PACS images can be treated byte by byte

January 12, 2005

The emphasis on image flow throughout the entire hospital supersedes operations within the image-generating departments. Image management outside the radiology department is not an afterthought, but rather an integral part in, and even a driving factor for, decisions that the department makes.

Our PACS project at the University Hospitals Leuven in Belgium radically follows from an overall IT perspective. The emphasis on image flow throughout the entire hospital supersedes operations within the image-generating departments. Image management outside the radiology department is not an afterthought, but rather an integral part in, and even a driving factor for, decisions that the department makes. Although images have specific characteristics and requirements, we often consider them yet another type of data or a collection of bytes.

This global perspective leads to solutions that would simply not be considered starting from the image side. The radiology department embraced this unorthodox view from the first experiments in PACS some 15 years ago. A few years ago, it teamed up with the IT department and Agfa as a commercial partner to realize a hospital-wide PACS.

We illustrate this approach using two different characteristics of the PACS. First, we completely delegated most of the image-related organization and workflow to the overall system, with image viewing and handling tightly integrated into that system. All clinicians retrieve and view images as an integral part of the electronic patient record (EPR). This extends to actions on the image exams, such as scheduling them for a case discussion or transmitting them to surgery planning workstations.

Second, images share a central storage system with most of the other hospital data. That storage system is unaware of the exact nature of the data. This abstraction provides economies of scale that enable us to hold all images online indefinitely.

This global approach not only resulted in a more functional PACS, but as a result, many traditionally difficult organizational aspects became easier.

INTEGRATING IMAGING HOSPITAL-WIDE

Cooperation between different subsystems such as the RIS, HIS, and EPR is often realized by exchanging information between them-for example, by using DICOM or HL7-while each subsystem works relatively independently. From our experience in implementing hospital-wide medical information systems, however, we have found the real problem to be harmonizing the working organization of the different subsystems. A subsystem such as PACS cannot provide the hospital-wide workflow we need, because it lacks the necessary information. Sending it the missing information is not a satisfying solution. That PACS would still have to duplicate internally some workflow functionality that took years to develop and tailor in other systems. Even worse, it is unlikely to implement that functionality in precisely the way we want it. In practice, we often find ourselves trying to overrule the organization imposed by some narrow-minded subsystem instead of getting benefits from it.

Our answer is not communication of more data. Instead, we radically assign specific functions to each system and enable one system to invoke functions from the other. It then becomes natural to concentrate overall organization into one system instead of dispersing or duplicating it over different systems.

In this view, the PAC Subsystem concentrates on image specifics. We implement overall organization instead in our in-house "active EPR," which includes aspects such as electronic requests and interdepartmental workflow. In this configuration, the PACS could be called a slave system, although the term symbiosis may be more appropriate.

An illustration of this approach in our hospital is clinical image access. We simply do not provide any means for our clinicians to retrieve images from a stand-alone PACS. Instead, we have tightly integrated the Agfa image viewer as a component into our EPR software. Physicians use that integrated system on 1500 PCs throughout the hospital: in outpatient consultation rooms, the wards, and case discussions. They select patients and exams in the overall system. The viewing component then displays the selected exam(s) and provides interactive viewing facilities for the images of that exam.

One advantage is ergonomic. Once the patient file is open, images can be displayed with one keystroke or mouse click, making an explicit search for the imaging exam unnecessary. But more important is the ability to exploit the complete context of the EPR and dedicated work lists to access information efficiently. The same argument holds for primary diagnosis. The RIS can better present organization-specific work lists or consider priorities than can a stand-alone PACS workstation.

The patient-based access control available in the EPR automatically extends to images, without need to duplicate access control functionality in the PACS. The latter was impossible to begin with: Not only are the rules too complex, the PACS does not have the information needed to establish whether a need-to-know link exists between a particular physician and patient.

We can relatively easily provide overall workflow support in our integrated system. That system knows at any time which image exam is the active one. We reuse our overall system to enable any physician to schedule an imaging exam for a particular case discussion. Physicians request generation of an image CD from the EPR, so we can reuse existing support mechanisms in that system. Similarly, any clinician can, from the overall software, initiate DICOM image transfer to a surgery planning workstation, for example. The actual image transfer is performed by the PACS. Having that action initiated from the overall system enables detailed access control and user-based selection of possible destinations. These organizational aspects were relatively easy to provide in the overall system, as overall organization is exactly what that system was designed for.

This strong integration improved efficiency for clinical image viewing. In a typical trauma case discussion, for example, more than 100 imaging exams are presented in 45 minutes. On an average working day, clinical users retrieve more than 5000 imaging exams. Approaching the problem from the overall information context instead of from the images has resulted in a more powerful image management system.

REUSING STORAGE INFRASTRUCTURE

PACS puts huge demands on storage and poses specific challenges in areas such as backup and disaster recovery. Storage demands for all kinds of applications have grown dramatically. Does that imply that PACS has to compete for expensive storage resources? We found that by looking at the data in the PACS as just another set of bytes, interesting possibilities for synergy become apparent.

It is less expensive to pool all storage needs than to provide separate storage systems. This applies not only to hardware costs but also to labor and expertise for building and managing the storage system and for keeping it up to date with technological evolution.

The lower marginal costs of a central storage system were a crucial argument in our decision to go for "all-online" storage for our PACS, with magnetic disks only. Another argument was the cost evolution of magnetic disks, combined with the fact that we can add disks gradually and thus exploit that cost evolution. The second argument holds to a large extent because of the economies of scale in the central system.

The consolidated storage system of our hospital provides 50 terabytes net, of which 25 TB are for the central PACS, and a few TB are for imaging applications outside that PACS. It is an NAS (network attached storage) configuration provided by Network Appliance. We hold images online indefinitely, despite a current growth of more than 8 TB a year (after compression but including two copies of lossy compressed images for clinical distribution), which is expected to increase due to advances in radiologic modalities and the inclusion of cardiology into that PACS.

That access to recent and historical images is equally fast is definitely an advantage. But that is a side effect of our decision to go for storage consolidation, more than a motivation.

Advantages that are more fundamental are management efficiency and flexibility at the level of the hospital. With prefetching gone, the need to tune the prefetching rules continuously has also disappeared. We can afford to spend more energy and expertise in backup and disaster recovery solutions for the central system (including for image-specific aspects) than we could for individual applications such as the PACS. Despite its huge data volume, PACS adds little management overhead to what we need for the other applications anyway. Project flexibility for imaging applications has increased because we can more easily react to opportunities by allocating more storage from the central pool. Upgrading the storage system may imply copying a few tens of millions of images during a few days, but the PACS software is oblivious to that.

Thus, the seemingly disrespectful view of images as "just" a collection of bytes managed on a general system enabled us to provide better service to these images.

REVERSED PERSPECTIVE ON IMAGES

We explicitly approached PACS from the global perspective instead of starting from the images. As a result, images have to share organization and even infrastructure with many other applications. Nevertheless, the result is a more functional PACS, more efficient access to diagnostic images, and a richer interaction between radiologists and clinical colleagues.

We believe that this approach will become even more relevant in the future. On one hand, hospital-wide information management will gain importance. But most of that is not specific for images, even if images are a bulkier data type. On the other hand, we must prepare for new ways to deal with the images themselves. How can we ensure efficient access to the essential information without drowning in the large amounts of images we will be able to generate? How can we integrate fundamentally new possibilities for image processing and analysis that are waiting to be released from the research labs?

In our approach, these two aspects are naturally separated. Hospital-wide information management is outside the PACS, leaving room inside PACS to excel in the image specifics. A crucial factor in this decoupling is an approach to integration that is not based on exchanging data but on subsystems providing functions to each other. Although technologically that is relatively common, it is not yet obvious commercially. Our commercial partners have proven that they can realize this kind of integration. It is our task to explore this vision of integration further in cooperation with them.

This more global perspective may seem to take away much of radiology's influence on PACS in favor of the IT department. But it has never been radiology's mission to provide the IT infrastructure. Radiologists are to provide the content, and they should embrace all synergy with the overall IT department that can make that content more accessible.

Radiologists may choose to focus on images in the basement, and PACS will help them do so. They can also opt to be partners and consultants to the clinicians and might even take up a role in new image processing and analysis services. PACS, when considered from the global perspective, can help them with that as well.

Mr. Bellon is PACS manager, Prof. Marchal is head of radiology, Mr. Feron is PACS senior developer, Mr. Vanautgaerden is chief technology officer, Mr. Reynders is front-end systems and storage manager, Mr. De Deurwaerder is network manager, and Dr. Van den Bosch is chief information officer, all in the radiology and information systems departments at University Hospitals Leuven in Belgium.