| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Editorials |
1 From the Biomedical Imaging Program, National Cancer Institute, Bethesda, Md (M.W.V.), and Clinical Center, Diagnostic Radiology Department, National Institutes of Health, Bethesda, Md (R.M.S.). Received December 8, 2002; accepted January 8, 2003. Address correspondence to M.W.V., Department of Radiology, University of Iowa Hospital and Clinics, 200 Hawkins Dr, 3966A JPP, Iowa City, IA 52242-1077 (e-mail: michael-vannier@uiowa.edu).
Index terms: Data analysis • Editorials • Images, storage and retrieval
Peer-reviewed archival publications are essential information resources for formal professional scientific communications in a complete and permanent form. Original work in radiology and the related sciences is published much the same way today that Roentgen’s was when he announced the discovery of x rays. Anyone can locate and access contributions to archival journals by using their bibliographic citations (1).
Today, it is increasingly common for libraries to be electronic repositories. The indexing schemes used in digital libraries permit searching for literature citations by using various descriptors, such as journal name and issue date, author names, or words in the title or abstract. Recent work published in the journal Radiology, for example, can be retrieved as an electronic version almost instantaneously by using any of these descriptors (2).
This is not true for images, however. In radiology, where imaging is central to everything we do, published images are neither indexed separately nor retrievable. To make matters worse, most authors decline to share their original source images, preferring to maintain them in private collections. It is impossible to reconstruct the results of published work, since the original source data (eg, images) are unavailable.
Sequestration of data in the imaging community is the antithesis of the situation in other related disciplines. Universities have established their own electronic repositories as a potential alternative to traditional publishing, partly in response to the high cost of academic journals (3,4). In molecular biology, for example, sharing of sequence and structural data is a requirement for publication in peer-reviewed journals. Authors must include an accession number in the article after they deposit the sequence or structure in a public database. The establishment of online publicly accessible scientific databases is a worldwide trend. Each January, the journal Nucleic Acids Research publishes a special issue on databases—347 of them in 2003 (5).
The publication policy of Science, a widely read weekly peer-reviewed journal published by the American Association for the Advancement of Science, illustrates the importance of data sharing in other disciplines (6):
As a condition of publication, authors must agree to honor any reasonable request for materials and methods necessary to verify the conclusions of experiments reported, and must also agree to make the data upon which the study rests available to the scientific community in some form for purposes of verification and replication. As a practical matter, for large data sets such as DNA and protein sequences and crystal structures, this generally means deposition of the data before publication in an approved public database, with the accession numbers provided for inclusion in the published paper. In selected cases, other repositories that allow free access to the data for purposes of verification and replication may be acceptable with the approval of the Editor-in-Chief.
According to Harold Varmus, a Nobel Prize winner and former director of the National Institutes of Health (NIH) (now president of Memorial Sloan-Kettering Cancer Center in New York, NY), "all modern biologists using genomic methods have become dependent on computer science to store, organize, search, manipulate and retrieve the new information. Thus biology has been revolutionized by genomic information and by the methods that permit useful access to it" (7).
NIH Data Sharing Policy
The NIH developed a statement on data sharing (effective February 26, 2003) to express the expectation and support of the timely release and sharing of final research data from NIH-supported studies for use by other researchers. Investigators who submit an NIH application will be required to include a plan for data sharing or to state why data sharing is not possible. This statement will apply to extramural scientists seeking grants, cooperative agreements, and contracts and will also apply to intramural investigators. This requirement includes clinical data, such as images, which are subject to confidentiality protection of individual subjects (8,9). The complexity of imaging data is a sufficient reason both for sharing and for not sharing primary data. Sharing of data should make research more efficient and should facilitate further investigations. The intellectual challenges are identical with or without data sharing. Sharing increases the value of the data and provides new knowledge and understanding (10).
There are a number of image-sharing projects underway in a variety of disciplines, including radiology, astronomy, and neuroscience. We will describe several of these projects to give a sense of the breadth of applications. These image archives also present possible avenues for implementation of a radiology image archive.
Medical Image Resource Center
The Radiological Society of North America has launched a project called the Medical Image Resource Center, or MIRC, to establish a community of Web-based libraries of imaging information, including teaching files, other educational materials, and research data. This system enables sharing of such materials with a Web-based interface (11,12). MIRC has the potential to satisfy the NIH data-sharing requirements.
American College of Radiology Imaging Network
Data sharing for a specific disease and radiologic imaging modality motivates the American College of Radiology Imaging Network, or ACRIN, computed tomographic (CT) colonography trial, which is currently in the planning stage. One component of this large clinical trial requires that 200 CT colonography studies of patients with polyps 1 cm or larger will be placed in an image database for use by researchers primarily in the computer-aided polyp detection community. Other applications for these data may well appear, such as observer performance studies and evaluations of interpretation software. The images will be annotated by three experienced reviewers, and CT colonography cases will come from multiple institutions. The final data set will be large (on the order of 50 GB when compressed). This goal is consistent with the ACRIN data sharing policy (13,14).
Digital Sky Project
In other areas of science, such as astronomy, data sharing is commonplace. Large-area digital sky surveys are a recent and exciting development in astronomic research. The combination of terabyte and/or teraflops computational resources with the recent large-area surveys in optical, infrared, and radio wavelengths provide unprecedented capability for astronomic research. With these capabilities, astronomers will have access to a multiwavelength digital library that covers a substantial fraction of the real sky. The Digital Sky and the tools for its exploration will revolutionize multiwavelength astronomic studies, both by increasing the data available and by providing faster and more sophisticated methods for its analysis. The Digital Sky offers a perspective of the universe that is statistics and data focused, in contrast to traditional work with individual stellar objects. The Digital Sky is funded through the National Partnerships for Advanced Computational Infrastructure, a program of the National Science Foundation (15).
The Digital Sky project provides simultaneous access to the catalogs and image data, together with sufficient computing capability to allow detailed correlated studies across the entire data set. Not all the data will be literally online. While the metadata are expected to be, the actual databases are so voluminous that they will reside in the high-performance archival storage system at the Center for Advanced Computing Research at California Institute of Technology (Pasadena, Calif) and at the San Diego Supercomputer Center, Calif (16).
Neuroscience
Neuroscience is generating vast amounts of highly diverse data that are of potential interest to researchers beyond the laboratories in which they are collected (17,18). In particular, quantitative neuroanatomic data are relevant to a wide variety of areas, including studies of development, aging, disease, and computational modeling. Moreover, the discrete and well-defined nature of the data make them an ideal application for the development of systems designed to facilitate data archiving, sharing, and reuse. At present, the only widely used forms of dissemination are figures and tables in published articles, which suffer from inaccessibility and loss of machine readability. Illustrations are a selected subset of the available data. Numerous database projects are in progress to address these shortcomings (19–22).
A series of demonstration projects on neuroscience information integration that involve multiple institutions working on several species (eg, mice and man) at multiple scale levels (molecular, cellular, and organ systems) are underway as components of the Biomedical Informatics Research Network, or BIRN, which is sponsored by the National Center for Research Resources at NIH. Several components of a software system and middleware to provide this infrastructure are now in place. BIRN should provide an integrated view of data stored at many different sites and the tools to analyze them (23,24).
Challenges and Benefits to Radiology
Examples in other sciences offer proof of the usefulness and benefit that data sharing provides through encouraging growth and development in those fields (10). The barriers to open data access include human subject confidentiality issues, assurance of investigator’s rights, description and organization of heterogeneous data, development of search tools, and transfer of data (21,24,25).
Image sharing in radiology may require a cultural change. Investigators already coping with limited research time and regulatory requirements will not welcome an additional burden unless they can be convinced of the benefits. Additional funding may be required to provide the needed resources (personnel, computer equipment) to prepare and transfer the images. Network infrastructure may require upgrading to transport large image data sets.
Nevertheless, the benefits are clear. By establishing rich linkages to other forms of biologic information, we can connect the imaging-based phenotype of disease with the underlying genotype and functional expression of key molecules. An integrated view of organisms, diseases, and therapies is essential to understanding the underlying mechanisms.
Implementation
Journal editors will likely play a key role in any effort to expand the sharing of images. However, journal editors cannot enforce data sharing unless an infrastructure exists to make sharing easy and inexpensive. Currently, establishment of an image repository is outside the technical capabilities and resources of most journals. To make image sharing feasible, the burden of establishing an image repository could be borne by organizations with the appropriate resources and skills, such as government or learned societies.
This requirement is not unique to radiology, and the National Digital Information Infrastructure and Preservation Program through the Library of Congress seeks to provide persistent infrastructure and aid in preserving electronic information (26).
A convenient mechanism for data sharing is the notion of a distributed image database infrastructure. Rather than developing new standards for databases that enable image sharing, one can leverage existing standards. For example, Z39.50 is a protocol of the American National Standards Institute and National Information Standards Organization that is widely used by research libraries (www.loc .gov/z3950/agency/; accessed April 16, 2003). This standard describes the search and retrieval of information from remote databases and addresses communication between the client and server. Z39.50 is implemented in inexpensive commercially available bibliographic software in use by many radiologists.
Ground truth is fundamental to utility of image databases and must be of high quality and defined and described explicitly. Similarly, the level of peer review of the data must be defined explicitly. One way to combine these two requirements is for images to be uploaded as supplementary information for a journal manuscript to be evaluated as part of the peer review process. For example, scholarly articles in the journals Radiology, Science, and Nature have supplementary information on dedicated Web sites.
Because region-of-interest and volume-of-interest information is fundamental to the understanding of complex images such as those in radiology, a common standard needs to be widely available. Examples of possible standards for region- and volume-of-interest information might include digital imaging and communication in medicine, or DICOM, structured reporting or extensible markup language, or XML. Development of a consensus on the appropriate standard by the radiology research community could go a long way toward easing incorporation of region- and volume-of-interest information into shared image databases. A major effort to develop such standards is underway in lung cancer CT screening as part of the Lung Image Database Consortium at the NIH (27).
Future Possibilities
Is all this extra effort worthwhile? It is valuable because the potential benefits to the public could be great. Lack of access to data is a common problem that may impede clinical research. Studies with large sample sizes usually have higher statistical power and greater clinical effects. Unfortunately, the development of large clinical databases is expensive, time consuming, and labor intensive. Case material may necessarily be limited or constrained geographically. Yet the "brain power" to analyze the data and originate new ideas may be more diverse.
The availability of large shared image databases could lead to new research directions in radiology, such as content-based image retrieval (28). For example, a radiologist could compare a patient’s images with those in the database. The database could contain a spectrum of images that depict normality and disease, a normal variant and developmental anatomy atlas, and a resource for the study of aging-related changes. A larger set of representative images might improve diagnostic confidence and lead to improved patient care.
Currently, most radiology research is taken on faith on the basis of a few published images and summaries of the data. Little research is verified at a deeper level; hence, the field moves forward more slowly than it could. The hindrance of sequestered data impairs the process for testing and rejection of hypotheses that is fundamental to scientific research.
There is a clear and immediate need to archive and share valuable imaging research data and to engage peer review processes to assess data quality and timeliness.
To Learn More
To track the progress in this area, as well as to establish and maintain communications among interested investigators, users, industry, and government, a Web site and listserver have been established at the National Cancer Institute Biomedical Imaging Program (29,30).
ACKNOWLEDGMENTS
We thank Andrew Dwyer, MD, for reviewing the manuscript.
REFERENCES
This article has been cited by other articles:
|
|
 |
|
  S. G. Armato III, G. McLennan, M. F. McNitt-Gray, C. R. Meyer, D. Yankelevitz, D. R. Aberle, C. I. Henschke, E. A. Hoffman, E. A. Kazerooni, H. MacMahon, et al. Lung Image Database Consortium: Developing a Resource for the Medical Imaging Research Community Radiology, September 1, 2004; 232(3): 739 - 748. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
