Grid computing makes most of finite resources

April 8, 2004

If you haven't heard the term "grid computing" yet, you undoubtedly will hear it soon; it's one of the hottest and more controversial topics in computer circles today. It refers to the use of multiple computers, typically in different locations, as a

If you haven't heard the term "grid computing" yet, you undoubtedly will hear it soon; it's one of the hottest and more controversial topics in computer circles today. It refers to the use of multiple computers, typically in different locations, as a unified resource to perform one or more tasks.

The term was probably selected to underscore similarities between the computer grid and the power grid, which represents a ubiquitous utility that provides easy, reliable access to electricity from multiple sources. Other related terms are global computing, metacomputing, peer-to-peer computing, and scalable computing.

As the complexity of our computing projects increases exponentially, computing power, rather than communication bandwidth or storage, is more often the rate-limiting resource. Grid computing takes advantage of the most plentiful and least expensive resource-bandwidth-and trades it for somewhat more costly archival space as well as relatively expensive and precious computing power. This is achieved by multiple computers donating processing cycles and storage capacity as a coordinated resource to solve a task.

The anatomy of the grid has been described as consisting of six layers. The bottom layer, referred to as the grid fabric, consists of the resources available to the grid: PCs, networks, software, databases, and other devices. Above this level is a security infrastructure; single sign-on, for example, is critical to the success of a grid. Above that are two middleware levels referred to as the core level and the user level, which handle job submission, storage access, accounting and tracking, resource management, and scheduling. The next level up consists of special grid tools and programming aids, including software that breaks up a large task into many smaller ones that can be performed in parallel. The top level represents commercial, engineering, and scientific applications.

As the grid matures, standards are beginning to emerge from organizations such as the Global Grid Forum and the Globus Alliance. The latter is a research and development project designed to make it easier for scientists and engineers to take advantage of grid computing. The Globus Alliance consortium has developed the Globus toolkit, which provides users with security infrastructure, middleware, and programming aids (the top four layers of the grid architecture). The toolkit has helped solve a variety of problems in engineering, mathematics, modeling, and scientific applications.

A recent convergence of grid technologies and Web services has resulted in a proposal for standards for grid services.

USING PACS

The medical imaging community has not taken advantage of grid computing for a number of reasons, including the general lack of applications requiring massive computational power and concerns about security. In addition, the most computationally intensive tasks, such as image segmentation and 3D display, typically need to be performed interactively or in real-time. The potential may exist, however, to utilize the large number of computers available on a PACS for some types of applications.

PACS generally use less than 5% of the computing power available from PCs on the network. Computer-aided diagnosis programs, for example, could potentially take advantage of grid computing for applications such as lung nodule detection on a CT or general radiographic study and detection of lesions or microcalcifications on a mammogram. Advanced image processing used in iterative techniques in nuclear medicine and even in general radiography could also potentially benefit from the power available in a grid of PACS computers.

Although grid computing itself may not have a major impact on radiology, the concept of a grid can be applied to noncomputer-related aspects of radiology, including the image interpretation process itself. The ability to create a patchwork grid of PACS and teleradiology services may ultimately be necessary to weather what Dr. David Levin has referred to as the "perfect storm" in radiology: the crisis precipitated by the shortage of radiologists amid an ever-rising tide of imaging studies. In a manner analogous to the mismatch in the growth of computer communication speed and computing power, the number and productivity of radiologists (interpretation power) is increasing at a much slower rate than our ability to rapidly and inexpensively send and store diagnostic images.

This mismatch has provided a major impetus for the transition from film-based to filmless departments. In a film-based department, radiologist interpretation power (RIP) was often squandered while radiologists waited for films to be brought from the film room, for a film alternator to be hung, and for clinical colleagues to finish looking at studies at their reading station.

PACS has made radiology departments more efficient and has substantially increased RIP per radiologist by using relatively inexpensive centralized storage and fast networks to eliminate delays related to film and radiologist transportation. It has also reduced the number of interruptions caused by conflicts for a single film.

Many facilities, including the Baltimore VA Medical Center, have seen further increases in efficiency when multiple hospitals or outpatient centers share a single PACS. A single nighthawk radiologist, for example, can provide evening and nighttime coverage for multiple imaging departments, working at a high level of efficiency.

Arguably, a subspecialist such as a neuroradiologist reading only neuroradiology cases is more effective in terms of quality and speed than a general radiologist. On the other hand, a general radiologist would be more efficient and possibly more accurate reading non-neuroradiology cases.

This increased efficiency made possible by a shared PACS or teleradiology system raises the question of whether a meta-PACS or grid of connected systems or radiologists could provide further improvements in efficiency and quality of care. A large, geographically distributed radiology interpretation grid could possibly offer deeper subspecialty expertise, as well as improved nighttime, weekend, and holiday coverage, taking advantage of multiple time zones. Additionally, a greater number of radiologists would become available to provide peer review, expert subspecialty coverage, and quality control.

Large networks of hospitals and outpatient clinics such as the VA, Kaiser Permanente, and the Department of Defense medical facilities already have much of the infrastructure in place to make use of shared radiology resources and other clinical and management functions within a grid, including a master patient index, communications, and security policies.

Dr. Anna Chacko and her colleagues in the Army set up an innovative "virtual teleradiology environment" in the late 1990s that was designed to take advantage of the radiology grid concept. The Department of Veterans Affairs has also discussed the possibility of using this approach to address its shortage of radiologists.

ONLINE RESOURCES

The increasingly severe shortfall of academic radiologists has made it difficult for many departments to provide the comprehensive training residents and fellows require for board review and clinical practice. A growing number of trainees are turning to online resources such as Internet-based teaching files, Web-based textbooks, and online lectures and other presentations.

The RSNA's Medical Image Resource Center (http://mirc.rsna.org) has created a mechanism to link multiple teaching files and other educational resources and research repositories into a virtual community of connected libraries. This effort, which has been released for unrestricted free distribution and eventually as open-source software, may be used to create an educational grid of radiology resources that can be accessed by a single query at a local, departmental, or global level. Similarly, clinical trial grids could be created using this approach.

Dr. E. Stephen Amis, chair of the American College of Radiology board of chancellors, wrote in the October 2003 ACR Bulletin that "clever minds are finding ways to meet the demands for 24/7 coverage by beaming their digitized images via satellite to countries such as Israel and India for interpretation. . . . Well-rested radiologists in these countries can have day jobs interpreting studies performed at night in departments in the U.S."

The article also raised the question of whether the ACR should push the Centers for Medicare and Medicaid Services for a new policy on payment for studies performed outside the U.S. Amis has appointed Dr. Arl Van Moore to chair a task force on international teleradiology to address this and related issues.

POTENTIAL PITFALLS

Amis's article was entitled "Does this sword have two edges?" The use of grid radiology may also have a sharp edge that cuts both ways. We have major concerns about serious pitfalls that might be associated with grid radiology. They include general issues inherent to all grid computing, such as security and performance, and clinical issues such as patient privacy, credentialing and privileging, turf disputes, lack of in-person communication, and decreased appreciation of radiologists' importance in clinical care.

In addition, a widespread grid approach to medical imaging risks reducing the radiology interpretation process to a commodity. Just as energy has become a commodity, wholesale interpretation of medical imaging could be provided at a fixed price, and the purchaser of these services would be tempted to make cost the most important variable in vendor selection.

The fundamental qualities of professional service and interpretation accuracy could become marginalized as large customers seek out solutions solely on a cost basis. This could eventually backfire for radiology if entrepreneurs enter the grid marketplace, possibly providing professional services using radiologists of questionable credentials and skills, physician assistants, and nonradiologist physician groups. It is therefore imperative that organized radiology continue to insist on strict professional standards for interpretation of medical images, regardless of the mechanism of delivery of those services.

Computer viruses and worms have used a concept similar to grid computing to take over multiple computers' resources for a single unified purpose (for example, to launch a denial-of-service attack on the Microsoft Web site). Grid radiology, if not implemented correctly, may increase our vulnerability to attack and disruption of services. The massive failure of the electrical power grid in the Northeastern U.S. and parts of Canada on Aug. 14, 2003, should remind us how vulnerable a grid can be to catastrophic failure and how highly we depend on it. If the medical imaging community decides to plug into the concept of grid computing, it is essential that safeguards be implemented to protect against this side of the sword.

Editor's note: This is the first of two columns. Next month: Grid radiology in practice.

Dr. Siegel is chief of radiology and nuclear medicine and Dr. Reiner is director of radiology research, both at the VA Maryland Health Care System. Dr. Siddiqui is an imaging fellow at the VA Maryland Health Care System and a body imaging fellow at Geisinger Medical Center in Pennsylvania.

Dr. Reiner receives grants or research support from GE, Fuji, TeraRecon, and Stentor, and is a shareholder in GE.