|
|
Teleological Modeling
This project is an effort to implement major changes in the way complicated
imaging experiments are conducted. In a modern imaging experiment (e.g.
confocal microscopy, CT, MRI, EM) the number of experimental parameters
requiring adjustment is often large and the effects of altering one, much
less several of them, is seldom obvious. Moreover, the resulting images
can be large; are often multi-dimensional (2, 3, or 4D); and may be scalar,
vector, or tensor valued. Teleological modeling is a facile way of dealing
with the complexities of both data collection and analysis where the data
collection and final image goals of the investigator are algorithmically
incorporated into a set of interactive computational tools. By fashioning
these tools into a "teleological pipeline", we will be developing a more
efficient environment for optical and MRI data collection, analysis, rendering,
and viewing. By a teleological pipeline we mean a computational and instrumental
device in which the user's goals (i.e. the qualities they wish the final
image to possess) guide the data collection and processing procedures.
The user's goals are figuratively entered in one section of the pipeline
and, ultimately, the images appear in another section where they can be
evaluated by the user, the goals modified, and the process repeated. In
this way, we are 'closing the loop' between the data collection, visualization,
and interpretation phases of the imaging xperiment. The teleological
pipeline is a robust process of converting goals into computational procedures.
Its application in this project will not only greatly increase the efficacy
with which data can be collected and analyzed, but will also point the
way to its application in other complex methodologies used in the neurosciences.
At the analysis level our goal-directed approach will enable us to increase
the information extracted from the multi-spectral MR and optical images
via the development of new methods for identifying tissues, volumetrically
rendering those different tissues, and creating animations. The same tools
will be useful for developing new and faster MR imaging methods by specifying
low level goals such as relationships between individual parts of an imaging
sequence and the characteristics of the final image (e.g. the effect of
TR and TE on image contrast, the relationship between the size of a diffusion
gradient and the resulting amount of diffusion weighting in the image,
partial volume effects due to under-sampling), or automatically compensating
for hardware limitations (e.g. non-linear gradients, inhomogeneous B1 fields).
Use of teleological modeling coupled with novel techniques of rendering
vector/tensor data (e.g. painterly visualization) and the semi-immersive
Responsive Workbench rendering environment will make complicated experiments
and the resulting images truly accessible to the working neuroscientist.
Hierarchical Multi-Resolution Image Matching
The elaboration of structure and function in the brain takes place through
a range of spatial scales (nanometers to centimeters) and temporal scales
(microseconds to months). Imaging modalities are typically limited to one
to three orders of magnitude in both time and space because of the physics
of the phenomenon being applied. Thus, the collection and analysis of the
diverse scales of information needed to produce a complete description
of brain development requires multiple imaging modalities. A new thrust
of our efforts is to investigate and implement methods of dealing with
the problem of differing scales. We employ optical (confocal and two photon
imaging) and MR imaging as examples of two complementary imaging modalities
with different applicable scales of temporal and spatial resolution. Optical
imaging techniques have high spatial and temporal resolution, but suffer
from limited field of view and the inability to see deep into dense tissue.
MR imaging is capable of wide fields of view and has the well recognized
ability to image deep within dense tissue in a non-invasive manner, but
it suffers from limited spatial and temporal resolution. "Hierarchical
Multi-Resolution Image Matching" is a joint experimental and computational
approach to melding optical and MR images of the same specimen into a single
package of information. The fact that image contrast in these two modalities
arises from qualitatively different mechanisms both complicates the analysis
and enriches the final product. Biocompatible agents that are visible in
both the optical and MR imaging modalities provide the experimental key
to joining images from the different techniques. These bifunctional tagging
agents provide high contrast targets for similar objects in both the optical
and the MR image data.
After automatically identifying similar features in the two image types,
the high resolution optical data is then "projected" down to the resolution
of the MRI data using our dyadic wavelet pyramid approach. The images,
which are now effectively at the same resolution, will be geometrically
matched using these common features as matching tie-points. Tie-points
for geometric matching will be automatically generated using a Fourier
phase correlation technique. This method is particularly suited to the
image registration problems encountered in these efforts because it is
relatively insensitive to image differences which are introduced due to
the different imaging modalities used. We are developing this approach
specifically for optical and MR imaging, but it is generally applicable
to other imaging methods.
|
The neuroscience aspects of this project address issues of how the brain
acquires its exquisitely complex structure - both in terms of anatomy and
neuronal patterns of connectivity. Experiments involve obtaining images
of the intact developing brain, in a number of species, repeatedly over
time, with different imaging modalities. As mentioned above, laser scanning
confocal microscopy, two-photon microscopy, and magnetic resonance imaging
microscopy will all be used in these studies. The different imaging modalities
emphasize different aspects of the developing brain. The high temporal
and spatial resolution of the optical techniques makes them ideal for real-time
imaging to follow the patterning of axons and cell bodies in the developing
nervous system. The ability of MR imaging to 'see' inside dense tissue
makes it ideal for following the elaboration of anatomical structure in
the developing nervous system. These efforts are a combination of tool
building (both hardware and software), data collection and analysis, sharing
of the voluminous data with the neuroscience community, and hypothesis
testing using the information obtained.
|
Past progress in this Human Brain Project has been fueled by ongoing and
dynamic interactions across the various individual Projects and Cores.
This is aided by the fact that many of the personnel have expertise and
interests that cross the artificial Project/Core boundaries. Work in the
informatics components provides the neuroscience components with novel
and more efficient ways of collecting, analyzing, and looking at information.
This necessitates the computational components learning from the neuroscience
components about how the data is collected, as well as what it is about
the data that is interesting and important. Thus, we have a win-win situation
where each component aids and is in return enriched by the other component.
The PI's have ongoing and complex collaborative relationships - e.g.
Fraser and Jacobs work on MRI micro-imaging of frog development; Barr and
Fraser collaborate on imaging and modeling biological systems; Laidlaw
and Jacobs collaborate on MRI diffusion tensor imaging; Allman and
Barr collaborate on MR imaging and extracting information from measurements
on the human brain; Solomon and Fraser collaborate on automated cell tracking
algorithms; Jacobs and Allman collaborate on in vivo MR functional
imaging in rodents and small primates; Barr and Jacobs collaborate on multidimensional
rendering and model acquisition in MRI. There are also many three-way collaborations
that are just as prevalent as the two-way efforts (e.g. Meade, Fraser,
and Jacobs work together on in vivo MRI contrast agent development).
The three individual Projects and four Cores revolve around the development,
application, interpretation, and presentation of high resolution multidimensional
images of the developing nervous system obtained on several model systems
both in vitro and in vivo. The collaborators bring a number of different
philosophies and types of expertise: computer graphics, modeling and control
theory; mapping brain physiology/function and cortical learning; embryology,
cell and molecular biology, and digital optical microscopy; microscopic
resolution MRI. While these appear to be disparate interests, the inherent
interrelationship of the research goals of the individuals makes the 'umbrella'
of a single research project an ideal way to coordinate and optimize our efforts aimed at better
understanding the brain.
Chemistry Core
Tom Meade
This core supports the design, synthesis, purification, and characterization
of novel contrast agents. These agents are
typically composed of two or more linked parts:
MRI 'active' part composed of a ligand which binds
Gd3+ (e.g. DTPA or DOTA);
optically 'active' part composed of a fluorescent
moiety (e.g. fluorescein or rhodamine);
targeting part whose properties are tailored to
the problem at hand.
For example, in neuronal tracing studies the targeting part is a lipid
that is incorporated into the membrane of the growing neuron, while in
cell lineage studies a high molecular weight dextran moiety is used because
it is inert and membrane impermeant. In other cases it is desirable to
have the separate parts of the agent interact with each other and to design
this interaction to be dependent upon the local environment. In these situations,
MRI contrast agent which are normally "silent" will be designed to "speak"
in the presence of a metabolic or physiological function (e.g. high Ca2+
levels or the gene marker enzyme b-galactosidase), thus indicating the
whereabouts of the function.
Animal Core
John Allman
All items pertaining to animal acquisition, care, breeding, surgery,
and disposal are dealt with through this Core. An experienced Animal
Technologist oversees all aspects of this project that involve live
animals. This includes being an integral part of the design and construction
of all MRI hardware planned for in vivo animal imaging. The services of
this core leave the investigators free to conduct their experiments,
while ensuring that the animals are cared for in a humane manner consistent
with Caltech and NIH guidelines.
Computation, Web, and Networking Core
Al Barr
This Core provides a stable and effective computational environment
for the other Projects and Cores. It aids in implementing algorithms
associated with goal-directed imaging experiments and dissemination of
results; provide routine hardware and software support for computers, networks,
Web page maintenance, and video environments. On-line digital archives
of images, models, and data from the other Cores and Projects will be maintained
so that all researchers in this Human Brain Project have access.
Micro-MRI Core
Russell Jacobs
This Core performs routine maintenance on the current MRI hardware/software
(Bruker AMX500 11.7 Tesla 89mm bore NMR spectrometer) and aids in implementing
new hardware and software (e.g. implementing fMRI routines, the experimental
portion of the "teleological pipeline") in support of individual projects.
New developments are aimed at optimizing image contrast and increasing
spatial and temporal resolution. A significant amount of effort is
directed towards MR probe development in order to maximize signal/noise,
deliver stimuli synchronized with the MR experiment, and insure animal
safety.
|