Special Lectures:


A Unified Approach for Identifying Behavior-Related Neural Processing
Alterations and Functional Connections in the Human Brain: A Spatial
Modeling Approach for fMRI Data.

F. DuBois Bowman
Department of Biostatistics and Bioinformatics
Center for Biomedical Imaging Statistics
The Rollins School of Public Health
Emory University

Functional magnetic resonance imaging (fMRI) studies attempt to
identify spatially localized brain regions that drive the execution of
experimental tasks, for example targeting behavior, cognition, or
emotion.  These studies may also provide neural representations of the
associations between task-related brain activity in spatially distinct
regions.  These objectives have previously been addressed by separate
analytic techniques.  There are numerous challenges in analyzing
functional neuroimaging data.  Neuroimaging studies produce massive
data sets comprised of serial scans on each subject, with each scan
containing hundreds of thousands of spatially localized measurement
sites (voxels).  Also, human brain function, viewed through fMRI,
presents complex patterns of spatial and temporal correlations.  In
this talk, we consider fMRI data from a study of inhibitory control
among cocaine addicts.  We present a Bayesian hierarchical model that
establishes a unified framework for addressing inferences concerning
task-related 1) changes in brain activity, both at the voxel-level and
regional level and 2) functional connections, both within and between
designated neuroanatomic structures.  We apply our model to fMRI data
from the inhibitory control study to assess localized (voxel-level)
changes in brain activity associated with cocaine addiction and its
response to treatment, regional changes in brain activity, and
measures of task-related functional connectivity.

______________________________________________________________________
_______________________________________________________________________

Making and Breaking Habits: The Basal Ganglia in Action

Dr. Ann M. Graybiel

The same brain that can construct language, music and mathematics also
lets us develop habits of thought and action. These semi-automatic
routines free us to think and attend to the world. But the habit
system can also be hijacked by disease and drug exposure. This lecture
will focus on the habit system of the brain and our remarkable ability
to switch from conscious activity to nearly non-conscious
behavior. The lecture will highlight research directed towards
understanding how we make and break habits and how the neurobiology of
the habit system is helping to advance understanding of human problems
ranging from Parkinson†¢s disease to obsessive-compulsive spectrum
disorders and addiction. Clinical and experimental evidence suggests
that our ability to acquire habits depends on the basal ganglia, deep
forebrain structures that are interconnected with the frontal cortex
in a series of loop circuits. The development of recording techniques
for monitoring neural activity in awake behaving animals now makes it
feasible to investigate what forms of neuronal representation are
built up in the basal ganglia and cortico-basal ganglia loops as
habits are acquired. Recordings from the striatum, the largest input
side of the basal ganglia, suggest remarkable plasticity in the
response properties of striatal neurons as animals learn sequential
procedures and also as they undergo bouts of learning, extinction and
reacquisition training. Single unit activity changes systematically
through these bouts and so does ensemble activity. These results
suggest that there is a form of `neural exploration' followed by
`neural exploitation" in the basal ganglia as procedures are
learned. After learning, the ensemble activity tends to emphasize the
beginning and end of such procedures, as though setting up boundary
states in higher-order representations. Such action-boundary
representations can be found also in multi-electrode chronic
recordings from the prefrontal cortex and striatum of monkeys and
information about time may be embedded as an infrastructure in such
representations. These and other findings support the view that basal
ganglia-based circuits can build representations of sequential actions
that facilitate their release or inhibition. Evidence suggests that
the laying down of such representations involves genes expressed in
basal ganglia-related networks, and there are promising molecular
approaches to understanding learning mechanisms of the basal
ganglia. Disorders of such basal ganglia plasticity could contribute
to behavioral fixity and difficulty of initiation of behavior, as in
Parkinson's disease, or to the excessive release of behaviors, as in
Huntington's disease, or to the repetitive behaviors and thoughts
characteristic of many neuropsychiatric disorders. The basal ganglia
thus may influence not only motor pattern generators, but also
cognitive pattern generators. 

____________________________________________________________________
_____________________________________________________________________

Progress and Challenges of Multimodal Imaging using MEG, 
EEG, MRI, and fMRI

Matti S. Hamalainen
Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts
General Hospital, Charlestown, MA, USA

Independently, electromagnetic and hemodynamic measurements of brain
activity offer compromises between spatial and temporal
resolution. Functional Magnetic Resonance Imaging (fMRI) is temporally
limited by the slow time course of the hemodynamic response, but can
provide a spatial sampling on a millimeter scale. Electroencephalography 
(EEG) and magnetoencephalography (MEG) in turn provide a temporal 
resolution of milliseconds, but the localization of sources is more 
complicated because of the ill-posed electromagnetic inverse problem. 
Elucidating the spatial distribution and temporal orchestration of 
human brain regions is thus facilitated by combining information 
provided by both anatomical and functional MRI with EEG/MEG data.

It was recognized very early on the source estimation problem in MEG
benefits from combination with anatomical MRI data. Presently, this
information is used routinely to delineate boundaries between regions
of different electrical conductivities for forward field computations,
to restrict the locations and orientations of the sources, and in
advanced visualization techniques involving three-dimensional
renderings of the cortical mantle and other structures.

The most obvious companion of MEG is EEG. Presently, many laboratories
are collecting MEG and EEG data simultaneously and exploring source
models which incorporate both types of data. Both simulations and
analyses of actual human data have shown that the combination of these
two methods yields more reliable estimates of the sources than using
one modality alone. Furthermore, these studies indicate that the
improvement is attributable to the different spatial sensitivities of
MEG and EEG.

The fusion of electromagnetic and hemodynamic data is still in its
infancy. In current modeling methods, this is usually accomplished by
confining the sources to the cortical gray matter and by computing a
distributed current estimate with a stronger a priori weighting at
locations with significant fMRI activity. More elaborate methods which
attempt to model the two data sets jointly under a common framework
are also emerging.

Email: msh@nmr.mgh.harvard.edu

_______________________________________________________________
_______________________________________________________________

Ruling out and ruling in neural codes

Sheila Nirenberg

The subject of neural coding has generated much heated debate.  A key
issue is whether the nervous system uses coarse or fine coding
strategies.  Each has different advantages and disadvantages and,
therefore, different implications for how the brain computes. For
example, the advantage to coarse coding is that it's robust to
fluctuations in spike arrival times.  Downstream neurons don't have to
keep track of the details of the spike trains.  The disadvantage,
though, is that individual cells can't carry much information, so
downstream neurons have to pool signals across cells and/or across
time to obtain enough information to represent the sensory world and
guide behavior. In contrast, the advantage to fine coding is that
individual cells can carry a great deal of information;
however,downstream neurons have to resolve spike train structure.
Here we address the question of what the neural code can and can't be,
using the retinal output cells as the model system. We recorded from
essentially all the retinal output cells an animal uses to solve a
task, evaluated the cells' spike trains for as long as the animal
evaluates them, and used optimal, i.e., Bayesian, decoding. This
approach makes it possible to obtain an upper bound on the performance
of codes and thus eliminate those that are not viable. Our results
show that coarse coding strategies are insufficient; finer, more
information-rich codes are necessary.

Email: shn2010@med.cornell.edu

_______________________________________________________________
_______________________________________________________________

Coding and Computation by Neural Ensembles in the Primate Retina

Liam Paninski, Columbia University

The neural coding problem --- deciding which stimuli will cause a
given neuron to spike, and with what probability --- is a fundamental
question in systems neuroscience. We apply statistical modeling
methods to analyze data recorded from a complete mosaic of macaque
parasol retinal ganglion cells in a small region of visual space. We
find that a surprisingly simple model with functional coupling between
neurons captures both the stimulus dependence and the detailed
spatiotemporal correlation structure of multi-neuronal responses; in
addition, ongoing network activity in the retina accounts for a
significant portion of the trial-to-trial variability in a neuron's
response. We assess the significance of correlated spiking by
performing optimal Bayesian decoding of the population spike
responses; we find that approximately 20% more stimulus-related
information is captured when correlations are taken into
account. Finally, we discuss work in progress on the following
questions: how much temporal precision is necessary to capture the
neural code in the retina? How can we adapt our optimal decoding
methods to estimate behaviorally relevant signals such as image
velocity? How do we perceive stable images when the retina must
contend with the constant motion due to small random eye movements?


Email: liam@stat.columbia.edu

_______________________________________________________________
_______________________________________________________________

Neuronal Oscillations as Instruments of Sensory Selection.

Charles E. Schroeder, Ph.D.
Cognitive Neuroscience and Schizophrenia Program
Nathan S. Kline Institute for Psychiatric Research
Professor of Psychiatry
Columbia University College of Physicians and Surgeons

It has been known for over 75 years that, independent of frequency,
neuroelectric oscillations reflect rhythmic shifting of neuronal
ensembles between high and low excitability states. Recent findings
indicate that neuronal oscillations have a highly structured
hierarchical organization across frequency bands. These findings also
strongly reinforce the conclusion that neuronal oscillations both
enable and constrain the brain's processing of sensory inputs, as
well as its generation of motor outputs. I will discuss recent
findings concerning the way the brain uses oscillations to process
input, and the methods used in these investigations. I will also
discuss a conceptual framework that can help to integrate the evidence
on the neuronal oscillation as a mechanism of brain operation, with
prior findings in traditional vision research paradigms (e.g.,
vigilance). The discussion will range across several sensory
modalities, and will end with a focus on selective attention's
manipulation of oscillatory phase as a mechanism of sensory selection
in primate V1.

Email:  schrod@nki.rfmh.org; cs2388@columbia.edu 

_______________________________________________________________
_______________________________________________________________

Challenges in the analysis and acquisition of 3-D in vivo imaging data

Dr. Philbert  Tsai
Kleinfeld Laboratory - UCSD

The analysis of functional and structural brain imaging data poses
nontrivial challenges and opportunities for the statistics and applied
mathematics communities.  Research in our laboratory focuses on
imaging vascular and neurovascular dynamics in cortex, using the
rodent as a model system and scanning nonlinear optical microscopy as
a tool.  I will discuss specific issues that arise in the analysis of
blood flow dynamics as well as issues that relate to large-scale
vectorization of vascular anatomy.  Lastly, I will present general
challenges for integration of on-line analysis and machine learning
with data acquisition and experimental control.
 
Email: ptsai@physics.ucsd.edu

_______________________________________________________________
_______________________________________________________________

Waveform based Spike Sorting yields Inconsistent Rate Codes Estimates

Valerie Ventura
Carnegie Mellon University

Extracellular electrodes are commonly used to monitor populations of
neurons.  The signal collected at an electrode is a mixture of
activities from different neurons.  Spike sorting consists of
clustering the recorded spike waveforms, with aim to determine how
many neurons contributed to the recorded data, and which neurons
produced which spikes.  The spike trains thus obtained are often used
to estimate how neural inputs modulate neurons firing rates.  This
sequence of first spike sorting, then estimating tuning properties,
has long been the accepted way to proceed.  Here, we prove that tuning
function estimates thus obtained are systematically biased, unless
spikes can be classified with perfect accuracy.  This means, for
example, that the commonly used peristimulus time histogram is a
biased estimate of the firing rate of a neuron that is not perfectly
isolated.  We further illustrate that inconsistent tuning curve
estimates are more than a statistical inconvenience.  Indeed they can
lead to erroneous scientific conclusions or loss of efficiency.  In
particular, substantial efficiency can be lost when decoding neural
spike trains to operate neural prostheses.

We further propose a new automatic spike sorter, which combines
waveform and tuning information. This spike sorter effectively
performs spike sorting and tuning function estimation simultaneously
rather than sequentially, as is currently done.  We prove that this
spike sorter yields consistent tuning curve estimates, so that
subsequent statistical procedures are fully efficient and scientific
inferences valid.