SUMMARY

NIMH and NINDS Joint Workshop

on Neuroimaging Informatics

Monday, April 24, 2000

Conference Room C, Neuroscience Center

6001 Executive Boulevard

Rockville, Maryland

I INTRODUCTION

The purpose of this meeting was to bring together key investigators involved in research and development of neuroimaging informatics tools to discuss several practical issues related to the usefulness and usability of such tools.

The primary question posed to the participants was: Is there a need to enhance existing neuroimaging informatics tools and approaches, making them more useful and useable? Important secondary questions relating to the manner in which this might be implemented successfully were also posed, including the extent to which extramural and intramural resources might be used.

II BACKGROUND

A variety of neuroimaging technologies allow the structure and function of the intact human brain to be studied. This presents a tremendous opportunity to better understand the human brain, both in healthy and disordered states.

The data generated by neuroimaging are vast, diverse and complex. To make sense and use of these data, neuroimaging relies upon informatics-a computerized way to handle data. Informatics is used at all stages of neuroimaging. For example, in magnetic resonance imaging, informatics is used to: design and implement the manner in which the imaging instruments capture signals generated by the brain, as well as the behavioral tasks used to probe particular brain systems, reconstruct the resulting signals into a three-dimensional representation of the brain, correct and suppress noise, statistically analyze the data, and visualize the results. This list is not comprehensive, and, of course, informatics is also used in storing, querying and retrieving data once it is collected.

Neuroimaging has the potential to reveal some of nature's most closely held and significant secrets, and informatics is key to realizing this potential. Paradoxically, however, seizing this opportunity has been impeded by inadequate coordination regarding the development and distribution of the informatics tools needed to meet this challenge, as outlined above. Existing tools have been developed piecemeal, primarily by students in laboratories interested in getting the answers to particular neuroscience questions, rather than in producing software products that are optimized for meeting the variety of needs of the broader community. It is, therefore, not surprising that these informatics tools are not as robust, or generally useful as they might be. While a handful of neuroimaging informatics tools are suitable for general use, and are used widely in the research community, many important tools are not widely available. Furthermore, even those that are in general use make varying assumptions, use different algorithms, or implement similar algorithms in different ways. This raises important concerns both about compatibility of use, and comparability of results that are reported in the literature.

Successfully addressing these concerns would expand access of neuroimaging approaches to wider research communities. Making powerful tools available to wider sets of users has transformed other areas of science, and this would likely transform and accelerate progress in neuroscience.

Other fields, such as graphic arts, clinical imaging and the internet community at large, have addressed these concerns by the establishing commonly agreed-upon conventions for informatics. This provides a foundation upon which individual development efforts can be based and insures compatibility of the results. These communities have profited greatly from the establishment of such conventions, both in terms of the ease of exchanging data, and the development of a rich set of software tools to meet their individual needs (for example, consider the number and range of "plug-ins" for Adobe Photoshop, or the development of the world-wide web, which was based on graphical standards). Research activities, however, require great flexibility that should not be unnecessarily limited by strict standards. The focus of much of the discussion at the workshop was on whether and how the neuroimaging community could benefit by the development of such informatics standards or conventions and, if so, how this could most effectively be initiated. Key to the success of finding common solutions that will be adopted by the neuroimaging research community will be to accommodate as wide a group as possible without forcing unneeded standards.

III DISCUSSION

The participants agreed that the manner in which neuroimaging informatics tools have been developed has produced results that are not as usable or useful as they could be. Among the many consequences of this history, as described by participants, were: difficulty in learning how to use poorly documented tools, fragile code, inefficiencies in porting tools to different computational platforms, incompatibilities among tools that serve complementary functions (the so-called "Tower of Babel" problem), no central point of origin or support for a variety of tools, a lack of validity testing of algorithms, and inadequate training opportunities for tool users at all levels of sophistication. These problems have also driven a large scale duplication of effort in the development of software tools within the neuroimaging community, which has produced significant inefficiencies at the level of individual laboratories and the support of the field by funding agencies.

In parallel to (and contributing to) problems with software development has been the lack of common data solutions (the data component of the "Tower of Babel" problem). This has impeded rigorous comparisons of results from different laboratories, the conduct of meta-analyses, and other means of exploring the already large and rapidly growing corpus of neuroimaging data. These limitations seriously compromise the ability of neuroimaging research to reach its full potential.

These concerns were viewed as reflecting the following set of needs:

• Field-wide data format conventions or other common solutions

• Interoperability of software tools (i.e., the ability of complementary tools to seamlessly communicate and interact with one another)

• Ability to disseminate software tools (i.e., their availability at accessible and supported locations, and their ability to operate transparently on the most commonly used hardware platforms)

• Clear and up-to-date documentation

• Comparability and reliability, which relies on commonly accepted benchmarking algorithms and datasets for evaluation

Participants noted that these concerns and needs have been voiced previously at workshops held by NIH and in other venues. However, as yet, no coordinated efforts to address these issues have emerged.

Staff from NINDS and NIMH presented an outline of one approach to addressing these needs through an envisioned effort referred to as the Neuroimaging Informatics Technology Initiative (NIfTI). This was discussed as a way to provide coordinated and targeted service, training, and research to enhance informatics tools used by the neuroimaging research community. Priority was described as making existing and widely-used tools more useable and useful. This could draw from both intramural and extramural aspects of NIH to make available to the effort all possible mechanisms. A close and ongoing interaction with the neuroimaging research community was considered key to the success of the effort, as was the use of experts in the tasks required (e.g., computer programming). Moreover, NIfTI efforts would be coordinated, where appropriate, with other pertinent informatics activities.

IV RECOMMENDATIONS/CONCLUSIONS

• Support for NIfTI: The participants strongly supported the need for NIfTI as an initiative, and agreed with the general outline of this initiative as thus far envisioned (e.g., using all mechanisms needed to accomplish goals, providing coordinated support for service, training and research to enhance neuroimaging informatics, etc.). Participants made clear that having unified leadership and resources for this effort are the bases for their support.

• Establishment of data format conventions or other common solutions: The participants provided a unanimous mandate for taking a first step, by addressing a fundamental aspect of neuroimaging informatics that will have significant and wide reaching implications for the tower of Babel problem: addressing the data interoperability problem.

Specifically, it was agreed that a group of a half-dozen or so people who are actually the ones writing the software used in the tools will meet and be charged with arriving at a technical solution for the problem presented by different labs using different data file formats. The solution might be either a standard format, an agreed-upon format translator, or some other common solution. Key is the understanding that a less-than-perfect solution is expected and is acceptable, but that some initial solution is required.

The problem of data interoperability is considered by many as one of the most important issues in imaging informatics. Despite this fact, there has been no coordinated, field-wide effort to address this issue, partly because past discussions of this issue have failed to produce a consensus on how to proceed. Representatives from virtually every major group involved in neuroimaging software development were present at this workshop. Their unanimous agreement about a process for addressing data interoperability, and their willingness to support and adhere to the results of this process, has the potential to open an important new chapter in neuroimaging. research

• Selecting tools for enhancement: A process was also suggested to begin a selection of specific tools for enhancement under NIfTI. The suggestion was made that a number of (15 or so) functions be identified (registration, detrending, etc.) that represent the most important and routine ones carried out in neuroimaging research. These functions would be matched with tools that perform them, and this would be the set of packages considered as a starting point.

• Development of new tools: For new tools, the importance of developmental pathways that would assure their compatibility was identified as an important aspect to consider for the intermediate future.

• Advisory board: Agreement was reached on the need for an advisory board that adequately represents the diverse interests of the neuroimaging community, including experts in neuroimaging informatics, software developers involved in both public domain and commercial efforts, less computer-savvy neuroimagers ("users"), and experts from relevant areas outside of neuroscience (e.g., computer science, applied mathematics, etc.).

V PARTICIPANTS

Peter Bandettini
Division of Intramural Research
National Institute of Mental Health
Bldg 10, Rm 4C104
Bethesda, MD 20892-1366
Phone: (301) 402-1333
Email: bandettini@nih.gov

Randy Buckner
Washington University
Department of Psychology
St. Louis, MO 63130-4899
Phone: (314) 935-5019
Email: rbuckner@artsci.wustl.edu

Jonathan Cohen (Co-Chair)
Princeton University
Department of Psychology
Green Hall
Princeton, NJ 08544
Phone: (609) 258-2696
Email: jdc@princeton.edu

Robert Cox
Medical College of Wisconsin
Biophysics Research Institute
8701 Watertown Plank Road
Milwaukee, WI 53226
Phone: (414) 456-4038
Email: rwcox@mcw.edu

Robert Desimone
Division of Intramural Research
National Institute of Mental Health
Bldg 10 Rm 4N 222
Bethesda, MD 20892-1381
Phone: (301) 496-3501
Email: desimonr@intra.nimh.nih.gov

Rainer Goebel
Max-Planck-Inst Brain Research
Dept Neurophysiology
Deutschordenstr 46
Frankfurt, 60528 Main, Germany
Phone: 49-69-96769-470
Email: goebel@mpih-frankfurt.mpg.de

Thomas J. Grabowski, Jr.
University of Iowa
Department of Neurology
2RCP, UIHC
Iowa City, IA 52242
Phone: (319) 356-8753
Email: thomas-grabowski@uiowa.edu

James Haxby
Division of Intramural Research
National Institute of Mental Health
Bldg 10, Rm 4C104
Bethesda, MD 20892
Email: haxby@box-h.nih.gov

William Heetderks
National Institute of Neurological Disorders and Stroke
6001 Executive Blvd.
Rm 2209
Rockville, MD 20852
Phone: (301) 496-1447
Email: heetderb@ninds.nih.gov

Michael Huerta
National Institute of Mental Health
6001 Executive Blvd.
Rm 7202
Rockville, MD 20852
Phone: (301) 443-3563
Email: mhuerta@helix.nih.gov

Alan Koretsky
Division of Intramural Research
National Institute of Neurological Disorders and Stroke
Bldg 36, Rm 5B05
Bethesda, MD 20892
Phone: (301) 402-0348
Email: koretskya@ninds.nih.gov

Story Landis
Division of Intramural Research
National Institute of Neurological Disorders and Stroke
Bldg 36, Rm 5A05
Bethesda, MD 20892-4150
Phone: (301) 435-2232
email: slandis@codon.nih.gov

Richard Leahy
University of Southern California
EE-System and Radiology
3740 McClintock Avenue, Suite 400
Los Angeles, CA 90089-2564
Phone: (213) 740-4659
Email: leahy@sipi.usc.edu

Yuan Liu
National Institute of Neurological Disorders and Stroke
6001 Executive Blvd.
Rm 2110B
Rockville, MD 20852
Phone: (301) 496-3108
Email: liuyuan2@ninds.nih.gov

Bruce Rosen
Massachusetts General Hospital
MGH-NMR-Ctr
13th St, Bldg 149
Charlestown, MA 02129
Phone: (617) 726-5122
Email: bruce@nmr.mgh.harvard.edu

John Strupp
University of Minnesota
Radiology
Box 292 Mayo, 8292, 420 Delaware
Minneapolis, MN 55455
Phone: (612) 626-0361
Email: strupp@tc.umn.edu

Arthur Toga (Co-Chair)
University of California at Los Angeles
Department of Neurology
710 Westwood Plaza
Los Angeles, CA 90024-1769
Phone: (310) 206-2101
Email: toga@loni.ucla.edu

Keith Worsley
McGill University
Montreal Neurological Institute
Burnside Hall, Room 1232
Montreal, Quebec, Canada H3A 2T5
Phone: (514) 398-3842
Email: worsley@math.mcgill.ca