung and colon, computer-aided detection (CAD) techniques could emerge as an important tool to help surf the surfeit of CT data.

“Without CAD, radiologists need to scrutinize every pixel on 500 to 700 images,” Napel said. “And they can’t do that with the kind of accuracy that is demanded. The appropriate tools, such as computer-aided exploration of the image volume, would help rule out or rule in the reason you scanned that patient in the first place.”

Raising The Bar

In some ways, CT is a victim of its own success. As the technology has advanced and resolution has increased, the diagnostic bar has been raised ever higher.

“Instead of looking for relatively large objects in a small number of slices, as we did 10 years ago, we’re looking for objects that are smaller because the resolution is better,” Napel said. “On top of that, we have hundreds of slices to view. So we still need solutions.”

The vendors jumping in to help, so far, are mainly the smaller companies specializing in software. But that is about to change.

“The major companies are beginning to join the smaller ones that have gone out on a limb with some of these new software technologies,” Napel said. “I think they are all motivated to provide a radiology solution that extends from patient to diagnosis. They recognize that a bottleneck exists between generating the images and providing the interpretation.”

Image Management

Interpreting images is just one side of the coin. The other is managing complex reconstructions and then archiving large volumes of generated images. The issue has become a headache verging on a migraine for PACS administrators like Mary Jo Olson, vice president of operations at St. Paul Radiology, a 75-member practice in Minnesota. The group practice has a network that links eight hospitals and five imaging centers.

“I asked PACS vendors at the last RSNA meeting how they were going to handle the management issue,” Olson said. “The scanners are creating 700 images per study. What are we going to do with all of these images? Nobody has an answer for that.”

The issue isn’t as pressing at the front end, however, as at the back end.

“The problem isn’t so great the first time around, when you are storing these images,” she said. “It’s later, when you call them up for comparison. The memory in the workstations fills up fast, and you can tie up your network moving these huge packets of information around.”

What’s needed are high-performance workstations that boast large storage capacity, fast processing boards to handle data, interactive real-time display boards that can show 20 to 30 frames per second, and fast Internet and intranet connections for distribution.

Vendors are beginning to look at that problem, too, but again it is the smaller companies that are stepping in with commercial products. Firms like TeraRecon and Hinnovation are tackling the problem of image management and distribution, while companies specializing in storage area networks (SANs) are tackling the archiving issue.

TeraRecon in San Mateo, CA, distributes a line of two-dimensional and three-dimensional Aquarius workstations, 3-D volume rendering accelerator boards, and 2-D and 3-D real-time reconstruction subsystems for optimizing multidetector CT and ultrasound systems. In May, the company introduced its AquariusNet server that supports streaming 2-D and 3-D image processing and review.

Expensive 3-D workstations may not integrate with existing clinical workflow. Rather than investing in more of them, a radiology department or imaging center could use AquariusNet to provide similar applications and functionality at the workstations used for soft-copy review.

“Modern imaging modalities are generating ever-increasing numbers of slices, requiring powerful workstations for review and manipulation,” said Robert Taylor, executive vice president of TeraRecon. “But workflow is disrupted when different workstations are used for 2-D review and 3-D postprocessing. It’s often impossible to gain quick, interactive access to 2-D and 3-D images from a physician’s desktop PC or consultation station.”

The company is marketing its server to radiology departments that seek the high processing power required for 3-D image reconstruction without having to also host the computing hardware. AquariusNet’s back-end server features support for image archiving and distribution as well as 3-D capability. In September, the company received FDA 510(k) clearance to market its 3-D workstation as a PACS product.

“Our server architecture opens up new possibilities, with the power to keep pace with modern multidetector CT scanners,” Taylor said. “We believe it will change the paradigm for PACS review and could redefine the role of stand-alone 3-D workstations.”

Flexible Storage

Hinnovation is moving in the same direction. At the Society for Computer Applications in Radiology meeting in May, it launched its iConnection 2-D software for distributing 3-D images among radiology departments and referring physicians. The company, based in Wauwatosa, WI, was started by a group of former GE Medical Systems scientists.

Actual reconstruction is performed on-site at workstations in the radiology department, but users at remote locations can view 3-D images from the central server over an Internet connection.

In the burgeoning 3-D imaging arena, these companies join a host of others, from Algotec to Vital Images. In addition to software and networking to support image management, other vendors are working on solutions for archive overload.

Storage area network (SAN) systems, designed to facilitate high-speed transfer and sharing of medical data and images without bogging down existing network infrastructure, could provide an answer to the problem of storing high-volume, data-rich images.

A SAN is a high-speed network that connects data storage devices and servers. In most installations, the SAN uses a fiber-optic telecommunications network for a dedicated server-to-server link. That link operates on the back end of a server independently of the local area network. Some SANs have been able to speed image transmission by more than 600% over a fast Ethernet.

Companies like Gadzoox Networks in San Jose, CA, have deployed SANs for governments, educational institutions, and e-commerce, and they are now turning to medicine. Sold through OEM companies including Hewlett-Packard, Compaq, and Data General, SAN hardware and software from Gadzoox can accommodate anything from chest radiographs (10 MB) to comprehensive MR angiograms (400 MB) in a matter of seconds.

Such high-speed networking can reduce image retrieval time to mere seconds, from the minutes or hours on a conventional PACS. By pooling devices on a SAN, any device on the network can be allocated as much or as little storage as it needs.

“SANs excel at applications that involve a lot of data, where the data are needed fast and have to be available around the clock,” said Spencer Sells, product marketing manager at Gadzoox.

Back To Front

Solving such back-end issues is important, particularly as radiologists begin working with data in more creative ways, and it is spurring R&D on ever more sophisticated, high-volume scanning technologies. An irony, however, is that the radiology community may have protested too much about data overload.

“Manufacturers are providing devices that give us exquisite resolution on huge numbers of images,” Napel said. “But what they are hearing from users is, ‘We can’t look at all these images.’ So the danger is that vendors may curtail development, and we will miss out on the next generation. Still, we want all the images we can get, we can do great things with them. In the future, radiology should see even greater things, in how fast we scan and how much we can do per unit of time.”