SUMMARY

First NIfTI Data Format Working Group Meeting

July 11 & 12, 2002

Conference Rooms B1/B2, Neuroscience Center

6001 Executive Boulevard

Rockville, Maryland

I           PURPOSE

The Data Format Working Group (DFWG) was established in response to specific recommendations made at two previous workshops held under the auspices of the Neuroimaging Informatics Technology Initiative (NIfTI). In those workshops, the use of a variety of different formats for functional magnetic resonance (fMRI) data by the research community was identified by tool-users and tool-developers as the major cause of a host of problems impeding progress in this important area of neuroimaging.

In accord with the recommendations from those workshops, the membership of the DFWG draws from research groups that are highly active in developing informatics tools for fMRI data, and it is charged with arriving at a technical solution to the problems engendered by the use of multiple, diverse, and evolving fMRI data formats. The DFWG, therefore, represents a community-based approach to solving a widespread problem in fMRI research.

The purpose of this meeting was to begin the process of identifying needs, constraints, and opportunities related to pursuit of this solution. The nature and range of problems were to be defined in the context of community goals. Priorities for addressing particular problems were to be discussed, and existing solutions to these problems were to be explored. Finally, immediate next steps in the process of arriving at a technical solution to the problem of multiple data formats were to be identified.

II          BACKGROUND

The use of neuroimaging to study the structure and function of the human brain has opened unprecedented vistas of understanding of this most complex and important organ. Among neuroimaging approaches to study brain function, fMRI stands out due to its versatility, compatibility with other key magnetic resonance imaging modalities, and its non-invasive nature. Not surprisingly, the use of fMRI has been embraced by neuroscientists and has greatly accelerated our understanding of the human brain in health and disease.

As neuroscience labs have developed and adapted fMRI technology, they have come to use a wide array of different formats for their fMRI data. And, as the use of fMRI has spread, so too have informatics tools that were written in particular labs to make sense and use of neuroimaging data. Such informatics tools perform functions like: spatial normalization, modeling the hemodynamic response, statistical analyses in space or time domains, data visualization, image segmentation, feature extraction, noise suppression, etc. Although different tools have been developed in different labs to perform the same function, it is significant to note that different tools seeking to accomplish the same goal (e.g., model the hemodynamic response) often do so in different ways. This means that the same input to the processing streams of different tools (that are supposedly performing the same function) can give rise to different outputs. Finally, each such informatics tool does not accommodate all data formats, so the data format in which the fMRI data are instantiated plays a big role in determining which informatics tool is used to subsequently process the data.

The fact that a given tool accepts a limited range of data formats together with the fact that different tools process the same data in different ways (and give different outputs) has several significant, negative consequences for fMRI-based neuroscience research. First, tools are not interoperable with each other, so the data format limitation places practical limitations on the manner in which fMRI data are processed and exploredÑthese limits to scientific inquiry are, therefore, imposed for reasons not driven by science. Second, rigorous comparison of tools is not possibleÑwhether one tool is better than another for a particular task, for example, is a straightforward question without a straightforward answer. Third, rigorous comparison of the neuroscientific results obtained in different labs using different tools is not possibleÑso when results differ across studies, it is not clear whether this finding is adding to our knowledge of brain function or merely an epiphenomenon of the software used to process the data.

Large labs with ample resources have overcome these obstacles by investing in programming capability and programming around them (e.g., writing software to convert data from one format into another), but a large number of labs do not have such resources at their disposal, and are greatly constrained by these circumstances. While there are several possible avenues through which to remove the constraints described above, members of the tool-developer and tool-user communities recommended that the best initial approach would be to establish a technical solution to the use of different fMRI data formats. A technical solution would have the potential of broadening the range of tools that could be used for processing data from a given study, and allow the rigorous comparison of the tools and, most importantly, of the neuroscientific results of fMRI studies.

To accomplish this, the NIMH and the NINDS, joint sponsors of NIfTI, convened the DFWG which comprises 11 experts who are deeply involved with development of informatics tools that are used widely by the fMRI research community. The DFWG is chaired by Stephen Strother (University of Minnesota, Minneapolis), and its members are: John Ashburner (Functional Imaging Laboratory, London), Robert Cox (NIMH Intramural Research Program, Bethesda), Kate Fissell (University of Pittsburgh), Rainer Goebel (Universiteit Maastricht), Christian Haselgrove (Massachusetts General Hospital, Charlestown), Colin Holmes (Silicon Graphics, Inc., Mountain View), Jack Lancaster (University of Texas Health Science Center, San Antonio), David Rex (University of California, Los Angeles), Stephen Smith (Oxford University Centre for Functional MRI of the Brain), and John Van Horn (Dartmouth College, Hanover). The DFWG is expected to meet several times over the next year, as well as communicating regularly via electronic means.

It is important to emphasize that at the outset of this activity, there was no preconceived notion of the form that the technical solution might take. Also important to emphasize is the fact that the technical solution arrived at by the DFWG will not be imposed upon the research community, but will be offered as a way to address the problems caused by the diversity of data formats in fMRI. Still, several groups responsible for developing fMRI informatics tools have already indicated their willingness to take advantage of such a solution in future software. Therefore, the DFWG has the potential to have a significant impact in this area.

III         DISCUSSION and FINDINGS

In the months preceding this inaugural meeting, members of the DFWG authored and circulated among themselves documents that provided their initial thoughts on the problem of multiple fMRI data formats. Papers were submitted both by those who were planning to attend as well as those who were unable to attend.

The meeting, itself, commenced with a morning of presentations. The first, by Michael Huerta, gave an overview of NIfTI and the charge to the DFWG. A series of presentations followed to provide technical orientation and provoke discussion, with Robert Cox presenting ÒXMLÕs Role in AFNI and Beyond,Ó Kate Fissell presenting ÒFormat Needs for Fiswidgets with Comment on ANALYZE,Ó David Rex presenting ÒData Format Needs of the LONI Software Pipeline,Ó and Colin Holmes presenting ÒOrigins and Capabilities of MINC with Comment on IndustryÕs Perspective.Ó

Over the course of the next day and a half, meeting participants discussed the range of problems caused by the use of multiple, diverse, and evolving data formats used in fMRI research, how these problems related to overarching goals of the fMRI research communities, and which problems should be given priority attention. Also discussed were whether solutions exist in whole or part to particular problems, and what steps should be taken next. The major findings and related action items that resulted from these discussions are summarized below.

• Number and nature of existing data formats used by the fMRI research community Despite the expertise and experience assembled at this meeting of the DFWG, it was clear that the full range of formats used in fMRI research was uncertain. Nevertheless, a systematic collection of different formats was considered by members of the DFWG as having importance for the fMRI research community at large for many reasons, as well as having importance to the DFWG in helping to identify next steps in the pursuit of a technical solution. The members of the DFWG volunteered to gather information about various formats; ultimately, this information will be compiled and disseminated on the NIfTI web site for use by the entire fMRI research community.

• Number and nature of existing data format converters used by the fMRI research community Many labs have written converters to move data from one format to another. How many such converters exist, and the details of their operation are not widely known, however. Since robust format converters with high fidelity in the conversion process would be extremely useful in addressing the problem of multiple data formats in the very near term, the DFWG volunteered to gather information systematically about existing converters. This information will then be compiled and disseminated on the NIfTI web site for the benefit of the entire fMRI research community.

• Simple data formats versus complex data formats The advantages and disadvantages of simple versus complex data formats was discussed at length by the DFWG. It was determined that both levels of complexity are useful. With this in mind, the DFWG decided to form two subgroups to discuss how existing simple and complex data formats might be enhanced and made available to the fMRI community at large. In addition, the idea of developing a robust and well-characterized converter between two such data formats was advanced as a way to facilitate problems of using different formats. These subgroups volunteered to draft documents that would outline how to enhance an existing simple format and an existing complex format. Drafts of these documents will be posted on the NIfTI web site, and comments on them will be solicited from the fMRI research community.

• Metadata and design matrix information An important component of data formats is the manner in which they address metadata. Since there is a range of opinions, however, about what constitutes metadata, the DFWG decided to draft a document addressing key issues. An aspect of particular interest to the members of the DFWG was the manner in which information about the experimental design be incorporated into the format; that is, how to generate a design matrix within a data format. A subgroup of DFWG will draft a document on metadata, with design matrix information considered in particular. This will ultimately be posted on the NIfTI web site for information to the fMRI research community

The DFWG decided to remain engaged electronically as these action items are carried out over the next several months, and will meet again in the first half of calendar year 2003.

IV        PARTICIPANTS

Stephen C. Strother (Chair)

V A Medical Center

International Neuroimaging Consortium (11P)

One Veterans Dr

Minneapolis, MN 55417

Tel: 612-725-2000 x4790

Email: steve@neurovia.umn.edu

Kate Fissell

University of Pittsburgh, LRDC

3939 O'Hara St.

Pittsburgh PA 15260

Tel: 412-624-5279

Fax: 412-624-9149

Email: fissell+@pitt.edu

Christian Haselgrove

Center for Morphometric Analysis

Massachusetts General Hospital

Building 149, 13th St.

Charlestown, MA 02129

Tel: 617-726-6018

Email: ch@cma.mgh.harvard.edu

William Heetderks

National Institute of Neurological Disorders

and Stroke

6001 Executive Blvd., Room 2207
Bethesda, MD 20892
Tel: 301-496-1447
Fax: 301-480-1080
Email: heet@nih.gov

Colin J Holmes

Silicon Graphics, Inc.

SGI 41-3-405, 1600 Amphitheatre Pkwy

Mountain View CA 94043-1351

Tel: 650-933-7512

Fax: 650-932-0332

Email: cholmes@engr.sgi.com

Michael F. Huerta

National Institute of Mental Health

6001 Executive Blvd., Room 7202

Bethesda, MD 20892-9645

Tel: 301-443-3563

Fax: 301-443-1731

Email: mhuerta@helix.nih.gov

Jack L. Lancaster

Department of Radiology

Research Imaging Center

University of Texas Health Science Center

7703 Floyd Curl Drive

San Antonio, TX 78229-3900

Tel: 210-567-8100

Fax: 210-567-8152

Email: jlancaster@uthscsa.edu

Yuan Liu

National Institute of Neurological Disorders

and Stroke

6001 Executive Blvd., Room 2110B

Bethesda, MD 20892-9523

Tel: 301-496-3108

Email: liuyuan2@ninds.nih.gov

David E. Rex

Laboratory of Neuro Imaging

Department of Neurology, Room 4-238

710 Westwood Plaza

Los Angeles, CA 90095-1769

Tel: 310-206-2101

Fax: 310-206-5518

Email: drex@loni.ucla.edu

Stephen M. Smith

Oxford University Centre for Functional MRI of the Brain

John Radcliffe Hospital, Headington,

Oxford OX3 9DU, UK

Tel: +44 (0) 1865 222726

Fax: +44 (0) 1865 222717

Email: steve@fmrib.ox.ac.uk

http://www.fmrib.ox.ac.uk/~steve

John D. Van Horn

Center for Cognitive Neuroscience

Dartmouth College

6162 Moore Hall

Hanover, New Hampshire 03755

Tel: (603)646-2909

Fax: (603)646-1181

Email: John.D.Van.Horn@dartmouth.edu

V         DFWG MEMBERS UNABLE TO ATTEND

John Ashburner

Functional Imaging Laboratory

12 Queen Square

London WC1N 3BG, UK.

Tel: +44 (0)20 78337491 or +44 (0)20 78373611 x4381

Fax: +44 (0)20 78131420

Email: john@fil.ion.ucl.ac.uk

http://www.fil.ion.ucl.ac.uk/~john

Rainer Goebel

Department of Neurocognition

Faculty of Psychology

Universiteit Maastricht

Maastricht, Netherlands

Tel: +31 43 3884014

Email: R.Goebel@Psychology.Unimaas.Nl