Core 3

Core 3.1 (Harvard-VA/Dartmouth) Response

Responses from Core 3.1 to the NAMIC Center Team of NIBIB are provided below, where the the question or concern is listed in boldface and italics, and our response follows in regular text.

The Center Team requests an electronic version of Table 3.11 and 3.21 on pages 486 and 546 of the original proposal.

These are included for both Core 3.1 and Core 3.2. File:P486-546 final-submission-01 24 04.pdf

Please provide a table of interactions for Core 3 sites with other NAMIC Cores.

Given the large number of interactions for Core 3.1, we are providing a summary of the interactions between Core 3.1 and the other Cores rather than try to put all of this information into a table. For year 2 of the progress report, however, we will put in separate tables for each of the interactions between Core 3.1 and the other Cores.

Core 1 interactions and collaborations with Core 3.1.

• Formatted and distributed data sets to the NA-MIC community. This involved a total of 188 structural scans and 80 diffusion tensor scans with unique identifiers. A great deal of time was spent by Sylvain Bouix (Harvard) in converting data sets to formats that could be easily read by collaborators.

• Kilian Pohl and Eric Grimson (MIT) and Sylvain Bouix (Harvard) designed and optimized an EM segmentation algorithm to classify our morphological MR images into gray matter, white matter, and CSF. The method is now used as Core 3.1’s primary segmentation technique.

• Lauren O’Donnell and Eric Grimson (MIT) worked closely with Carl-Fredrik Westin and Marek Kubicki (Harvard) to develop better visualization and analysis tools of fiber bundles. One project is the analysis of anisotropy measures along fibers, including left and right fornix, uncinate fasciculus, corpus callosum, and occipito-frontal fasciculi. White matter tractography was performed using the Runge-Kutta method to interpolate paths (fiber traces) following the major directions of diffusion, where traces were seeded only within the defined regions of interest. Next, a fully automated procedure was applied to the fiber traces, grouping them according to a pairwise similarity function which takes into account the shapes of the fibers and their spatial locations. Results showed the ability of the clustering algorithm to separate several fiber tracts which are otherwise not easy to define. Multiple hypotheses regarding specific fiber connections (i.e., the 5 regions of interest highlighted for Core 3.1 – can be seen in Table 3.11, see hypertext link, above) and their abnormalities in schizophrenia.

• Algorithm development of non-rigid EPI registration using ITK (MGH, Kitware) for use by Andy Saykin and colleagues (Dartmouth). This collaboration between Dave Tuch (MGH) and Andy Saykin (Dartmouth) has developed algorithms for robust non-rigid, intrasubject registration between an EPI image and a conventional structural image. This has been tested on data sets from Dartmouth for effective application of structural analysis tools in regions of interest.

• Algorithm development of ITK system for quantitative fiber tract analysis (UNC, GE, MIT, MGH) for use by Shenton and colleagues (Harvard). This collaboration across multiple sites is developing a complete tool kit for analysis of white matter tracts extracted from DT-MRI images. This includes tracking of fibers, clustering of tracks, and measurement of features on track clusters. The prototype system has been tested on data from the Brockton repository. This set of tools will be integrated with the fiber bundle tools mentioned previously to allow for careful measurement of properties of a number of structures of interest: including left and right fornix, uncinate fasciculus, corpus callosum, and occipito-frontal fasciculi.

• Several ongoing collaborations are working to develop or refine connectivity analyses central to Core 3.1 Specific Aims. David Tuch (MGH) and Andy Saykin (Dartmouth) have had ongoing discussions regarding novel strategies for integration of DTI and fMRI connectivity data. Tuch and colleagues at MGH have developed a prototype tool for analyzing connectivity among a set of anatomically- or functionally-defined regions of interest. Plans are underway for installing this tool at Dartmouth for further testing and application to initial combined DTI and fMRI data sets acquired on the same participants. The Dartmouth team has also had ongoing discussions with Sandy Wells (BWH/MIT) and his students regarding novel approaches for modeling connectivity among targeted ROIs within fMRI time series data sets. The work by Wells and colleagues may have important design implications for optimized fMRI task timing when capturing connectivity related information is a goal.

• The UNC (Gerig, Styner) and Dartmouth teams (Saykin, West, Shen) have held discussions and exchanged technical information by email regarding software approaches and file formats for manual segmentation of ROIs required for volume and shape analyses, as well as possible collaborative work to enhance SPHARM shape modeling.

Core 2 interactions and collaborations with Core 3.1.

• Alexander Yarmakovich and Steve Pieper (ISOMICS): Development of a new module in Slicer to improve image resampling and registration. The module was based on specifications given by Sylvain Bouix and it can handle multiple transforms and single click batch resampling.

• Gordon Kindlmann (BWH), Bill Lorensen (GE) and Steve Pieper (ISOMICS): Standardization of diffusion MR data format based on specifications and needs from Core 3.1, highlighted by Sylvain Bouix. A single format (NRRD) can now handle structural data as well as diffusion data. The format allows for an arbitrary number of dimensions, and has information about the patient orientation as well as all information regarding gradient directions and strength in the header. This format is developed and maintained by Gordon Kindlmann and has been incorporated into ITK by Bill Lorensen.

• Functionalities of Slicer, with ISOMICS. Core 3.1 initiated the addition of a resampling module into Slicer and extensively tested the linear and non-linear registration as well as automatic segmentation modules. Core 3.1 also pushed for more flexible file format readers in Slicer. These components were part of the latest version of the software presented at the dissemination seminar at Dartmouth, May 26-27.

• Katie Hayes (BWH) and Steve Pieper (ISOMICS) and the DTI schizophrenia group in Core 3.1 have regular and ongoing interactions for bug removal in Slicer as well as testing of new features. Researchers can now report “bugs” they notice while using software on the Wiki and these problems are discussed weekly at DTI/Slicer development meetings, where the website is updated weekly, though not during the summer. This website also includes a wishlist of the neuroscience investigators, for the developers.

• At Programmer’s Week in Boston, Core 3.1 personnel provided structural, DTI, and fMRI test cases for developers. Core 3.1 also offered suggestions and goals for modules and updates in Slicer to be developed by the programmers.

• Core 3.1 members (A. Cohen, C. Wible, K-U Lee) met with Wendy Plesniak (BWH) to discuss development of the “fMRI engine” module in Slicer. The discussion focused on adding functions to the fMRI engine in order to make it comparable to SPM.

Core 4 interactions and collaborations with Core 3.1. No interactions at this point.

Core 5 interactions and collaborations with Core 3.1.

• Core 3.1 members participated in the first user training workshop in May 2005 at Dartmouth where Sylvain Bouix gave several presentations on how Slicer has been used at Harvard and also gave hands on training sessions on manual and automatic segmentation, as well as registration. Three research assistants, A. Cohen, M. Dreusicke, and L. Rosow gave individual help during the hands on sessions. These three research assistants were instrumental in assisting researchers in how to use Slicer.

• Core 3.1 members provided internal training material to Sonia Pujol to help Core 5 prepare for the user workshop presentations.

Core 6 interactions and collaborations with Core 3.1.

• Work with Steve Pieper has involved testing some of Slicer tools to try to remove “bugs” before tools are disseminated to NAMIC community and beyond.

Core 7 interactions and collaborations with Core 3.1.

• Work with Steve Wong and coworkers has involved applying a 76-space analysis of gray matter diffusivity to normal and schizophrenic data sets provided by Core 3.1. This technology enables the parcellation of MRI gray matter into 76 regions and it also makes possible the measurement of apparent diffusion coefficient in each segmented region.

The Center Team requests a detailed breakdown of the tasks/aims assigned to the four Core sites, and the progress associated with each site. ''''' This information is provided above for Core 3.1 under collaborations between Core 3.1 and the other Cores.

Specifically, what is the progress the four studies in Core 3 towards driving the computational goals of NAMIC?

This information is also provided above for Core 3.1 under collaborations between Core 3.1 and the other Cores, which involve driving the development of computational tools to address the specific hypotheses of the studies proposed.

Please delineate the year 2 goals for Core 3.

Below, we delineate goals for year 2 for Core 3.1.

• Lauren O’Donnell, Carl-Fredrik Westin and Eric Grimson will collaborate on developing better visualization and analysis of fiber bundles to address the fronto-temporal hypotheses of Core 3.1. One immediate project is the analysis of anisotropy along the fiber bundles (e.g., along the uncinate fasciculus, left and right fornix, corpus callosum, and occipito-frontal fasciculi).

• Tom Fletcher and Ross Whitaker (U of Utah) will test a new statistical method to analyze diffusion tensor images. The group visited Harvard in March of 2005. This is an ongoing collaboration with expected results in year 2. The contact person at Harvard is Marek Kubicki.

• Martin Styner and Guido Gerig (UNC) will investigate new shape analysis of caudate data with Sylvain Bouix and Martin Styner who met most recently during Programmers’ Week in Boston.

• Ramsey Al-Hakim and Allen Tannenbaum (GTech) will develop semi-automatic techniques for segmentation of brain areas. The method is currently being tuned for the segmentation of the basal ganglia as defined by Jim Levitt in Core 3.1 (Harvard).

• Katie Hayes and Steve Pieper will continue to work with Core 3.1 in tracking bugs and in new functionality testing of Slicer.

• Alexander Yarmakovich and Steve Pieper will create a registration module which can perform rigid, affine and non linear registration between multiple different modalities based on ITK. The functionality already exists, but an effort for better user interface and easier integration within the Slicer core framework needs to be done. This work will be done in close collaboration with members of Core 3.1.

• Kilian Pohl and Steve Pieper will implement a single click brain tissue classification module based on the EM segmentation algorithm in conjuntion with members of Core 3.1.

• fMRI data will be collected at both Dartmouth and Harvard to complete data sets prospectively.

• New hires Marc Niethammer and Katharina Quintus (Harvard) are needed to facilitate the interface with Core 3 and Core 2. This interface is critical as Katharina will work on customizing software for Core 3.1. needs as well as perform batch processing, and Marc will serve many of the functions that Sylvain Bouix has served so successfully in the first year of NA-MIC.

Please provide the animal studies approval details for the study mentioned on page 177.

• In the context of a potential future development, we mentioned in the Progress Report that the Dartmouth team is involved in a NICHD-sponsored fMRI study of developmental brain injury using a piglet model. The ongoing piglet study has the requisite IRB approval at Dartmouth. We believe that in future work the NAMIC Slicer package and other tools could be very helpful in analyzing these novel data, particularly after we add DTI acquisition when our 3T magnet is installed this fall. Should others wish access to piglet data for software testing, we will obtain all necessary approvals for data sharing prior to uploading any animal data.

Core 3.2 (UCI/Toronto) Response

Please provide a table of interactions of Core 3 sites with other NAMIC cores.

• Core 1 (Algorithms): UCI formatted and distributed data sets to the NA-MIC community. This included a combined dataset of structural, functional, PET, clinical, and genetic (SNP) data.
  • Jim Fallon and Martina Panzenboek met several times in person and over the phone with the Georgia Tech personnel, in order to translate domain expertise in cortical segmentation into automated rules (see more detail below).
  • UCI has provided Dartmouth with an agar phantom courtesy of FBIRN so that they can calibrate their new 3T scanner.

• Core 2 (Engineering): The probabilistic DTI atlas (see more detail below), in which UCI is offering neuroanatomical expertise and development of training sets, has been being developed in Slicer. This has required substantial interaction with the Core 2 team at BWH as Core 3.2 has pushed for the development of improvements to the interface and ease of use.

• Core 4 (Service): No interactions with Core 4.

• Core 5 (Training): Core 3.2 members have participated in several of the Slicer workshops, including the San Diego programmers course, and the Dartmouth workshop.

• Core 6 (Dissemination): Core 3.1 hosted the Slicer workshop in Irvine, CA, March, 2005. This workshop included non-NAMIC investigators from BIRN, and from the general imaging community at UCI.

• Core 7 (Leadership): Core 3.1 has been in constant contact with NAMIC leadership as needed for organization, administration, and communication.

Specifically, what is the progress the four studies in Core 3 towards driving the computational goals of NAMIC?

The ability to correctly identify cortical and subcortical structures is a prerequisite to circuitry analysis, either through DTI or functional scans. The collaborative work with Cores 1 and 2 has helped determine those methods. The UCI Core 3.2 (Fallon) and Georgia Tech Core 1 (Tannenbaum) team has successfully developed a new semiautomated anatomically accurate cortical segmentation method.

Fallon identified lengthy and complex neuroanatomical rules for defining a cortical areas and in collaboration with Tannenbaum creating a program that is anatomically accurate, but takes only a fraction of the time to carry out. Previously, manual tracing of the DLPFC required 1-2 hours; with this method it can be done in 3 minutes.

The algorithm was originally developed in Matlab and returned a VTK file that is a 3D model of the DLPFC. These segmentations can be viewed in 3D Slicer of Brigham and Women’s Hospital (Core 2); future work is to implement the algorithm into Slicer.

This NAMIC Core 1-Core 3 collaboration was facilitated by two face to face conversations and presentations at two NAMIC meetings, and through individual and conference calls, and email. In June of 2005, Allen Tannenbaum and Ramsey Al-Hakim traveled to Fallon’s lab for four days to expand and deepen the Core 1- Core 3 interaction in order to determine how the neuroanatomical rules, i.e., both the quantitative definitions but importantly the more qualitative and intuitive features of the neuroanatomist’s perceptions, estimations and determinations of location, shape, ‘neighborhood’ rules can be automated in such a way that retains the precision, accuracy, (and variability) of neuroanatomical expertise but which requires undergraduate level expertise at only a fraction of the time necessary for the manual segmentation.

The FBIRN legacy dataset of 300 subjects with imaging and genetic data that was planned is currently being collected and will be available in Year 2. A combined clinical/imaging/genetic dataset of schizophrenic subjects has been made available to NAMIC algorithm developments.

Please describe further how NAMIC algorithms will help represent genotypes and phenotypes and how genetic, imaging and clinical data will be combined.

There are a large number of tools available for identifying haplotypes related to various illnesses, including the DBP of schizophrenia. A visualization of haplotypes is shown in the figures in the progress report included below. Core 1 will refine and extend the current Toronto Phase v.20 algorithm to study the more complex gene-gene interactions and epistasis. The implementation of this algorithm will allow the exploration of more complex genotypes. Using data mining strategies, we will be able to cluster these more complex genotypes with functional imaging circuitry, as well as clinical variables.

Please delineate the year 2 goals for Core 3.

Core 3.2: Through coordination with primarily Cores 1 and 2, in the next project year, we will continue with our proposed timeline to

1. implement the DLPFC segmentation algorithm into 3D Slicer of Brigham and Women’s Hospital,
2. apply the semi-automated techniques to other forebrain structures to determine the level of organization at which the circuitry produces different schizophrenic syndromes,
3. refine and make robust the in-house multi-gene analysis software, and
4. use PLS analysis to explore the functional connectivity between DPFC structures and other variables, such as behavioral performance and genetic profile.

Specifically, does NAMIC have interest in single cell measurements of biochemical, genetic regulatory networks related to schizophrenia (e.g. dopamine-related networks)? This would help to link the bottom-up approaches (e.g. Greengard studies) with the topdown or more general approaches in schizophrenia research. It would also help to better explain the genotype-phenotype relationship.

Core 3.2 investigators have done original work in post-mortem schizophrenic brains, in collaboration with Paul Greengard, identifying decreased DARPP-32 in the prefrontal cortex. DARPP-32 is involved in dopamine regulation and other biogenic amines, modulating both D1 and D2 receptor function (Albertka, et al., Arch Gen Psych, 2002). These insights on the role of dopamine and glutamate in DLPFC dysfunction in schizophrenia have informed the Core 3.2 aims and targeted the DLPFC as a region of interest. The current resolution of fMRI and DTI does not allow a one-to-one mapping to single-cell recordings; however, the concepts have been most useful.

Progress Report: UCI/Toronto

A portion of the progress report for Core 3.2 was not included in the original submission through scanner error. The entirety of the report is included here, some of which was noted to answer the reviewer questions above. File:UCI Toronto2005.pdf