Biomedical Informatics Research Network: Integrating multisite human functional imaging acquisition

1
Biomedical Informatics Research Network
Integrating multi-site human functional imaging
acquisition and analysisTurner, J.A.1 Potkin,
S.G.1 Brown, G.G.2Glover, G.H.4 Greve, D.3
Keator, D.B.1 McCarthy, G.5 Grethe,
J.S.2Wible, C.G.6 Lim, K.O.7 Toga, A.W.8
Andreasen, N.C.9 OLeary, D.10 FIRST BIRN111.
University of California, Irvine CA 2. University
of California, San Diego, La Jolla CA 3. Mass.
General Hospital, Charlestown, MA 4. Stanford
University, Stanford CA 5. Duke University,
Durham NC 6. Brigham Womens Hospital, Boston
MA 7.University of Minnesota, Minneapolis MN 8.
University of California, Los Angeles CA 9.
University of New Mexico, Albuquerque NM 10.
University of Iowa, Iowa City IA 11.
www.nbirn.net
Introduction
Methods Human Calibration
Methods Data storage and access

• Five male right handed subjects traveled to all
  ten scanning sites over the course of August and
  July 2003, and were scanned on two separate days
  doing the same tasks repeatedly. They served as
  human phantoms, being the same brain doing the
  same task.
• The primary tasks used were multiple repetitions
  or runs of the following
• Resting The subject laid with eyes open, 4.5
  minutes two runs per session.
• Breathholding The subject alternated between
  breathing normally and holding his breath every
  15 s, for 4.5 minutes two runs per session.
• Sensorimotor task The subject tapped his fingers
  and heard tones in time with a 3Hz flashing
  checkerboard (left), alternating with rest
  periods every 15 s, for 4.5 minutes four runs
  per session.
• This collection of multiple runs, visits, tasks,
  and scanners across a consistent set of subjects
  is a unique fMRI dataset that allows us to assess
  the relative variability introduced by each of
  these factors.

SCIENTIFIC OBJECTIVES OF THE BIOMEDICAL
INFORMATICS RESEARCH NETWORK (BIRN) Many research
groups at medical schools and universities in the
United States have projects in neuroscience
involving studies of the brains of humans and
animals in health and disease. The projects use
a variety of experimental and imaging techniques
and are focused at all scales from the molecular
to the whole brain. For most studies, the sample
sizes are small and restricted to specially
chosen populations. What has never been
attempted is the intelligent combining of data
from such studies. To do this, groups of
neuroscientists must be ready to account for
differences in technique, equipment, species and
population focus and to overcome the challenges
of accessing and sharing large datasets. BIRN
represents the first attempt to develop a
"protocol" for this kind of collaborative
research among neuroscientists and medical
scientists. It depends on the new computational
and networking technologies that have been
developed to bring researchers together over the
Internet, and on still newer technologies for
federating data from multiple sources. It also
depends on the social integration of the
scientific groups that will be sharing their data
and, ultimately, making their collective results
public. Thus BIRN is a test of a new mode of
doing large-scale medical science research.
BIRN leverages a number of ongoing studies in
the hope of improving both biomedical insight and
statistical accuracy. BIRN is developing useful
ways of sharing data rapidly over the Internet,
with appropriate controls over data and patient
confidentiality and privacy, under model
agreements among the many scientists involved.
BIRN is constructed to take advantage of three
revolutions the revolution in biology that can
link processes and events from the molecular to
the behavioral scale the revolution in computers
that can link researchers across distances,
databases to databases, and data-gathering
facilities to large-scale analysis facilities
and the revolution in technology that has led to
advanced, high-speed, and in vivo noninvasive
brain imaging and 3-D microscopic imaging
techniques.
The data are complex Experiments have many
subjects, each subject may undergo several
different procedures, and the same subject may be
in different experiments. The project
information is stored and can be spliced into the
subject specific information. Individual
scanning sessions are stored, as well as
statistics compiled from the data and
subject/session specific annotations.
The imaging data is uploaded to the SRB and
shared with all the consortium sites using the
tools described above. Analysis occurs on a site
by site basis. All sites have access to all the
human phantom data and can download and analyze
it as they wish. The current goals are to
assess the variability within the imaging data
and develop methods to counteract it.
Functional MRI Results
ABOUT THE FUNCTIONAL IMAGING RESEARCH IN
SCHIZOPHRENIA TESTBED FIRST BIRN The FIRST
BIRN is developing and validating functional
magnetic resonance imaging (fMRI) methods to
combine data from different sites using a common
protocol, for a large-scale multi-center study of
schizophrenia. The proposal is crystallized
around a high-speed broadband network supporting
a federated database, specifically developed for
the pooling and sharing of data across sites, to
facilitate large-scale hypothesis testing and
intelligent data mining to address complex
scientific problems. Each center will also
conduct its own unique studies on the same
patients being studied by the FIRST BIRN
consortium, to provide added value. Sites
participating in the FIRST BIRN are shown in the
image to the left.
Sensorimotor task Site effects
Methods XML application to Image QA Calibration
The differences between sites in the overall
amount of brain which shows an activation to the
task, with the same 5 subjects doing the task in
each scanner. 1.5T scanners are shown on the
left 3 and 4T scanners on the right. Even
within the 1.5T scanners, there are site to site
differences in fMRI sensitivity, which need to be
accounted for before data can be combined.
Non-human phantom scans are used to (1) perform
a one time system check to test a scanners
geometric accuracy using a specialized
cylindrical geometry phantom and (2) perform
ongoing quality assurance scans using a spherical
agar phantom to verify inter-image scanner
stability and drift. The phantoms, protocols, and
measures were developed at Stanford the phantoms
were built at UCSD the measures were made
useable across different file formats by scripts
developed at UCI. The data obtained from scanning
the spherical agar phantom are analyzed to
measure the Signal to Noise ratio, the drift and
fluctuation over time, and spatial correlations.
a) One of the subjects doing the sensorimotor
task in a 1.5T scanner b) the same subject and
task in a 3T scanner. The stronger field strength
shows the expected increase in sensitivity to the
BOLD response, seen here in the motor and
auditory areas. (Data from Duke and MGH images
from MGH.)
The advances of the FIRST BIRN federated database
will specifically facilitate this combining of
site-specific data with the consortium data.
This value added sharing will enhance the
FIRST BIRNs scientific discovery process. GOALS
OF THE FIRST BIRN The FIRST BIRN has two major
goals Technological development of a
distributed network infrastructure that will
support the creation of a federated database
consisting of large-sample fMRI datasets
contributed by the 10 centers. Clinical
Study brain function in schizophrenia with fMRI.
We propose to use the technology developed herein
to study changes in brain function in the
development, progression, and treatment of
schizophrenia as assessed by fMRI. We will
investigate two domains that have been
demonstrated to be dysfunctional in
schizophrenia early auditory sensory processing,
and working memory/executive functioning.
Collaborative studies will focus initially upon
first onset, chronic, and late-onset
patients. FIRST BIRN will perform a.
Standardization and calibration of equipment and
imaging activation paradigms across sites using
mechanical and human phantoms b. Collection of
imaging data using the consortium fMRI protocol
on populations of persons with schizophrenia at
different sites c. Combining of unique imaging
data collected with diverse activation methods
into the federated database. We present here the
initial progress toward the standardization and
calibration of equipment and activation paradigms
made by the FIRST BIRN over the past year.
Example QA results from two different 1.5T
scanners at two different sites. Three different
measures can independently indicate the
functionality of the scanner, detecting drift and
increases in noise, as well as malfunction prior
to component failure. The collaborative efforts
in sharing this data have forced some consistency
across the sites, since the evidence regarding
each scanner is available to all. Scanner
manufacturers are using these data to improve
imaging performance.
Integration of different methods and measures
The kind of data the FIRST BIRN will be combining
across the country a) Cortical activations to a
thought or perception, from an individual b)
Measures of group differences in activations on a
partially-inflated cortical model c) The
macro-circuitry of the brain involved in
schizophrenic syndromes d) The micro-circuitry
of the cortical layers e) Areas with differences
in cortical thickness over the brain, between
patients and controls f) Data from other species
or studies regarding cell types, locations,
receptor densities, etc.
e)
c)
a)
d)
b)
f)
XML Common Data Interface The XML common data
interface header describes the data and metadata
of each FIRST BIRN native image file format. The
XML header now encapsulates DICOM and accounts
for those differences in DICOM implementations
between Siemens, GE, and Picker. Other currently
supported formats include Stanfords F-files,
Analyze, GE, and raw formats. The XML common
data interface allows for collaborative access to
the imaging data without the need for format
conversions which often result in loss of data
resolution and/or image metadata. The XML data
interface is being extended and modularized to
capture domain specific information about
studies, protocols, patients, data provenance,
statistical outcomes, and derived data. Below is
an example of the use of XML in the BIRN QA
process.
Example Query of Federated Database
Data Integration A users query to the federated
database requires integrating the different kinds
of data shown above. The software architecture of
such integration systems consists of a two-part
middleware, called the wrapper and the mediator,
that are between the information sources and the
user. The wrapper converts the data from the
respective information source to a form that the
mediator can accept and manipulate. The
mediator converts a users query into smaller
sub-queries that are sent to each source, and to
integrate the results returned from each source.
The integration of the returned data is
directly dependent on the ontological information
available. The FIRST BIRN group is actively
working to define the ontology describing patient
populations and experimental characterizations
needed to share imaging data.
Are chronic, but not first-onset patients,
associated with superior temporal gyrus
dysfunction?
Methods Datasharing
Integrated View
Human Privacy Issues The concept of a federated
database of human data is unusual. A more common
model is a centralized database, where the data
may be collected at several sites but stored in
one physical location. The centralized model
allows for a single entity (a university,
company, or foundation) to be responsible for
following all federal, state, and local
regulations regarding human subject data. The
challenge to the FIRST BIRN is to develop
precedents for collecting human neuroimaging data
and storing it in multiple locations, with
approval to allow access by other sites, who did
not necessarily collect the data and whose
identity may not be known at the time the data
are collected. The solution was to request
authorization from the subjects to upload the
images and the Protected Health Information (PHI)
necessary for the purposes of the study into the
BIRN database. That the data would be put in the
database for those members currently in the BIRN
consortium and for those to join later, was made
explicit in the protocol and in the consent form.
Sites whose IRBs require HIPAA authorization
forms for these data will use them. The FIRST
BIRN has decided as a policy to request
permission from the IRB and authorization from
each subject to upload to the BIRN database as
much information as possible, with the explicit
statement that the subjects name, address, and
Social Security Number will not be in the BIRN
database, either with the imaging data or with
the clinical assessment data.
Mediator
Wrapper
Wrapper
Web
Wrapper
Wrapper
Wrapper
Wrapper
PubMed, Expasy
PET fMRI
Clinical
ERP
Receptor Density
Structure
This research was supported by the Functional
Imaging Research in Schizophrenia Testbed (FIRST)
Biomedical Informatics Research Network (BIRN,
http//www.nbirn.net), which is funded by the
National Center for Research Resources at the
National Institutes of Health (NIH).