Lung Imaging Workshop: Technology Transfer
Diagnostic Imaging Program (DIP), DCTD, NCI
Washington, D.C.
Jan 12-13th, 1998

1. Report: Technology Transfer Workshop

     A. Lung cancer imaging and image processing as a model for technology transfer
     B. Specific areas in lung cancer imaging that could be enhanced by technology transfer
     C. Recommendations and priorities for technology transfer
     D. Recommendations regarding mechanisms

2. List of Attendees

     A. Presenters
     B. Business Representatives
     C. NCI Representatives
     D. Invited Presenters (unable to attend, reviewed report)

3. Presentations and Abstracts

     A. Monday, January 12, 1998

          1. Introduction and Overview of Current Imaging Methods
          2. Feature Extraction, Lesion Characterization and Classification
          3. Multi modality: Image Reconstruction and Image Restoration Methods
          4. Planer: Image Registration and Change Point Analysis
          5. Hardware Implementation
          6. DARPA: Representative Image Processing Applications of Dual Use

     B. Tuesday, January 13, 1998

          7. Summary of Workshop

1. REPORT: TECHNOLOGY TRANSFER WORKSHOP

Section A. Lung cancer imaging and image processing as a model for technology transfer

Clinical Rational. The issues relating to the clinical need for lung cancer imaging are well recognized in the published literature and are briefly summarized below:

• Lung cancer is usually not detected early, and thus may be diagnosed at an advanced stage, where intervention or therapy is less effective. Although the incidence rate of lung cancer is lower than for breast and prostate cancer, the mortality rate of lung cancer is the highest for all cancers in both men and women.
• The sensitivity for lung nodule detection of peripheral lung cancers by conventional chest radiographs is still very low. There is a critical need for alternative imaging strategies and image processing techniques to improve the sensitivity of detection. Similarly, the limitation of the response characteristics of screen-film detectors, and the wide variety in image quality that is seen in chest radiographs, greatly impacts the accuracy of image interpretation.
• There is a critical need to consider longitudinal studies for early detection of lung cancer for asymptomatic patients, using different imaging modalities, such as for high-risk patients. A cost effective means to automatically and quantitatively measure these early changes as soon as possible by image processing is a worthwhile goal.
• There is a need to improve follow up for diagnostic imaging studies to improve intervention, such as surgery or medical therapy, perhaps using tomographic and functional imaging methods, and provide a more effective and quantitative measure of the progress or response to treatment of the disease.

Digital Planar Chest Radiography. Currently the majority of lung cancers are detected from conventional chest radiographs, because of the cost effectiveness of this imaging modality. The technical requirements for film digitization are straightforward with standard film digitization equipment. In addition, since the late 1980's there has been a major transition from conventional film-based X-ray techniques to digital sensors, using either storage phosphors or, more recently, flat panel digital detectors, that allow real-time imaging and potentially real-time quality control procedures. These digital sensors permit several improvements that include: (a) dual-energy imaging to partially remove the rib cage to improve cancer detection by image subtraction, and (b) the introduction of image processing methods to improve and potentially standardize image interpretation. The use of digital sensors and the potentially widespread implementation of PACS systems in the near future pose very critical problems for image interpretation using computer monitors, computer-processed images using hardcopy devices, and related problems for remote diagnosis or teleradiology using low-end computer monitors.

X-Ray Tomography. There is an increasing interest in performing longitudinal studies using either digital X-ray chest imaging or X-ray CT. Similarly, alternative cost effective X-ray imaging methods such as tomosynthesis, or limited angle tomography, have recently been proposed for longitudinal studies, that may allow an approximate 3D image representation of the chest. Tomosynthesis may partially reduce the problems with overlapping structures in relation to lung nodule detection, where the nodules may be wholly or partly overlapped with the rib cage. Tomosynthesis is still in the early development stage. Although it offers the advantage of a 3D approach without significantly increasing radiation dose, it poses additional new problems in terms of optimized reconstruction methods, image artifacts, and need for image processing to improve in-plane image contrast. Similarly, the introduction of low dose X-ray CT, such as spiral imaging methods, either for longitudinal studies or diagnostic follow-up investigations, has stimulated increasing interest in alternative image reconstruction methods to potentially reduce dose, and methods for improved feature extraction using raw data in the frequency domain. Although the practicality of some of the alternative reconstruction methods needs further review, the advances in low dose X-ray CT clearly pose new image processing problems for image reconstruction, image restoration, image registration and 3D image fusion with other modalities. Similar problems arise for other tomographic modalities such as MRI, PET, SPECT and coincidence SPECT, that are proposed for improved tissue characterization of lesions such as lung nodules. Allowance needs to be made in imaging processing methods for the varying response characteristics of these sensors.

Computer Assisted Diagnostic (CAD) methods. The 2D and 3D planar and tomographic modalities generate new issues in terms of cost effective and efficient image interpretation, and the potential improvement of observer performance or reduction of observer variability in reading images. Further development of optimized computer assisted diagnostic (CAD) methods are required, as a A second opinion @ strategy, for the very extensive 2D or 3D image data sets generated by different sensors, where image reading requirements pose a serious obstacle for efficient patient throughput. The performance of CAD methods has been recently shown to be dependent on the response characteristics of film digitizer, such as the gray scale characteristics or spatial resolution. There is a need to develop CAD methods that are less dependent on the digital sensor characteristics, i.e. a more generalized CAD method is required. CAD methods also need to be developed for longitudinal studies with different sensors. One clinical example is the CAD detection of peripheral lung nodules, but other applications for lung imaging, such as extension to hilar lymph nodes, or airway involvement, should be further explored.

Virtual bronchoscopy. Virtual bronchoscopy is an emerging visualization technology that may become a useful diagnostic tool for evaluation of the airways. Several published have shown it to provide a more complete evaluation of stenoses of the airway (due to inflammation, scarring, tumor, or post-lung transplant). One of virtual bronchoscopy's most important strengths is how well it displays three-dimensional data to referring clinicians. Potential applications of virtual bronchoscopy still under investigation include detection of endobronchial lesions which do not cause stenosis (e.g., early detection of carcinoma) and its use to guide transbronchial biopsy of lesions just beyond the airway wall (hidden from, yet accessible to the bronchoscopist). Such use could decrease the need for mediastinoscopy. Several commercial virtual endoscopy systems which include the capability to perform virtual bronchoscopy are currently available. However, research on virtual bronchoscopy is still in its infancy, and validation in larger numbers of patients, methodology and user interface issues, and potential artifacts are still being evaluated. Image processing methods which may prove useful for virtual bronchoscopy include 3D CAD methods and 3D image registration of functional imaging data sets with the anatomical image data sets.

Technology Transfer Issues. It is clear from workshop discussions that new algorithm approaches in image processing for all modalities are required to enhance the outcome of the rapid advances in hardware development. These methods are pre-eminently mathematical and the subject of intensive research in the fields of mathematics, computer science, electrical engineering, and physics. These investigators are predominately university-based and funded by NSF and DOD/DARPA. This workshop demonstrated the critical need for both timely access to these more sophisticated image processing techniques and the necessity for translational research to allow these software tools to be tailored and optimized for specific medical applications such as lung imaging. The implementation of some form of centralized imaging processing software resource was considered important to provide a means for reduction of research time lines for lung imaging within the medical imaging research community. This could be similar to the NCI implementation of common resources for the human genome project.

Summary. There was a strong consensus that lung imaging was an important clinical model and image processing was an excellent technical model for technology transfer. There are not only problems to be solved but exciting new techniques that can solve them. Similar technology transfer for hardware research is also an area to be actively considered in the future. Specific research areas where technologies were identified as suitable for technology transfer are identified in Section B, recommended actions items are briefly outlined in Section C and priorities for technology transfer are described in Section D.

Section B. Specific areas in lung cancer imaging that could be enhanced by technology transfer

1. Image registration and change point analysis.

• Need for optimized inter-modality image registration, as required for time series or longitudinal studies, using chest radiographs or perhaps x-ray limited angle tomography.
• Need for inter-modality registration, e.g., radiographs registered to tomographic modalities such as x-ray CT, PET or SPECT images or with x-ray limited angle tomography.
• Need to examine new techniques, for example:
  1. using internal anatomical landmarks as opposed to rigid registration, such as feature-based or site-model image registration
  2. deformable registration models such as image-warping registration methods
  3. multi-resolution wavelet approaches with selective sub-image registration at different scales for computational efficiency
  4. evaluation of constraints for stabilizing image-matching procedures for small localized structures such as lung nodules; i.e. to allow for deformation of normal anatomy over time or anatomical variation in imaging protocols using all imaging modalities.
• Need for exploring alternative and more robust metrics for classification of changes in images over time should be investigated that are also site specific, such as spatial scan statistics and modified statistical approaches for improved likelihood estimation of change.

2. Image fusion.

• 3D multi-modality image fusion and related image segmentation, single-scale spatial image modeling approaches that include image warping, knowledge-based methods, and multi-spectral feature-space methods such as clustering methods including fuzzy logic. These methods are required for integrated 3D displays, visualization and virtual reality such as virtual bronchoscopy, i.e. for anatomical and functional imaging for tomographic modalities as required for classification, or follow-up studies in lung cancer using x-ray CT, PET and SPECT.

3. Image restoration.

• New approaches for image resolution or image contrast restoration are required to compensate for either:
  (a) the limited response characteristics of the detector such as for tomographic nuclear imaging;
  (b) to compensate for influence of photon transport, such as for x-ray detectors (planar or tomographic imaging) and nuclear detectors (SPECT, PET).

4. Image reconstruction.

• Need alternative 3D tomographic image reconstruction methods such as for x-ray CT, MRI, PET or SPECT, tomosynthesis or limited angle tomography. Examples include the feasibility of the limited field of view or localized reconstruction methods for all tomographic modalities. There is a need to investigate the stability of linear inverse problems and feasibility of non-linear solutions to non-ideal image tomographic reconstruction problems to facilitate, for example, the use of x-ray flat panel detectors in limited angle tomography.

5. Image enhancement.

• More optimized techniques are required for improving the visual interpretation of images using soft copy devices, such as high resolution computer monitors, or hard copy image techniques and related issues of image perception.

6. CAD: Automatic feature extraction methods.

• Improved feature extraction is clearly required from both the perspective of CAD methods and longitudinal studies. Feature extraction is also important for image registration. This should include all imaging modalities, such as planar film (digitized), digital x-ray detectors, and tomographic imaging modalities. and related longitudinal studies. It is therefore important to develop feature extraction for both 2D and 3D image sets, as required for lung nodule detection, and to allow nodule and related features to be computed in cases where the nodule is overlapped with the rib cage. Similarly, extraction of improved features to distinguish nodules from other similar structures, such as blood vessels, is necessary to reduce the false positive detection rate. Feature extraction and related segmentation is required for functional tomographic modalities for diagnostic follow-up, including PET and SPECT.
• The feasibility of alternative multi-resolution approaches for feature extraction may be of interest, as opposed to single scale methods, such as the use of wavelet transforms that have both direction and frequency information within discrete frequency band widths. Several new methods have been actively developed for defense application in the DOD sector including automatic target recognition (ATR). Methods for selection of wavelet bases should possibly be considered for potentially more optimum performance for certain applications.
• Other alternative transforms for improved feature extraction, sensitive to small structures such as nodules, are important to explore. Similarly, feature extraction using the raw image in the frequency domain for tomographic images may be promising as an alternative to the spatial domain (reconstructed images).

7. CAD: Automatic classification methods.

• Alternative classification methods should be explored that may provide for more robust and stable solutions to difficult pattern recognition problems. Approaches may include nonlinear discrimination and recognition techniques, such as neural networks and/or fuzzy systems, based on significant advances made over the last decade. In addition, these may include recent advances in evolutionary computation, where optimal fuzzy/neural classifiers may be optimized through random variation and selection. Similarly, evolutionary methods (such as genetic algorithms and evolutionary programming) should prove useful for feature subset selection and transformation, leading to parsimonious models that reduce the false positive detection of lung nodules.
• There is a critical need to reduce both the training time and classification speed for pattern recognition methods as applied to medical image analysis as addressed below without sacrificing sensitivity or specificity.

8. Hardware implementations.

• The feasibility of close-to-real-time and cost-effective image processing or CAD methods should be explored using hardware approaches. Examples include field programmable gate arrays, as required for: (a) training and application of neural networks, or (b) application of wavelet transforms or other transforms implemented on compatible filter banks for feature extraction.
• Alternative approaches using advances in workstation parallel CPU design and efficient parallel code implementation should be considered.
• The practicality of other non-traditional methods need to be reviewed carefully such as optical image processing as applied to high-order Fourier and wavelet transforms for feature extraction or optical correlation of images.

Section C. Recommendations and Priorities for Technology Transfer.

1. Information Dissemination.

NCI should be proactive for reaching out and identifying leading non-medical researchers working on image processing and drawing them into imaging problems such as lung imaging. NCI should be specifically proactive in promoting interactions between such investigators and medical imaging specialists using approaches such as:

• Workshop format with input from specialists outside the medical imaging sector such as the current workshop.
• Internet Web site with links to other federal resources such as NSF and DARPA. (Not required to build an all-encompassing Web site, but to provide links to resources.)
• Presentations by NCI representatives at a broad spectrum of regional and national meetings. (Example: both medical and non-medical meetings such as SPIE Medical Imaging, IEEE signal processing and IEEE EMB-sponsored national meetings.)
• Contacting university-sponsored research or grant offices with program announcements or problem statements to determine the level of expertise and interest within basic science departments in these institutions. Alternatively, one could coordinate this effort as an institutional effort within NIH and link, for example, with computational mathematics divisions of NSF and DARPA. This is being pursued in the broader field of biomedical engineering by seminars and programs recently sponsored by NIH.

2. Infrastructure.

• Development of image databases and mechanisms for distribution. Issues that need to be addressed include generation of multiple modality image data bases, ground truth, format (DICOM/ACR) standards, other database information (such as ACoS and SEER tumor registry findings).
• Timely availability of image data bases with ground truth from leading imaging technologies to allow non-medical image software investigators to test the feasibility of new algorithms.
• Standardized metrics for objective comparison and evaluation of algorithms. This issue is critical as metrics used to evaluate detection of an abnormality greatly influence the measured performance. Similarly, standards for training algorithms are important, for example, classifiers such as neural networks or the incorporation of a priori knowledge in the image processing chain.
• Access to data could be possibly limited to investigators that are funded by any federal agency or private foundation. For example, they may state requested access to this data base in submitted proposals on algorithm design and adhere to recommended standards and metrics for evaluation. (This will allow controlled access to the data that should not be disseminated or published without the corresponding evaluation techniques.)

Section D. Recommendations regarding mechanisms.

1. Existing Models and Possible NCI/NIH extension.

Consider emulating the DARPA model in the area of computational mathematics, where formulated contracts within academia and industry are developed. This was specifically performed proactively to establish scientific program priorities and thus produce specific software components or modules for broad DOD program driven applications. DARPA is currently successful in getting its grantees to move more rapidly by constraining the infrastructure, i.e. by reducing the number of different software platforms, languages and data formats. The users built new infrastructure on the common platform and exchanged the results of their work in a highly interactive way. This approach is intended to enhance progress at multiple sites and reduce duplication of effort in building software tools and infrastructure to perform research tasks. DARPA also proactively fosters interchange of ideas and results at workshop meetings so that tools and successful ideas move quickly from grantee to grantee. One could consider interagency agreement/programs such as with NSF and/or DARPA to leverage off existing programs, for dual use technology or technology transfer, and enhance timely dissemination to the medical imaging community. NSF and DARPA representatives at the workshop have expressed interest in this inter-agency approach at the program level for computational mathematics.

2. Alternative approaches using NCI/NIH programs.

NCI could consider using existing resource grants or equivalent programs, but with the suggested modifications:

• Technology transfer funding from NIH is nontraditional, and should have some differences in program management built-in. This includes issues such as:
  1. Partnering with medical and non-medical scientists from universities, national and DOD laboratories and industry
  2. Infrastructure support for all participants
  3. Award size to accommodate multi-center efforts and software dissemination.
  4. Grant review process needs to emphasize either research-and-development-type proposals or contract-type proposals that may be required for infrastructure support.
  5. Cooperation among technology-transfer grantees, and with other grantees including industry.
  6. Infrastructure support: Preliminary funds to stimulate new developments through common software, data interchange, platforms and image data bases. These issues are considered critical for partnering with either non-medical institutions or industry.

3. Concerns in relation to success of technology transfer.

• There is a need to critically address continuity of support for technology transfer to ensure that real progress can be made throughout both the development and evaluation research timetables.
• Continuity of experience and knowledge is important for effective management and related technical and scientific communication between medical and non-medical scientists, recognizing differences in cultures, and allowing for continued research transition of non-medical scientists to new imaging problems. Some form of exchange of scientists between such laboratories is necessary.
• Intellectual property issues need to be addressed in relation to technology transfer mechanisms similar to that practiced within the DOD/ DARPA.
• Timely and wide dissemination of technology is required, with full documentation for software implementation, to allow improved research timetables similar to the very successful NCI Human Genome Project.

2. LIST OF ATTENDEES

Workshop attendees are from university and medical research institutions, IPA=s from NCI, NSF and DARPA, and Editors or Associate Editors from leading IEEE journals in computer software.

A. Presenters.

Carlos A. Berenstein, Ph.D.

IPA: (University of Maryland): Program Director, Division of Mathematical Sciences, NSF, Arlington, VA
National Science Foundation, 4201 Wilson Blvd., Room 1025, Arlington, VA 22230
(703) 306-1870, FAX (703) 306-0555, cberenst@nsf.gov

Laurence P. Clarke, Ph.D.

IPA: Special Assistant, DCTD, NIH, Bethesda, MD
Associate Editor: IEEE TMI, Editorial Board, SMRI and MRI
Professor of Radiology and Physics, Director of Research, Department of Radiology,

Program Leader: Moffitt Digital Medical Imaging Program (DMIP)
Department of Radiology, College of Medicine, University of South Florida, 12901 Bruce B. Downs Blvd., MDC 17, Tampa, FL 33612-4799
(813) 975-7833 or 7835, FAX (813) 979-6724, clarke@rad.usf.edu

Carey Floyd, Ph.D.

Professor of Radiology and Biomedical Engineering
Duke University, Radiology/Biomedical Engineering, Medical Center, Box 3949, Durham, NC 27710
(919) 684-4138, FAX (919) 684-2711 or 3934, cef@deckard.mc.duke.edu

David Fogel, Ph.D.

Editor: IEEE Transactions on Computational Methods
Executive Vice President and Chief Scientist
Natural Selection, Inc., 3333 N. Torrey Pines Ct., Suite 200, La Jolla, CA 92037
(619) 455-6449, FAX (619) 455-1560, dfogel@natural-selection.com

Matthew Freedman, MD, MBA

Associate Professor of Radiology and Clinical, Director of the Division of Imaging Science and Information Systems, Georgetown University Medical Center
Freedman@isis.imac.georgetown.edu

Dennis Healy, Ph.D.

IPA: (From Dartmouth, Mathematics Department) Applied and Computational Mathematics Program, DARPA/DSO, 3701 N. Fairfax Drive, Arlington, VA 22203-1714
(703) 696-0143, FAX (703) 696-3999, dhealy@darpa.mil

Maryellen L. Giger, Ph.D.

Editor: Academic Radiology, Associate Editor: IEEE-TMI and Medical Physics.
Associate Professor of Radiology, University of Chicago
University of Chicago, Radiology Department, MC2026, 5841 So. Maryland Ave., Chicago, IL 60637
(773) 702-6778, FAX (773) 702-0371, m-giger@uchicago.edu

Michael McNitt-Gray, Ph.D.

Assistant Professor
Medical Imaging Division, Department of Radiological Sciences, CHS, AR-265, UCLA School of Medicine, 10833 Le Conte Ave., Los Angeles, CA 90024-1721
(310) 206-3285, FAX (310) 206-2967, mmcnitt-gray@mail.rad.ucla.edu

Loren T. Niklason, Ph.D.

Assistant Professor of Radiology, Harvard Medical School, Boston, MA
Radiological Sciences and Tech., Mass. General Hospital, Edwards Res. Bldg., Room 517, Fruit Street, Boston, MA 02114
(617) 726-6757, FAX (617) 726-5123, niklason@helix.mgh.harvard.edu

Tim Olson, Ph.D.

Assistant Professor
Department of Mathematics, University of Florida, 358 Little Hall, Gainesville, FL 32611
(352) 392-0281, FAX (352)392-8357, olson@totcon.com

Carey Priebe, Ph.D.

Assistant Professor,
Department of Mathematical Sciences, 104 Whitehead Hall, Johns Hopkins University, Baltimore, MD 21218
(410) 516-7200, FAX (410) 516-7459, cep@jhu.edu

Charles Putman, M.D.

Senior VP for Research Admin. and Policy, Professor of Radiology and Medicine, Chair of ISWG lung imaging evaluation. Multi modality lung imaging.
Duke University, 012 Allen Building, Box 90026, Durham, NC 27708-0026
(919) 684-3403, FAX (919) 684-8525, cputman@mail01.adm.duke.edu

Anthony P. Reeves, Ph.D.

Associate Professor
School of Electrical Engineering, 331 Rhodes Hall, Ithaca, NY 14853-5401
(607) 255-2342, FAX (607) 255-9072, reeves@ee.cornell.edu

Harold S. Stone, Ph.D.

Fellow: NEC Research Institute, Princeton, NJ
NEC Research Institute, 4 Independence Way, Princeton, NJ 08540
(609) 951-2999, (609) 951-2488, hstone@research.nj.nec.com

Ronald M. Summers, M.D., Ph.D.

Department of Radiology, National Institutes of Health, Building 10, Room 1C660, 10 Center Drive MSC 1182, Bethesda, MD 20892-1182
(301) 496-7700, Fax: (301) 496-9933, rms@nih.gov

Raoul Tawel, Ph.D.

Jet Propulsion Laboratory, California Institute of Technology, Mail 302-231, 4800 Oak Grove Drive, Pasadena, CA 91109
(818) 354-4951, FAX (818) 393-4272, raoul@neuron6.jpl.nasa.gov,
raoul@brain.jpl.nasa.gov

Yue Joseph Wang, Ph.D.

Assistant Professor of Radiology
Electrical Engineering, Computer Sci. & Radiology, The Catholic University of America, Michigan Avenue 620, Washington, DC 20064
(202) 319-5879, lab (202) 319-5249, FAX (202) 319-5195, wang@pluto.ee.cua.edu

John B. Weaver, Ph.D.

Associate Professor of Radiology
Department of Diagnostic Radiology, Dartmouth-Hitchcock Medical Center, One Medical Center Dr., Levanon, NH 03756-0001
(603) 650-5846, FAX (603) 650-5455, John.B.Weaver@Hitchcock.ORG

B. Business Representatives.

James R. Aloise

Business Development Manager
GE Corporate Research and Development, Bldg. KW, Room C284C, P.O. Box 8, Schenectady, NY 12301
(518) 387-5068, FAX (518) 387-5449, aloise@crd.ge.com

Nadeem Ishaque, Ph.D.

Manager, CT Program
GE Corporate Research and Development, Bldg. KW, Room C284C, P.O. Box 8, Schenectady, NY 12301
(518) 387-7603, FAX (518) 387-6030 , ishaque@crd.ge.com

John C. Huffman, Ph.D.

Director of Marketing, Medical Imaging
MTS Corporate R&D, Silicon Graphics, Mail Stop 005, 2011 N. Shoreline Blvd., Mountain View, CA 94043-1389
(415) 933-6037, FAX (415) 969-6289, jhuffman@sgi.com

C. NCI Representatives

Daniel Sullivan, M.D.
Associate Director, Diagnostic Imaging Program (DIP)
NCI, DCTD, DIP, 6130 Executive Blvd., Rm. 800, Rockville, MD 20852
(301) 496-9531, FAX (301) 480-5785, sullivand@dtpepn.nci.nih.gov

Anne Menkins, Ph.D.

Program Director, Diagnostic Imaging Program (DIP)
NCI, DCTD, DIP, 6130 Executive Blvd., Rm. 800, Rockville, MD 20852
(301) 496-9531, FAX (301) 480-5785, menkinsa@dtpepn.nci.nih.gov

Manuel Torres-Anjel, Ph.D.

Program Director, Diagnostic Imaging Program, National Cancer Institute
NCI, DCTD, DIP, EPN/800, 6130 Executive Blvd., MSC 7440, Bethesda, MD 20892-7440
(301) 496-0735, FAX (301) 480-5785, torresm@dtpepn.nci.nih.gov

Donald Henson, M.D.

Medical Officer, Division of Cancer Prevention, National Cancer Institute
NCI, DCPC, 6130 Executive Blvd., Rm. 330, Rockville, MD 20852
(301) 496-9424, FAX (301) 496-8667, deh@helix.nih.gov

D. Invited Presenters (unable to attend, reviewed report)

James C. Bezdek, Ph.D.

Editor: IEEE Transactions on Fuzzy Logic
Nystul Professor of Computer Science, Computer Science Department, University of West Florida
Computer Science Department, University of West Florida, 11000 University Parkway, Pensacola, Florida 32514-5750
(850) 474-2784, FAX (850) 857-6056, jbezdek@argo.cs.uwf.edu

Ronald A. DeVore, Ph.D.

Professor, Department of Math & Statistics
University of South Carolina, Columbia, SC 29208
(803) 777-2632, FAX (803) 777-3783, devore@oregano.math.scarolina.edu

David Donohoe,

Professor, Department of Statistics, Stanford University
Sequoia Hall, Stanford University, Stanford, CA 94305
(415) 723-3350, FAX (415) 725-8977, donoho@playfair.stanford.edu

Heber McMahon

Professor of Radiology, University of Chicago, Chief of Thoracic Imaging
University of Chicago, Radiology Department, MC2026, 5841 S. Maryland Ave., Chicago, IL 60637
(773) 702-4532, FAX (773) 702-1161, macm@midway.uchicago.edu

Michael W. Vannier, M.D.

Editor-in-Chief: IEEE Transactions on Medical Imaging
Department of Radiology, University of Iowa Hospitals & Clinics, 200 Hawkins Dr., 3966A JPP, Iowa City, IA 52242-1077
(319) 356-3372, FAX (319) 356-2220, michael-vannier@uiowa.edu

3. PRESENTATIONS AND ABSTRACTS