2007 Annual Scientific Report

From NAMIC Wiki
Revision as of 14:04, 30 April 2007 by Will (talk | contribs) (→‎Progress)
Jump to: navigation, search
Home < 2007 Annual Scientific Report

Back to 2007_Progress_Report

For reference:

1. Introduction (Marty Shenton)

2. Four Main Themes

This year's activities focus on four main themes: Diffusion Image Analysis, Structural Analysis, Functional MRI Analysis, and the NA-MIC Kit. Each of the following sections begins with an overview of the theme, provides a progress update and list of key investigators, and concludes with a set of links to additional information for individual projects in that theme.

These thematic activities involve scientists from each of the 7 NA-MIC cores (Appendix).

  • Core 1 Algorithms-Ross Whitaker PI
  • Core 2 Engineering-Will Schroeder PI
  • Core 3 DBP1-Martha Shenton PI / DBP2-Andy Saykin PI / DBP3-Steven Potkin PI
  • Core 4 Service-Will Schroeder PI
  • Core 5 Training-Randy Gollub PI
  • Core 6 Dissemination-Tina Kapur Co-PI; Steve Pieper Co-PI
  • Core 7 Leadership-Ron Kikinis

2.1 Diffusion Image Analysis Theme (Marek Kubicki, Guido Gerig)

Progress

Key Investigators

  • BWH: Martha Shenton, Marek Kubicki, Marc Niethammer, Sylvain Bouix, Katharina Quintus, Mark Dreusicke, Carl-Fredrik Westin, Raul San Jose, Gordon Kindlmann, Doug Markant
  • Harvard/MGH: Bruce Fischl, Denis Jen, David Kennedy
  • MIT: Lauren O'Donnell
  • UCI: James Fallon, Martina Panzenboeck
  • UNC: Guido Gerig, Isabelle Corouge, Casey Goodlett, Martin Styner
  • Utah: Tom Fletcher, Ross Whitaker, Saurav Basu
  • Georgia Tech: Eric Pichon, John Melonakos, Xavier LeFaucheur, Allen Tannenbaum
  • Dartmouth: John West, Andrew Saykin, Laura Flashman, Paul Wang, Heather Pixley, Robert Roth
  • Isomics: Steve Pieper

Additional Information

For details of each of the projects in this theme, please see NA-MIC Projects on Diffusion Image Analysis.

2.2 Structural Analysis Theme (Allen Tannenbaum, Martin Styner)

Progress

Shape Driven Segmentation

The characterization of local variations specific to a shape population is an important problem in medical imaging since a given disease usually only effects a portion of an organ’s surface. In particular, one of the driving biological projects that motivates our work is the study of schizophrenia. Yet the clinical study of schizophrenia is only now beginning to take concrete form, primarily because neuroimaging techniques are finally providing a sufficiently detailed picture of the structure of the living brain and tracking the way the brain functions in controlled experimental settings. One important aspect of such an analysis of schizophrenia is the segmentation and shape analysis of selected brain structures, such as the hippocampus or the caudate nucleus, in order to find differences between groups of healthy and diseased patients. An automated segmentation of such structures must therefore be highly accurate and include high frequency variations in the surface. Since shape representation is a key component of the segmentation, it must be rich enough to express shape variations at various frequency levels, from low harmonics to sharp edges. Additionally, a shape representation that encodes variations at multiple scales can be useful in itself as a rich feature set for shape analysis and classification.

Medical object segmentation with deformable models and statistical shape modeling may be combined to obtain a more robust and accurate segmentation. To address this issue, a decomposable shape representation targeted to the population seems natural, where the shape parameters are separated into groups that describe independent global and/or local biological variations in the population, and a prior induced over each group explicitly encodes these variations. Wavelet basis functions are useful for such a representation since they range from functions with global support to functions localized both in frequency and space, so that their coefficients can be used both as global and local shape descriptors, unlike Fourier basis functions or principal components over landmarks which are global shape descriptors. Our work presents three novel contributions for shape representation, multiscale prior probability estimation and segmentation.

Key Investigators

  • MIT: Kilian Pohl, Sandy Wells, Eric Grimson
  • UNC: Martin Styner, Ipek Oguz, Guido Gerig, Xavier Barbero
  • Utah: Ross Whitaker, Suyash Awate, Tolga Tasdizen, Tom Fletcher, Joshua Cates, Miriah Meyer
  • GaTech: Allen Tannenbaum, John Melonakos, Tauseef ur Rehman, Shawn Lankton, Ramsey Al-Hakim, Eric Pichon, Delphine Nain, Oleg Michailovich, Yogesh Rathi, James Malcolm
  • Steve Pieper, Bill Lorensen, Luis Ibanez, Karthik Krishnan, Michael J. Pan, Jagadeeswaran Rajendiran, Jim Miller, Karthik Krishnan, Luis Ibanez
  • Harvard PNL: Sylvain Bouix, Motoaki Nakamura, Min-Seong Koo, Martha Shenton, Marc Niethammer, Jim Levitt
  • Dartmouth: Andrew Saykin
  • UCI: James Fallon

Additional Information

For details of each of the projects in this theme, please see NA-MIC Projects on Structural Image Analysis.

2.3 Functional MRI Analysis Theme (Polina Golland, Andy Saykin)

  • Imaging Phenotypes in Schizophrenics and Controls (Turner, Kennedy, Potkin, UCI)

Functional connectivity of the DLPFC by genotype was investigated using partial least squares (PLS) correlation analysis. PLS is a multivariate analytical technique used to summarize large neuroimaging data sets in such a way as to correlate patterns of activation with a variable(s) of interest (i.e., DLPFC activity). In the most recent analysis, the DRD1 genotype was used as a grouping variable. This analysis has been submitted for publication (Tura, Turner, Fallon, Kennedy, and Potkin. Genetic Impact on Functional Connectivity in Schizophrenics During a Working Memory Task)

Working memory performance did not differ significantly between the two cohorts. However, imaging-genetic analysis showed a significant difference (P< 0.05) between the circuitry engaged by each group. Significance and reliability of the resulting imaging-behavioral patterns within each genotype were assessed by 200 bootstrap and 500 permutation tests, respectively. In one group, the circuitry included the temporal pole, the insula, the dorsolateral prefrontal cortex, and the Brodmann Areas (BA) 1,2,3,4,6,11 and 21, while the other group showed a network comprising the tectum, precuneus retroplenial, vermis, substantia nigra, BA 22,39,8, and 9. The DRD1 polymorphic site may characterize circuitry differences in schizophrenic patients.

  • Path-Of-Interest Analysis (joint DTI/fMRI modeling)

We collected preliminary data using an application of the “optimal path analysis”. In this analysis, we extracted group fMRI activation due to the Stroop effect (attentional experiment where incongruent color, in which the word is written, competes with name of the color itself, activating areas responding to conflict monitoring and selection) separately for controls and schizophrenics. This resulted in three clusters of activation, one in the right Dorsolateral Prefrontal Cortex, a second in the Anterior Cingulate Gyrus, and a third in the Medial Parietal Lobe. In the next step, we placed activation clusters in each individual space, by reversing normalization parameters used during fMRI analysis. Finally, EPI fMRI scans were co-registered to DTI scans, and the same registration parameters were applied to activation maps (fMRI results).

Regions of activation were used as start and destination points for optimal path analysis, which resulted in three separate paths of optimal connectivity for each subject. The probability of the connections were then calculated for each path and each subject, and compared between groups. In our preliminary analysis we included 10 control subjects and 10 chronic schizophrenics. Our results demonstrated a relationship between Stroop Effect fMRI activation in the medial parietal area and optimal path connectivity between parietal and cingulate activation sites in schizophrenics (rho=-0.56; p=0.047), which was not observed in controls. These findings suggest that decreased connectivity may result in schizophrenics relying more on posterior parts of the executive attentional network during performance of the Stroop task.

  • Hippocampal and frontal memory circuitry abnormalities in schizophrenia: Relation of diffusion, morphometric and fMRI markers

We performed a combined DTI, fMRI, and morphometric study on 13 patients with schizophrenia (SZ) and 14 HC. We identified areas of increased trace diffusivity (TD) in the hippocampal and insular regions as well as areas of reduced fractional anisotropy (FA) in left frontal white matter in SZ relative to HC (p<.01). Voxel based morphometry analyses in a subset of these subjects showed corresponding reductions in gray matter density in hippocampal and insular regions in patients relative to controls (p<.01). Analysis of fMRI results from the novel vs. repeated word contrast from the event-related auditory verbal episodic memory encoding/retrieval task in a subset of the subjects indicated reduced activation in frontal and temporal regions, as well as increased activation in posterior cingulate, retrosplenial, and thalamic regions in SZ relative to HC (p<.05). Further analysis showed that left frontal white matter FA was associated with activation in the left and right hippocampi as well as other frontal and temporal regions, but inversely related to activation in the retrosplenial/posterior cingulate region (p<.05). These initial findings indicate a pattern of relationships between of structural and functional brain abnormalities in schizophrenia and demonstrate the feasibility of integrated quantitative analyses across modalities.


West JD, Saykin AJ, Roth RM, Flashman LA, Koven N, Pendergrass JC, Arfanakis K. Hippocampal and frontal memory circuitry abnormalities in schizophrenia: Relation of diffusion, morphometric and fMRI markers. 13th Annual Meeting of the Organization for Human Brain Mapping, Chicago, IL, USA, June, 2007. Journal paper in preparation.

2.4 NA-MIC Kit Theme (Will Schroeder)

Progress

The continuing vision of the NA-MIC Kit is to provide an open source set of software tools and methodologies that will serve as the foundation for medical image computing projects for both academic and commercial use. Key elements of this vision are:

  1. Unrestrictive License. Users of the Kit are free to distribute their derived works under any license suitable to their needs.
  2. Cross Platform. This software set can be adapted to the best available price-performance computer systems for any particular use.
  3. Extensible Application Framework. New techniques and algorithms can be quickly integrated into a working system. Sophisticated user interfaces can be generated easily through automated processes. Sophisticated toolsets such as ITK, VTK, and KWWidgets are available to create and deploy applications quickly.
  4. Quality Software Process. Developers and users can rely on accurate and well documented behavior from all the parts of the Kit.
  5. Sustainable Community. Users are actively involved in the design process of the Kit. Documentation, training materials, and hands-on sessions are available and well publicized to the community.

Slicer3

A major focus of the third year was the implementation of the Slicer3 in the NAMIC Kit. This effort addressed Item #3 above Extensible Application Framework. The previous two years of the NAMIC project, which entailed gathering requirements from Cores 1 and 3, and developing the computational foundation, toolsets, and software process, came together in the Slicer3 application platform.

Core 2 worked hard to insure that the Slicer 3 application serves, and will continue to serve, as a productive technology deployment platform. The application framework was designed carefully to provide ease-of-use, both in terms of interaction and software integration. Advanced capabilities, including the ability to launch large-scale grid computing, was designed into the application. Some of the key features of the Slicer3 application completed in the third year include the following.

  • Advanced application framework including a tuned GUI for ease of use, undo/redo capabilities, 2D/3D view windows, and support for advanced interaction techniques such as 3D widgets. The application provides viewers for displaying slices, volumes, and models including the ability to edit properties. A built in transformation pipeline enables users to confidently import, edit and display data in a consistent coordinate system.
  • The application is data-driven based on the next generation MRML scene description file format. Backward compatibility to Slicer 2.x MRML files is preserved.
  • A module plug-in architecture and execution model that enables researchers to package and integrate their software into the Slicer3 framework. Plug-in modules can be implemented in a variety of programming languages, and are described using a simple XML description. These modules, when located and loaded into Slicer3, have the capability to automatically generate their user interface, which is seamlessly integrated into the Slicer3 GUI.
  • Support for editing and marking data including support for fudicials, paint and draw editors.
  • The creation of several simple plug-in modules, including the conversion of previous Slicer 2.x modules to the new Slicer3 architecture.

EM Segment

As an application framework, Slicer3 provides tools for loading, viewing, measuring, editing, and saving data. To support advanced medical image analysis, plug-in modules are required in conjunction with the Slicer3 core. To demonstrate the capabilities of the framework we implemented the EM Segment module, a sophisticated and proven method for automatically segmenting complex anatomical structures.

To use this module, users specify parameters defining the image protocol and the anatomical structures of interests. This process results in a template that the module uses to automatically segment large data sets. The template is composed of atlas data and a non-trivial collection of parameters for the EM Segment algorithm (Pohl et al.). Once the parameters are specified, the target images are segmented using the algorithm; if the results are satisfactory, the template is saved and can be used later to segment new images (via the GUI or batch processing). If the results are unsatisfactory, the parameters can be modified and the segmentation re-run.

Besides successfully demonstrating the use of complex algorithms in the Slicer3 framework, this effort also led us to develop tools, including modifications to the underlying KWWidgets GUI toolkit, to support module workflow. With these tools, it is possible to simplify complex modules by dividing the complicated template specification task into a number of smaller, intuitive steps. These steps are enforced by the GUI and reduce the potential for user error, while improving the overall user interface.

Quality Software Process

Key Investigators

  • GE: Bill Lorensen, Jim Miller, Xiaodong Tao, Dan Blezek
  • Isomics: Steve Pieper, Alex Yarmarkovich
  • Kitware: Will Schroeder, Luis Ibanez, Brad Davis, Andy Cedilnik, Sebastien Barre, Bill Hoffman
  • UCLA: Mike Pan, Jagadeeswaran Rajendiran
  • UCSD: Neil Jones, Jeffrey Grethe, Mark Ellisman
  • Harvard: Nicole Aucoin, Katie Hayes, Wendy Plesniak, Mike Halle, Gordon Kindlmann, Raul San Jose Estepar, Haiying Liu, Ron Kikinis
  • MIT: Lauren O'Donnell, Kilian Pohl

Additional Information

For details of each of the projects in this theme, please see NA-MIC Kit Projects.

3. Highlights (Will Schroeder)

3.1 Slicer3

  • Slicer3
  • EM Segmenter (http://wiki.na-mic.org/Wiki/index.php/Slicer3:EM)
    We constructed a module that integrates the EMSegment algorithm---an automatic segmentation algorithm for medical images---into the Slicer3 platform. The project was a joint effort between the NAMIC engineering, algorithms, and biological problem cores. As in Slicer 2, the user is able to adjust the algorithm to a variety of imaging protocols as well as anatomical structures. However, the configuration of the algorithm is greatly simplified in Slicer 3 as the user is guided by a new wizard-style workflow interface.
    A tutorial session for the EMSegment module was given at the January 2007 NAMIC all-hands meeting (notes and data available online at http://wiki.na-mic.org/Wiki/index.php/Slicer3:EM). The module is available with the beta release of Slicer3. Future work includes testing, validation, and the integration of data preprocessing steps into the EMSegment module.

3.2 Algorithms

  • As we published for the first time an open-source framework to do shape analysis in this NAMIC year, I think this is a highlight. It has been downloaded many times since the first online publication (around October 06) and it now used by several image analysis groups (not yet by the clinical researchers, we still need to package it nicely into Slicer v3). I updated our shape analysis page last week. The webpage with the information is here http://www.na-mic.org/Wiki/index.php/Algorithm:UNC:Shape_Analysis.
  • Highlights [Algorithmic Development and Software Transfer at Georgia Tech]:
    • The spherical based wavelet shape analysis package has been put into ITK, and in the next few months the multiscale segmentation work will be incorporated as well. We then intend to import this menu of algorithmsin to 3D Slicer. All of our algorithms are open source, and are intended to be user friendly to give them the widest possible accessibility.
    • We have implemented a very fast method for the optimal transport approach to elastic image registration. This will be made available via ITK in the next year.
    • Our tool for the semi-automated segmentation of DPFC areas in coronal sections of from brain MRI data developed in collaboration with our Core 3 colleague Dr. James Fallon is in Slicer 2. This will be ported to Slicer 3 in the next year as well.
    • ITK code has been for the conformal flattening procedure has been ported to an ITK filter and is in the NAMIC Sandbox.

3.3 Outreach

  • Papers- Number of algorithms papers (count the number of NAMIC MICCAI papers accepted - it was significant). In fact, all 3 DTI papers presented at MICCAI last year were NAMIC associated.
  • Highlight a tutorial or other training event. E.g. attendance at the UNC-DTI workshop was good and there will be another such workshop at HBM

4. Impact and Value to Biocomputing (Jim Miller)

4.1 Impact within the Center

4.2 Impact within NIH Funded Research

4.3 National and International Impact

5.NA-MIC Timeline (Ross Whitaker)

This section provides a table of NAMIC timelines from the original proposal that graphically depicts completed tasks/goals in years 1, 2, and 3 and tasks/goals to be completed in years 4-5. Changes to the original timelines have also been described.

2007 Scientific Report Timeline

6. EAB Report

The NA-MIC External Advisory Board (EAB), chaired by Prof. Chris Johnson of the University of Utah, met at the annual All-Hands Meeting. After individual presentations by NA-MIC investigators and open as well as closed-door EAB discussion, the Board provided its independent expert assessment of the Center (Appendix 2).

Logistics

Schedule and process for preparation of this report

  • March 30 - Assign section/theme leads (Ron). Last year: introduction (marty), structural analysis (allen, martin), dti (guido, marty/marek), fmri (polina, andy), namic kit(bill), timeline (ross), highlights (will), impact (bill).
  • April 7,10 - tcons with Marty, Ross, Will, Tina to finalize the layout, process, and timeline of report.
  • April 13 - update projects list using last year's NA-MIC_Collaborations and projects pursued at the half week in SLC. Remind investigators to update individual pages. (Tina)
  • April 23- complete project description pages in updated list: NA-MIC_Collaborations (all project owners).
  • April 30 - complete section summaries, introduction, highlights, impact, timeline (owners of these topics)
  • May 3 - submit wiki report to NA-MIC editor, Ann (Tina)
  • May 17 - submit Edited report to Rachana (Ann)
  • May 31 - ship final package to NIH (Rachana)

Guidelines from NIH Program Officer

The following guidelines were provided by Grace Peng, NA-MIC program officer, in Feb 2006.

The key is to synthesize all the individual elements into bigger picture stories that really speak of each area’s impact to the community.

The specialized scientific report should have the following format:

  1. Introductory page describing the new grouping of NAMIC project themes.
  2. A description of progress in each NAMIC project theme (not to exceed 2 pages each), tying together relevant activities from participating subcomponents and referencing cores in parentheses.
  3. A table of NAMIC timelines (from original proposal), graphically depicting completed tasks/goals in years 1,2, and 3 and tasks/goals to be completed in years 4-5. Changes to the original timelines should be described.
  4. A description of 3 highlights selected from all NAMIC projects to showcase NAMIC.
  5. A discussion of NAMIC’s impact and value to the biocomputing community this year.