PACS takes on unfamiliar research role in cancer project

Surprisingly little advancement has occurred in the area of PACS research in digital imaging's first 12 years. Now that research is about to enter a growth phase, nourished by the National Cancer Institute's effort to band cancer centers together in a nationwide cancer bioinformatics grid called caBIG.

The voluntary caBIG network, also referred to as the Grid, creates an open-source, open-access cancer research web by connecting more than 500 researchers from 50 NCI-designated cancer centers and other organizations to enable sharing of imaging studies, data, and tools. The unprecedented project is part of the NCI's ambitious 10-year plan to eliminate cancer suffering and death.

The lack of a robust infrastructure for developing new cancer therapies and technologies means that many new drugs never make it into patient treatment plans. Developing and obtaining regulatory approval for a single new therapeutic drug is estimated to take 10 to 15 years and cost over $800 million. These fiscal barriers prompt the pharmaceutical industry to focus efforts primarily on billion-dollar markets at the expense of other potentially beneficial but less financially promising therapies.

THINKING CABIG

To help reach its 10-year goal, the NCI, which is part of the National Institutes of Health, has fostered a number of projects to address this research dilemma. Current NCI investments-such as the Cancer Genome Anatomy Project, Academic Public-Private Partnership Programs, the Specialized Programs of Research Excellence, Rapid Access to Intervention Development, Rapid Access to Preventive Intervention Development, and now the $20 million caBIG project-are all part of a strategy to optimize development and delivery of more effective molecularly targeted drugs and technologies to detect, diagnose, and treat cancer.

PACS and digital imaging will play a key, if unfamiliar, role. To date, PACS tend to be patient-centric, which is of enormous clinical benefit but makes scouring PACS archives for medical research data difficult if not impossible.

If personalized medicine is to become a reality within a decade, as the NCI envisions, it will require the creation of massive bioinformatics databases, including research access to PACS archives. This information will eventually allow researchers to tailor cancer therapy not just to a generic diagnosis but to specific aspects of a particular tumor.

"The only way to achieve what we think will be the next era of medicine is by paying better attention to informatics than we have in the past," said Dr. Eliot Siegel, a professor and vice chair of information systems at the University of Maryland and chief of radiology and nuclear medicine at the Maryland Health Care System.

Today, informatics data are saved for only about 2% of cancer patients who enroll in clinical trials. Potentially useful information is dying with the patients rather than being captured to add to the collective medical knowledge, Siegel said.

Medical image databases have largely been an isolated institutional activity to this point. Some medical groups, such as Partners Healthcare in Boston, have created their own localized databases containing hundreds of thousands of images. Partners' CIRAS (clinical image repository access system) drew its data from the Research Patient Data Registry, an accumulation of 450 million diagnoses, medications, procedures, reports, and laboratory values gathered over 25 years from 1.8 million patients treated at Massachusetts General or Brigham and Women's hospitals.

Until recently, however, there has been no national effort to connect these centers and their collections. Now, for the first time, such an attempt is under way.

REVOLUTION IN MEDICINE

Many medical experts see healthcare on the threshold of a new era of personalized medicine. One manifestation of this revolution will be that patients will routinely have their DNA sequenced, Siegel said.

"By then, we will have genomic, proteomic, and pharmacokinetic information that is specific to individual patients," he said. "Medicine will be much better able to do an array of targeted diagnostic studies that, hopefully, will allow us to predict what diseases patients are most likely to get."

Once a patient does develop cancer, the ultimate goal is to provide oncologists and other investigators with the tools and resources to query the Grid to determine whether certain types of radiation or chemotherapy were effective depending on both the tumor's and the patient's genomic and proteomic makeup, Siegel said.

The caBIG project also treats a pervasive issue in basic cancer research: Even though steady advances in recent research have produced an abundance of new findings, medical researchers have had no direct method for sharing data and analysis methods. The potential power of research collections can be fully realized only by enabling individual investigators and research teams who are otherwise isolated to combine and leverage their findings across the entire cancer research community.

The caBIG initiative is now providing the necessary unifying infrastructure. It enables research teams to pursue new collaborative efforts while supporting the needs and contributions of individual investigations.

An example of the benefits of caBIG can be found at Harvard University and the Massachusetts Institute of Technology, where researchers are joining forces on the development of computational models that simulate cancer as complex dynamic biosystems across several scales in space and time. The effort requires input data from the experimental side on the molecular, microscopic, and macroscopic level, including digital imaging from a variety of modalities.

"Any infrastructure that promises to enable and facilitate standardized acquisition, archiving, sharing, and mining of such biomedical data and to support design, development, and exchange of in silico models generated from this type of data is tremendously helpful for us in accomplishing our long-term goal of predictive cancer modeling," said Dr. Thomas S. Deisboeck, director of the Complex Biosystems Modeling Laboratory at the Harvard-MIT Athinoula A. Martinos Center for Biomedical Imaging.

IMAGING WORKSPACE

The NCI has already established five caBIG workspaces: Clinical Trial Management Systems, Integrative Cancer Research, Tissue Banks and Pathology Tools, Architecture, and Vocabularies and Common Data Elements. Within each workspace, cancer centers will be funded to develop new data management applications and to adapt existing analytical research tools.

An In Vivo Imaging Archive (caIMAGE) imaging workspace was announced in April. According to Siegel, who has been asked to lead it, the mission of the imaging workspace is to advance the field of imaging informatics to extract meaning from in vivo imaging data, thereby improving outcomes for cancer or pre-cancer patients. The image archive could eventually store up to a million images from 7000 sites.

It is hoped that when the primary analysis of clinical trials is complete, most investigators will make their image archives and associated clinical and laboratory data available to the general public over the Internet via caIMAGE, in accordance with the NIH's data sharing policy, said Dr. Daniel C. Sullivan, associate director of the Cancer Imaging Program at the NCI. This will allow medical researchers to conduct a variety of secondary analyses, such as correlating certain image features with genomic, proteomic, or other laboratory or clinical data.

"We expect that many of these image archives will be particularly useful to developers of computer-assisted diagnosis algorithms and other types of image processing algorithms, such as automated methods for change analysis of lesions," Sullivan said.

Images from clinical trials collected in the In Vivo Imaging Archive will allow investigators to more easily perform quality assurance assessment, conduct blinded, independent readings or measurements of tumor response, and maintain a secure, auditable archive that will be acceptable to the FDA for drug approval purposes, he said.

"Overall response rate in cancer therapy trials has been shown to be more accurate when assessed from blind, independent reads than from unblinded reads by study investigators," Sullivan said. "Thus, cancer patients will benefit from more accurate results from cancer therapy trials."

IMAGING GOALS

The various protocols, standards, and guidelines that will be needed to define and implement the In Vivo Imaging Archive will largely be determined by the In Vivo Imaging Workspace, a group of experts who meet regularly by conference call and occasional in-person gatherings.

The near-term goals of the imaging workspace include seeding the clinical trial image archive; establishing standards for clinical trial image data collection, transmission, and archiving; and developing open-source software and nonproprietary workstations for data sharing, Siegel said. Longer term goals involve artificial intelligence-enhanced data mining and clinical decision support.

The imaging participants are initially expected to include 10 to 15 funded cancer centers and other, unfunded, participants from academia, professional organizations, nonprofit groups, federal agencies, and the imaging industry.

Participation is democratic, even if funding is not extravagant. About 15 academic institutions will receive partial funding for approximately one person-day-per-week. Nevertheless, organizers expect a large number to participate without funding.

The In Vivo Imaging Workspace is a public, inclusive activity, Siegel said. Everyone with related expertise and interest is invited and encouraged to participate.

"Those who are unfunded are invited to the table on equal footing; when there are votes and prioritizations, whether you are funded or not, you have an equal vote," he said.

RESEARCH ENTERPRISE

The first $2 million In Vivo Imaging Archive grant will be used to create a process wherein academia, government, and the imaging industry will attempt to find ways to cooperatively share imaging data throughout the research enterprise, Siegel said.

The idea is to have participants identify the major issues in promoting research and image interoperability. PACS, hospital and radiology information systems, and the electronic medical record are not currently designed to allow searches of the type needed in an era of personalized medicine.

"The biggest challenge is figuring out how one creates incentives for the major vendors to include research as a primary focus," he said.

Circumstances such as GE Healthcare's acquisition of Amersham and Siemens' emphasis on molecular imaging are helping to change the industry's thinking about the future of medicine. But the ability to effect those changes rapidly by persuading imaging equipment makers to emphasize research will be a challenge. The effort will in some ways resemble the struggle involved in persuading vendors to embrace the Integrating the Healthcare Enterprise initiative.

"We have to capture the imagination of both industry and academia in order to reengineer the way we organize our information systems," Siegel said.

REINVENTION OF INFORMATICS

The reinvention of imaging informatics promises to position radiology squarely in the realm of e-science. In this setting, scientific and medical research is increasingly conducted through distributed global collaborations enabled by the Internet, using large data collections and terascale computing resources. In the case of caBIG, the computing resource is the Grid.

Grid computing represents the latest technology to evolve from the familiar realm of parallel, peer-to-peer, and client-server models. Under caBIG, cancer centers will be connected through a test-bed architecture called caGRID 0.5, which contains tools for creating and deploying caBIG-compliant grid services.

"We are trying to create a virtual community of interconnected applications, tools, and individuals to redefine how research is conducted and how care is provided," said Sue Dubman, director of applications for the NCI's Center for Bioinformatics.

To achieve this goal, the NCI installs remote field center software at participating sites and acquires images coming from local PACS, which are then de-identified. The system uses anonymization routines from the RSNA's MIRC (Medical Image Resource Center) software customized for each site, since different facilities have different requirements for anonymization.

"We take images from the field center, de-identify them, then extract them to our repository," Dubman said.

Each field center also installs caBIG curation routines to validate image quality and ensure that image annotations are complete, accurate, and linked to the appropriate studies.

The NCI then provides a Web portal, the caBIG Clinical Protocols Portal, that allows researchers to share their clinical trial protocols among members of the caBIG community.

CONNECTING TO GRID

Dubman said caBIG creates a middleware layer that provides API (application program interface) access to images, a feature that also connects to the Grid so that researchers will be able to query individual PACS across the Grid in subsequent program phases.

What makes this particularly attractive is that limited integration between clinical and imaging systems presents one of the challenges in imaging today. By providing the API, caBIG can facilitate this integration.

"In effect, by using caBIG standards, we're supporting semantic interoperability so we can query across different modalities," Dubman said.

The Center for Image Processing and Integrated Computing (CIPIC) at the University of California, Davis is one site anxious to profit from caBIG standards. CIPIC researchers are developing statistical and bioinformatics tools to exploit the statistical properties of functional MRI and PET scans for cancer research.

"While there are several sites working on these tools, with caBIG we won't have to reinvent the wheel," said David Rocke, Ph.D., codirector of CIPIC.

If a tool is developed at UC Davis, it will fit into the caBIG infrastructure so that it can be used by any participating cancer center. Similarly, UC Davis will be able to use tools developed at Johns Hopkins or the University of Pittsburgh.

"This is potentially an enormous increase in the collective power of research," Rocke said.

Although medical researchers publish papers and distribute software, others may find it difficult to use their tools if they don't adhere to standards, he said.

"If you get somebody else's software to analyze images, and it's written for a different brand of imaging device than you're using, it may be proprietary software or it may be written in a language you don't use," Rocke said.

With caBIG, any software for analysis of imaging data that adheres to these standards will be usable by any participant.

"Standardization is a huge leap forward in our ability to share technical accomplishments and accelerate cancer research," he said. "We're all quite excited about it. This is a big opportunity to make all of us more effective."

Additional information about the caBIG workspaces and working groups is now available online at http://caBIG.nci.nih.gov. An interactive overview of caBIG and an inventory of tools can be found by selecting "caBIG Interactive Overview" from the "News/Events" sidebar.

caBIG Principles

- Open Source

Publicly funded development must yield openly distributable products.

- Open Development

Community-driven development aligns needs with development priorities.

- Open Access

Data have value beyond original purpose for collection. Scientific method demands verification

by peers. Obligation to share publicly funded data products.

- Federated

Local control of deployments. No central "ministry of information." Scalable.