fmri methods lecture 12 – adaptation & classification
TRANSCRIPT
Integration of information
Retinal ganglion cell receptive fields
V1 neuron receptive field(Hubel & Wiesel)
Integrate
VisionComponents of visual computation:
Locations in visual field
Contrast
Orientation
Spatial frequency
Direction of motion
Categories – objects, faces, houses
Neural selectivity!
Test selectivity
Change a stimulus attribute in a controlled manner and see how the neural response changes...
Neural tuning curves
Particular neurons prefer specific stimulus attributes:
“right” neuron “left” neuron
0° 45° 90° 135° 180° 225° 270°
Similar tuning within a column
Organization of orientation selective neurons in primary visual cortex.
200-300 µm wide
Cortex is 2-4 mm thick
Each voxel = 3 mm3
40-50,000 neurons per mm3
~ 1,000,000 per voxel
Resolution
Fine for distinguishing between gross neural systems in separate brain areas (“modality selectivity”).
How can we tell what intermingled neural populations within a particular area are selective for?
Location of adaptation
Different forms of adaptation take place at different synaptic and cellular locations and have different time-scales…
There are multiple sensory neurons on the siphon, if adaptation is pre-synaptic will it generalize across locations?
Location of adaptation
Many possibilities:
Amount of neurotransmitter released.
Amount of receptors available (receptor traffic) at the post synaptic cleft.
Activity dependant ion channel changes (conductance): either pre-synaptic or post-synaptic.
Protein dependant structural synaptic changes (longer term).
Membrane conductance change
Contrast adaptation in V1 neurons – 100’s of milliseconds
Carandini et. al. Science 1997
Tester strength (% contrast)
1.5% adapt
47% adapt
1.5% adapt
47% adapt
Models of fMRI adaptation
Grill-Spector et. al. TICS 2006
Adaptation carry over to higher cortical areas?
Models of fMRI adaptation
Grill-Spector et. al. 2001
Different number of stimuli types within a block of 32 stimuli
Selectivity, adaptation, & invariance
Grill-Spector et. al. 2001
A true test of “high level” selectivity would be invariance across “low level” changes…
Selectivity, adaptation, & invariance
Every neuron from retina up will adapt…
None will adapt…
Generic “face” neurons will adapt…
Only Individual “face” neurons will adapt…
fMRI adaptation study
Yellow: overlap of motor and visual adaptation
Mirror system areas
Dinstein et. al. 2007
Repeat Non-repeat
How consistent are the patterns?
Trial #1
….
….
….
Trial #2
Trial #3
Trial #4
Trial #1
Trial #2
Trial #3
Trial #4
Trial #1
Trial #2
Trial #3
Trial #4
Response per trial…Y (bold) = X (model) * b (response amplitudes) + error
Y
=
Tim
e po
ints
X (column for each trial)
010000000000
000010000000
000000000100
100000000000
000100000000
000000010000
001000000000
000000001000
000000000001
… … …
b for each trial
*
b1
b2
b3
b4
000....
000b1
b2
b3
b4
00..
0000000b1
b2
b3
b4
Y
=
Tim
e po
ints
X (column for each trial)
… … …
b for each trial
*
b1
b2
b3
b4
000....
000b1
b2
b3
b4
00..
0000000b1
b2
b3
b4
Response per trial…Y (bold) = X (model) * b (response amplitudes) + error
Multivariate pattern classificationResponses of a single voxel to different direction trials
Resp
onse
Am
plitu
de
Multivariate pattern classificationResponses of two voxels to different direction trials
Res
pons
e A
mpl
itude
(vo
xel 1
)
Response Amplitude (voxel 2)
Left
Up
Right
Multivariate pattern classificationR
espo
nse
Am
plitu
de (
voxe
l 1)
Response Amplitude (voxel 2)
Left
Up
Right
Responses of two voxels to different direction trials
Multivariate pattern classificationDecode direction using brain pattern
Res
pons
e A
mpl
itude
(vo
xel 1
)
Response Amplitude (voxel 2)
Left
Up
Right
Left?
Multivariate pattern classificationDecode direction using brain pattern
Res
pons
e A
mpl
itude
(vo
xel 1
)
Response Amplitude (voxel 2)
Left
Up
Right
Decode trials (leave one out)
Trial #1
….
Trial #2
Trial #3
Trial #4
Trial #1
….
Trial #2
Trial #3
Trial #4
Trial #1
….
Trial #2
Trial #3
Trial #4
Decoding attended stimulus
The orientation that the subject attends is evident in the distributed response pattern of V1 neurons!