P1252109103Qemzu

1
University of Washington Human Brain Project
2004 Representative Projects
Supported by Human Brain Project grant R01
DC02310
James F. Brinkley, David Corina, Dan Suciu,
Andrew V. Poliakov, Richard F. Martin, Kevin P.
Hinshaw, Eider Moore, Hao Li, Xenia Hertzenberg,
Veronica Smith, Erin Gibson, Chris Re, Etorre
Lettich, Hansang Cho, Linda Shapiro and George
Ojemann
Uses for Web-based Visualization of Brain Data
Next Generation Visualization and Mapping
Evaluating Anatomical Normalization Methods
Embedded Data Management for fMRI Analysis
MRI from database
BrainJ3D is a portable, Java3D software toolkit
for brain visualization and mapping that will
permit more widespread distribution of the tools
we have developed to date (left hand panel). In
addition, it is designed to be general-purpose,
flexible and interoperable with popular
neuroimaging software packages (SPM, FSL etc.)
As reported earlier, we have developed a web
interface for our interactive 3D visualization
and analysis tools for integrated brain data. We
utilize surface-based rendering of the cortex and
other 3D models. Functional data from various
modalities can be overlaid onto individual
subjects anatomical models. Both scene
rendering and data storage are performed on the
server, while a simple web client can run on
virtually any computer with internet access.
Goal Provide transparent data management and
batch processing for users of the popular SPM2
fMRI analysis package. Solution X-Batch, a
plug-in utility for SPM2 which links user-defined
SPM2 batch scripts to the Data Management
ontology developed by the Dartmouth fMRI Data
Center. The fMRIDC ontology is implemented in the
Protégé knowledge acquisition framework developed
at Stanford .
P54

• Features
• Supports neuroimaging data
• structural MRI volumes functional volumes
  (fMRI, ERP etc.) surfaces (cortex, veins,
  arteries etc.) surgical photographs maps
  (collections of labeled points)
• Written in Java/Java3D
• cross-platform
• hardware accelerated
• Supports remote visualization client/server
  and standalone modes one server can serve
  multiple clients uses the Java3D off-screen
  rendering

Features Ø Implemented as an SPM toolbox Ø
Accesses the fMRIDC ontology through the Protégé
API Ø Allows user definition and storage of
batch processing parameters Ø Processing
sequences may be applied to multiple data sets Ø
Metadata are collected automatically and stored
in the ontology Ø Results files are saved to
directories defined by the protocol
Left Hemisphere w/ CSM sites
Flat Map

• Motivation
• A valuable quantitative approach for relating
  anatomy and function of multiple human brains
  employs the deformation or anatomical
  normalization of individual target brains to a
  canonical brain chosen as a template.
  Neuroscientists have a variety of anatomical
  normalization methods to choose from. The method
  selection impacts result quality and study
  validity. Therefore, it is important to establish
  criteria of success and validity of statistical
  models for anatomical variation appropriate for
  the specific research project.
• We evaluate two anatomical normalization methods,
  CARET and SPM2, for comparing and analyzing data
  collected from epileptic patients who have
  undergone neurosurgery. During surgery, the
  cerebral cortical surface is electrically
  stimulated in multiple locations (CSM sites) to
  identify cortical loci associated with language
  disruption. The CSM site data for 80 patients
  have been entered into a relational database. By
  anatomically normalizing these patients brains
  with the CSM site data, neuroscientists believe
  that the remaining functional variation of the
  language sites may exhibit organizational
  patterns that are not apparent otherwise,
  providing more insight into language function in
  the human brain.
• Methods
• Normalize 20 brains to the colin27 brain atlas
  using both methods. Criteria for success includes
  preservation of known anatomical locations and
  reduction of anatomical variation between CSM
  language sites. Post-normalization, we record 3-D
  coordinates of CSM sites in MNI152 coordinate
  system.
• Metrics
• The smaller the average distance between
  normalized CSM sites, the smaller the observed
  anatomical variation. A neuroanatomist compares
  pre- and post-normalization anatomical locations
  of CSM sites. The greater the number of sites
  that remain in the same anatomical location after
  normalization, the better the preservation of
  anatomical locations.
• Preliminary Results

SPM2
Screenshot of the Brain Visualization client
This approach helps facilitate remote
collaboration and data sharing among a
distributed group of researchers, and our
collaborators find it productive and convenient.
This system is now being used in several
neuroimaging studies by research groups that
utilize various techniques.

• SPM toolbox
• Default ontology loads automatically
• View/delete/run previously defined processing
  sequences (protocols)
• Create new processing sequences

Pre-surgical visualization of fMRI data on the
reconstructed cortex surfaces In an ongoing
study, language mapping data are collected from
patients undergoing neurosurgery for intractable
epilepsy. Structural and functional MRI data are
acquired prior to surgery. During the surgery,
cortical stimulation mapping is performed and
single neuron activity may be recorded from areas
that will be resected. Reconstructions of 3D
models and fMRI analysis can now be performed
prior to surgery and the neurosurgeon can access
this integrated data on the web before or during
the surgery. In particular, this helps planning
experimental recording of single neuron activity
in the areas that were found to be active during
language tasks, as revealed by fMRI.

• Visualization
• Slice viewer and 3D surface rendering
• Cutaway view utilizes structural volume and
  brain mask
• Overlay of functional volumes onto the
  structural data
• An arbitrary number of volumes can be overlaid
• fMRI can be projected to the cortical surface
• Multiple models can be displayed at once
  (Cortex, Veins and Arteries)
• Extracts surfaces from masked volumes

• Batch script creation
• No programming necessary
• Protocols consist of processing stages, each
  stage defined by its parameter set
• Use previously created parameter sets, define
  new ones, or use a combination
• Specify order of processing
• All parameters saved into the ontology

SPM2 with x_batch

• Batch processing
• Process any number of datasets from beginning
  to end
• Relevant result files are moved into a
  designated directory

TMS localization on cortex models Another group
of researchers is evaluating TMS (Transcranial
Magnetic Stimulation) in severely depressed
patients as a potential treatment technique, a
less traumatic alternative to ECT ("shock
therapy"). Using our tools, they are able to
localize the position of the coil (yellow) with
respect to the individual patients cortex model
and show its projection on the cortex surface
(blue)

• Mapping
• Map points using pick operation
• Assign any text or number label to point
• Display label text and/or a choice of shapes
• Nodes can be arbitrarily grouped
• Each group has its own appearance.
• Can arbitrarily place points in space

Data management In the background, metadata
are collected and stored in the fMRIDC Data
Management Tool
P54
P117
Surface-Based Method CARET

• Integrating functional data of multiple
  modalities
• In a study of Autistic patients, researchers are
  collecting functional data of multiple
  modalities, including
• Spectroscopy (PEPSI)
• Electromagnetic tomography (EEG/ERP source
  localization)
• functional MRI

• Workspace
• Organizes and manages patients data
• XML format, easy to customize

• Current and Planned Work
• Incorporate results from other applications
• Manage coordinate systems in workspace
• Spatial Queries
• Basis for further work in visualization and
  mapping

Volume-Based Method SPM2
Metadata automatically stored into the ontology
Defining new protocol
ERP source localization
Colin Atlas with normalized CSM sites