Transcript 14.FARS 3.Synthetic PET(2001) - University of Southern California

Michael Arbib: CS564 - Brain Theory and Artificial Intelligence
University of Southern California, Fall 2001
Lecture 14.
FARS and Synthetic PET
Reading Assignment:
Arbib, M.A., Billard, A., Iacoboni, M., and Oztop, E., 2000,
Synthetic Brain Imaging: Grasping, Mirror Neurons and
Imitation, Neural Networks, 13: 975-997.
In addition to the material on FARS and Synthetic PET covered in
class, the paper contains material on imitation - fMRI data, a
model by Aude Billard and a Synthetic fMRI study, as well as
some notes on the MNS model.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
Synthetic PET: Analyzing Large-Scale Properties of
Neural Networks
Arbib, M.A., Bischoff, A., Fagg, A. H., and Grafton, S. T.,
1994, Synthetic PET: Analyzing Large-Scale Properties
of Neural Networks, Human Brain Mapping, 2:225-233.
The issue here is to how to map
simulated activity of the neurons in models of interacting brain regions
based on, say, single-cell recordings in behaving monkeys

into
predictions of activity values to be recorded from corresponding regions of
the human brain by imaging techniques such as positron emission
tomography (PET).

Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
Modeling activation
PET typically measures local cerebral blood flow (CBF).
The key hypothesis of our method is that
the counts acquired in PET scans are correlated with the synaptic activity
within this region.

However, PET studies typically do not work directly with
raw PET activity but rather with the
comparative values of this activity in a given region for two different tasks
or behaviors.

to estimate task specific modulations of local activity.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
Localization
Each array in the neural network model represents a
neural population in a region identified anatomically and
physiologically in the monkey brain.
A synthetic PET comparison requires explicit hypotheses stating that
each such region A is homologous to a region h(A) in the human brain.
Comparison of a synthetic PET study with the results of a human brain
scan study will, inter alia, be a test of the hypothesis

"h(A) in human is homologous to A in (a given species of) monkey".
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
Computing raw PET activity
t1
rPET A   wBA (t) dt
t0 B
Assumptions:
H1. All synaptic contacts are made within the region in which the cell
body is located.
H2. The contribution to the blood-flow measured by PET in both
inhibitory and excitatory synapses is defined by the integral of the
absolute value of the synaptic weight times the spike rate incident
upon the synapse.
H3. We can lump all cells within a neural region into a single sum for an
individual region. duration of the scan.
Note that our methodology will put H2 to specific test.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
Creation of the synthetic PET comparison
PET A 1 / 2 
rPET A 1  rPETA 2
rPET A 2
where rPETA(i) is the value of rPETA in condition i.
We may convert the value of the "synthetic PET comparison" PETA(1/2)
to a color scale, and display the colors on the region h(A) homologous
to A on slices based on the Talairach Atlas.
As a computational plus (going beyond the imaging technology), we
can also collect the contributions of the excitatory and inhibitory
synapses separately, based on evaluating the integral in (1) over one set
of synapses or the other.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
Another View of FARS
The precision pinch and power grasp
pools in F5 and AIP are connected
through recurrent excitatory
connections. The precision pinch
pool contains more neurons than
other grasps, which effects the
Synthetic PET measure in these and
downstream regions. F6 (pre-SMA)
represents the high-level execution
of the sequence, phase transitions
dictated by the sequence are
managed by the basal ganglia (BG).
The dorsal premotor cortex (F2)
biases the selection of grasp to
execute as a function of the
presented instruction stimulus.
F6
instruction
stimulus
F2
AIP
F5
power
power
precision
precision
SII
BG
F1
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
Conditional Tasks and Area F2
In addition to simulating the Sakata task, we simulate
conditional tasks. Here the crucial observation is that:
Dorsal premotor cortex (F2) is thought to be responsible for the
association of arbitrary stimuli (an IS) with the preparation of
motor programs.
In a task in which a monkey must respond to the display of a pattern
with a particular movement of a joystick:
some F2 neurons respond to the sensory-specific qualities of the input.
However,
many F2 units respond in a way that is more related to the motor set
that must be prepared in response to the stimulus.
When a muscimol lesion in this region is induced, the monkey loses the
ability to correctly make the arbitrary association.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
Modeling Interactions of F2, F5, and AIP
F5
F2
AIP
instruction
stimulus
Basal Ganglia (BG) appears implicitly - mediating the recurrent
inhibitory connections back to F5 and AIP.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
The PET Grasping Experiments: Conditions
Control

No movement; only watch lights
Power Grasp

Green Light indicates which block to grasp
Precision Pinch

Green Light indicates which switch to pinch
Conditional vs. Non-Conditional Task
Grasp a cylinder using either a precision pinch or a power grasp, the
choice being determined by an instruction stimulus (the color of a
light).

Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
The PET Grasping Experiments: Apparatus
FSR
LED
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
Precision Pinch versus Power Grasp
Basic Model Assumptions Reflected
Here:
F5 has more cells to code precision than
pinch


SII has more cells to code expectation

MCx has to do more work for precision
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
PET Results : Precision versus Power
Precision versus power grasp (lower panel).
+50:
L - SMA Proper
R - Dorsal Premotor Cortex.
No show: F5  Ventral Premotor Cortex!!
+32: L - Inferior Parietal  AIP
- 16: L - Cerebellar Vermis
Also: Contralateral Occipital Lobe
Model predicts increases in AIP and F5. AIP and F5
are co-sensitive in the current model.
Data only see AIP. Why no F5?
 AF:
Stereotypical grasp or Force recruitment in
power grasp
 MA: Few cells code a specific pinch
Coding issues are crucial.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
PET Results: Precision versus Power
+50:
L - SMA Proper
R - Dorsal Premotor Cortex.
No show: F5  Ventral Premotor Cortex!!
+32:
L - Inferior Parietal  AIP
- 16:
L - Cerebellar Vermis
Also: Contralateral Occipital Lobe
Model predicts increases in AIP and F5. AIP and F5 are co-sensitive in
the current model.
Data only see AIP. Why no F5?
 AF:

Stereotypical grasp or Force recruitment in power grasp
MA: Few cells code a specific pinch
Coding issues are crucial.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
Conditional Task versus (Precision, Power) Average
Basic Model Assumption Reflected
Here:
F2 processes the instruction stimuli
only in the conditional task

Second order effect on F5 due to F2
input: this activity level is passed to AIP
and BG

Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
PET Results: Conditional versus Non-Conditional
Conditional versus Non-Conditional (upper right panel).
R - Area 18 (2 sites)
L - Area 18/19
R - Cerebellar cortex
F2 and AIP are activated; F5 is not.

F2: Data and model agree
 AIP:
Human has more activation
than model
F5: Human has no change but
model does increase slightly.

Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
Model and Data: Toward Reconciliation
In current model,
F2  F5 is a free
parameter.
F5
AIP
F2  F5  AIP forces
AIP to change less than
F5.
F2
F5
BG
ins truction
stimulu s
AIP
AF: A possible
solution - reroute F2
 F5 to
F2  AIP
F2
instruction
stimulus
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
Should we Accept this Perspective on F5-AIP Interactions?
tell PM the
possible
categories
affordances for PM
precise motor coordinates
for motor cortex
“here’s my
choice”
PM (Premotor
Cortex)
codes action fairly
abstractly
PP (Posterior
Parietal)
overall
action
the details action parameters
Motor
Cortex
Cerebellum
Tuning and
coordinating MPGs
and MPGs
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
Synthetic PET: A Tender Seedling
Synthetic PET forces attention to details of human data
while highlighting assumptions made in monkey
models.
 Assumptions
must be made to bridge from a limited data set to explanations of
coherent functioning at cellular and behavioral levels.
Synthetic PET both benefits from and contributes to better understanding of
homologies between human and monkey.
We need further research on metabolic correlates of neural information
processing:
synaptic activity
axonal activity
synaptic plasticity
gene expression, glial function, etc., etc.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET
And the Method is Much More General
And the Method is Much More General

Homologies with non-primate species
Extension to fMRI and other imaging techniques - with further research on
metabolic correlates of neural information processing.

The Grand Aim:
Increased Progress in Systems Neuroscience by Developing Modeling
Tools that Catalyze Progress through the Integrated Use of Human and
Animal Data

Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 14. FARS and Synthetic PET