Object Processing
Visual agnosia is marked by elementary visual capacity (acuity, visual field)
intactness, but severe perceptual deficits.
VA can be broken into apperceptive (early stage: elementary shape perception) and
associative (shape intact, but can’t associate percept with meaning / name /
pantomime use. must recognize by sound, touch, taste, questions).
Apperceptive patients can’t tell objects apart or shapes. Can’t draw contours. can’t
copy shapes. Also cannot disentangle contours for overlapping objects.
Associative are still good at copying drawings, but not good at drawing from
emmory. printing flaws of the same object cause these patients to think they are
different objects. Totally summed up by people not knowing what is on their plate
until they taste it.
The Kluver-Bucy syndrome is associated with hyper-orality where animals have a
visual agnosia and need to interpret objects with mouth.
Lesions to Inferotemporal cortex causes an inability to distinguish different objects,
a lack of retinament in previously acquired visual discriminations.
IT usually spans across viewing conditions (view-independent-invariance), so if you
lose a concept of an object, you lose all viewpoints of it. This is also true for changes
in illumination.
IT lesioned monkeys can’t transfer visual discirimination from half visual space to
the other.
Dorsal stream for “where”. ventral stream for “what”. IT lesions affect
discrimination but not visuo-spatial tasks like reaching or visospatial judement.
Parietal lesions do the opposite. Ventral stream responds to features. Dorsal
respond to spatial aspects like direction and speed.
As we move from V1 to higher order regions, the receptive fields get larger, like in
TE. There is no retinotopy in TE.
When subjects attend to particular aspects of visual stimuli, PET scans show
preferential increases in blood flow suring this selective attention (dorsal for
motion) ventral for shape, color. This is evidence of top-down modulation. There is
more evidence where TMS over V5 induces moving phosphenes and over V1 creates
static.
Neurons in macaque area V4 aquire directional tuning after adaptation to motion
stimuli. V4-V5 interactions represent modulation between areas at similar level of
visual hierarchy. They also represent cases of cross-talk between ventral and dorsal
streams. V4 neirions are not normally motion selective but can become so due to an
adaptation in V5 neurions that changes V4 neuronal properties.
Connections of these streams with hippocampus (via EC) and amygdala make things
more valent and memorable.
As you move further along the ventral stream, there is an increase in invariance.
There is also a grandmother cell hypothesis where every object is coded in a single
cell in IT, but the population code hypothesis seems more fit where each specific
object activates patterns in multiple cells.
Sparse coding shows that a few cells might code for very selective stimuli (Jennifer
Anniston).
Cells in TE that respond to common visual features are grouped in columns running
perpendicular to cortical surface. Columnar organization is reminiscent of V1.
Temporal neurons show selective responses to biological stimuli (faces – FFA).
some respond to gaze direction, hand-object interactions, biological motion
(walking), and point light displays (lights at joints like making CGI).
Viewpoint dependence in object recognition: object centered is that recognition is
by components and objects can be defined by a small set of primitive shapes
(geons). viewer-centered has a key factor where its objects need to be seen as they
originally were presented. A small set of prototypical views would suffice due to
mental rotation.
Object recognition and experience can be seen under repetition suppression and
mnemonic enhancement. More neurons will be recruited for a trained object.
repetition suppression can happen under anesthesia and unattended stimuli.
Cutting the line between TE and hippocampus reduces pair coding.
TE neurons extract invariant aspects of objects, preferentially respond to
biologically relevant stimuli, and they learn (increased frequency of selective cells
after learning and pair coding.)
Dorsal and ventral double dissociations extend to imagery as well.
Lateral occipital complex is concerned with object processing (actual vs. scrambled
objects shows preferred activation to non-scrambled) . Object sizes don’t matter. All
that matters is object like. LOC is intermediate between V1-5 and TE/IT.
There is also an extra-striate body area that responds to body parts, PPA, STS-FA is
also responsive to the changeable aspect of faces (smiling). FFA more lateral than
PPA.
N200 wave forms show shape selectivity that is strongly induced by faces but not by
changes in low level visual features.
Prosopagnosia – can’t recognize faces. can classify objects or faces from other
objects, but not between faces. Could be due to lack of expertise.
Haxby studied showed that we can classify faces even when the FFA is removed,
which is a great case for distributed coding.
Representational Dissimilarity Matrices that show animate and inanimate
categories in IT as having the most dissimilarity in humans and monkeys alike. Also
there is an unsupervised multidimensional scaling approach that groups neuronal
responses to objects into bodies, faces, natural, and artificial objects. This
dissociation does not extend to V1 and the dissimilarity remains even if highest
BOLD responses are removed….more evidence for distributed coding.
pictures of tools will activate not only visual areas, but also motor areas necessary
for grasping in parietal and premotor. About 50% of grasping motor neurons in
ventral premotor cortex also have visual properties. They fire at the site of
graspable objects. They are called grasping visuo-motor neurons or canonical
neurons.
AIP (anetior intraparietal ariea) in the intraparietal sulcus shows motor and visual
selectivitiy in a mirror neuron fashion.
Space Processing
PPC known for integrating perception and action. Broken up into SPL and IPL.
SPL lesions cause optic ataxia. IPL causes hemispatial neglect.
Simultagnosia is the inability to attend to multiple objects simultaneously.
Ocular apraxia is the inability to move eyes to new target locations.
Optic ataxia is basically the inability to use visual information to guide reaching
movements.
In hemispatial neglect (since its usually seen with lesions to the right IPL) the
patients neglect the left side and “squeeze” all parts of a scene into the right side or
ignore the left all together. Bisecting a line shows a serious rightward bias. In a
whole page of lines, with instructions to cross the lines, they will only cross out one
half.
Its thought that right hemisphere attends to both hemispheres, wheras the left
attends only to contralateral space. Imaging supports this.
Hemispatial neglect can be object-centered where they can attend to both sides of
space, but the frame of reference is the object. They will draw only the right side of
each object if two are presented in both visual fields.
If a monkey makes an identical saccade, but it is either to the right or left of an
object, then neuronal firings will be diminished when its to the right of the bar.
Particularly in the supplementary eye field.
in real life scenarios, this is witnessed as well (Duomo study). If patients stand and
describe a scene, they will neglect the left. If they are asked to imagine standing the
opposite way, they will report details about things they omitted before. This is
evident of implicit processing without conscious awareness.
Anosognosia neglect part of their body and can’t map their condition to their own
body.
with body schema problems, patients will deny that their contralateral limb is their
own. object sin hands are reported to be held by someone else until moved to
ipsilateral hand.
Patients shown red or green lights to L or R visual field and asked to name color.
They respond to right visual field, but not left. however, if response keys are on the
left side, patients can’t even detect the stimuli.
In the line cancellation task, if you provide a mirror so that patients can see both
sides of the paper, neglect patients cancel the left side of the page (perceptual
neglect). However, some patients, with motor neglect, will cancel the right side of
the page.
There is near and far with spatial neglect. Some patients can’t bisect lines in front of
them but can use a laser to bisect a line far away. Vice versa exist as well. This makes
a case for extrapersonal and peripersonal space.
Extrapersonal might be related to FEF lesions, since that region is associated with
focusing the eyes outside of peripersonal space. Ventral premotor lesions might lead
to peripersonal neglect because these concern arm and mouth movements. Thus LIP
and FEF are for extrapersonal and VIP and F4 are for peripersonal.
Thus, space maps are supported by fronto-parietal networks and damage to the SLF
(connects posterior parietal and frontal love) is related to patients with neglect.
Arcuate fasciculus is also apparently important (connects Superior Temporal cortex
with frontal lobe).
LIP neurons show strong responses at the onset of a visual stimulus in the RF. this is
modulated by attention. When still attending (working memory) to that RF, there is
activity. There is a dynamic remapping of visual memory trace
VIP neurons have RFs that are smaller around the mouth.
There are bimodal cells in ventral premotor area F4 that respond both to tactile
information and visual info of things approaching that sector of space around the
tactile receptive field. This moves with the body part, not the retina. The objects that
come close are typically 3D, graspable objects.
Extinction is when there is neglect only if there is a bilateral presentation of stimuli.
Cross-modal neglect is when a visual stimulus near the ipsilesional hand induces
tactile extinction of the contralesional hand. If the view of the hand is occluded, the
effect is reduced. The same effect us observed with a rubber hand.
Peripersonal space can be extended with the use of tools, like a rake. The visual
receptive field around the hand of the monkey will extend to the tip of the rake. You
can see inhumans with near space hemineglect that their near space extends if they
use a stick to bisect a line far away. But…if they use a laser pointer they can do it just
fine.
Stimulation of motor cortex can induce final posture regardless of initial. However,
the EMG profile is different depending on the start state. That means we are
stimulating intent and the intent gets completed regardless of starting position.
The velocity of these movements are the same as naturally spontaneous
movements.
You can induce defensive movements as well with VIP and F4 stimulation. Monkeys
will move in a manner that tries to protect the receptive field. Don’t just move the
body part with the RF, but also other parts to help block/avoid.
These movements can be evoked even under anesthesia.
There is a “hand-final-location” mapping where more dorsal representation for final
hand location is located more ventrally and vice versa. Models that try and figure
out the topographic continuity of actions ends up showing that the body motor map
in cortex is configured as such due to things necessary for coordinated movements.
hippocampal place cells are for particular locations in space and persist in darkness
and remap for new environments.
post subicular neurons are sensitive to head direction, even in dark, as long as
animal knows spatial orientation.
grid cells fire in regular hexagonal patterns across an environment. They can be
anchored to landmarks.
Attention
Covert orientation is attending to something without moving the eyes.
Slience drive exogenously driven attention.
Attending to a feature of an object tends to facilitate processing of other features of
the same objects.
Posner paradigm: spatial locations are cued with a peripheral cue or central cue(at
fixation with an arrow). Valid trials are when the actual stimulus goes to clued. RT
are faster for valid trials. The validity effect. RT gets penalized linearly with distance
of actual stimulus from cued location. This is the cost associated with switching the
direction of the covertly planned saccade.
The premotor theory of attention is that enhancement achieved by planning
(without executing) a saccade. This was tested by showing that attention shift a
saccade’s trajectory: saccades have trajectories that are not straight, with an initial
trajectory in the opposite direction of the cued location.
Both covert and overt attention recruit eye fields necessary for saccades (FEF, SFS,,
LIP, IPS, etc.). BOLD maps are overlapping.
However, peripheral cues can slow RT if the delay between cue and stimulus is long
enough(SOA: stimulus onset asynchrony). Known as Inhibition of Return, its
thought that this improves visual search by preventing return to previously
searched locations. Gaze cues don’t produce this IOR. IOR is thought to involve
cerebellum and superior colliculus since they involve peripheral cues (the only
situation where you observe this effect). These are olderstructures (and since IOR is
an old philosophy) it might follow that these regions evolved with the visual search
efficiency algorithms.
Gaze cues are important as well (looking at someone orient their eyes in a particular
direction). They are very powerful. They orient attention in an automatic manner.
Even when the gaze is not predictive of the target. Even when they are counter
predictive they will get faster RT to the gazed location.
FEF, SEF, and LIP(sometimes called PEF) are activated by both gaze and peripheral
cues.
Extrastriate areas are more active for gaze cues as compared to peripheral cues.
LIP neurons respond when stimulus is flashed onto its RF. However, does not
respond when a stimulus stable in the scene appears in the RF following a saccade.
Firing is high until right after saccade is executed. It does respond if the stimulus is
made salient by flashing it on and off, though. This suggests that LIP neurons
integrate bottom up and top down factors that modulate attention.
LIP neurons attend to task relevant or conscipuous objects by coding a salience
representation/priority map that specifies only a small number of stimuli.
Inactivating LIP or FEF produces deficits in target selection during search. Reward
expectations modulate this effect. However, they don’t change selectivity, but
instead more reliable response selectivity.
LIP neuronal responses are modulated by visual, cognitive, motivational, and motor
factors. The motor demands necessary for a task that involves the detection of
oriented stimuli in an LIP’s RF will be modulated by which motor response needs to
be made (left or right hand).
Attention modulated visual acuity changes activity in V4 before stimulus
presentation and during stimulation. Does attention modulate visual processing by
amplifying neural signals for attended stimuli? V1 doesn’t show any modulation.
Attention effects aremore pronounced in higher level processing areas. Top down
signals through feedback connections. Attentional effects have a 150ms delay in TE
and 230 in V1.
Attention modulates ERP by enhancing waveform in the opposite hemisphere.
Attention reduces signal ambiguities (separates the tuning curves of responses). Can
increase ability to discriminate orientations if attending to the RF where stimuli are.
Thus, attention increases discriminability by reducing variability of responses
(reduces the width of the response curve). This effect is stronger in interneurons
than in pyramidal cells. Perhaps this means that the effect of attention it so suppress
signal from distractors? Attention also reduces the degree to which ow-frequency
fluctuations in the neuron’s spiking were correlated with fluctuations in the activity
of other neurons. Spatial attention decorrelates intrinsic activity fluctuations in V4.
Decoupling intrinsic activity might help us dissociate things.
Furthermore, a decrease in correlation increases signal to noise ratio far more than
an increase in firing rate, as the neuronal pool size increases past 10.
If you stimulate FEF and record from V4 you can increase neuronal response in V1
to a receptive field that would be active if the saccade was made that would have
been made if the FEF stimulation was above threshold. This shows that preparing to
make a saccade to a location is similar to covert attention.
efficient search is pop out. ineddicient requires conjuntions of elementary elements
of which there are multiple of each type. each item in inefficient search adds
50msec.
Feature integration theory (FIT) is where elementary features are bound into
coherent objects. Pop out doesn’t require this. In inefficient, we need to integrate.
FIT assumes that object completion happens after attentional search. However,
completions that hide popouts (where occlusion could mistake someone for a
target) shows that object completion precedes attentional search. The
computational limits of search are after features are integrated into wholes. This is
probably because the RF of higher order visual areas are limited. Neurons cannot
simultaneously send signals about all the stimuli inside the RF.
Firing rate of infero-temporal neurons is increased more for preferred stimuli than
non-preferred.
If a stimulus is preferred and presented non competitively (isolated) or
simultaneously, there is no difference in V1, but there is in V4 where there is greater
activity than V1 (Bigger RFs, bigger effects).
Suppressive interactions among multiple stimuli are eliminated in extrastriate
cortex when they are presented in the context of pop-out displays, in which a single
item differs from the others, but not in hereogenous displays, in which all items
differ from eachother. This effect is exaggerated as you move up the visual stream.
Basically, there is a suppression when images are viewed in heterogeneous display
as opposed to in isolation. But, if the images have a pop-out, there is no relative
suppression.
Mirror Neurons
AIP-F5 are for grasping.
Mirror neuron system is generally PF, PFG-F5
Grasping neurons code for the goal, not specific movements. This changes whether
the intended grip is precision or whole hand grips.
There are excitatory mirror neurons where the cell fires when monkey grasps and
also watches experimenter grasp. There are also inhibitory where the cell stops
firing at the sight of the experimenter grasping and when the monkey grasps. The
firing rate, however, is different (perhaps preventing unwanted action). These cells
also fire when the monkey grasps in darkness (not strictly visual). These mirror
neurons still fire even if the intent of the grasping is clear, but not seen (due to a
screen put up to cover an object). The neurons won’t fire if the grasp was made in
vain (pantomime). Prior knowledge drives this.
Audiovisual mirror neurons will respond to the sounds of things that required
action/grasping. A common code for sender and receiver?
strictly congruent mirror neurons mirror the exact action. Broadly congruent mirror
exact actions and similar actions.
F5 neurons code for hand to mouth actions. STS actually codes a bit for vision of
observed actions (not motor). Could plausibly send visual info to F5. PFG connects
with both F5 and STS and could relay the signals of STS to F5.
Intention to eat are more powerful than just picking up and even picking up and
placing by the mouth.
Pliers study (reverse and normal grip) shows that n spite of completely different
movements, the cells fire in both situations suggesting that the coding is for an
action goal, rather than the finger movements.
Mirror neurons can be modulated by actions performed in peripersonal vs.
extrapersonal space. This is relevant for the potential interactions between observer
and observed action. However, this is more so for plausibility of interaction. If a
shield is put up where it removes the action from the monkey’s workspace (but still
would have been in peripersonal space), then you get responses similar to
extrapersonal space.
Piramidal tract mirror neurons can show excitatory reponses during action, but
inhibitory during observation which might be important for the inhibition of
movement during action observation.
We also see LIP mirror neurons for when a monkey observes another monkey
making a gaze in that neurons preferred direction.
There are also VIP mirror neurons for defensive movements.
Songbirds also show mirror neurons. identical patterns when listening to certain
note sequences as when producing them. Even disrupting auditory feedback during
singing does not alter the activity.
mirror neurons in humans have been found in SMA and MTL (for memory of making
the action when observing it)
humans usually have more prolonged responses (maybe a more complex neural
signature).
LIP mirrors eye and attention.
Dorsal PMC and M1 for reaching
VIP for peripersonal space
SMA movement initiation
MTL memory mechanisms
Watching grasping actions increases MEPs in muscles. This is also seen with
increases in contralateral premotor cortex. (just one synapse away from muscle).
Mirroring also suppresses central rhythms (beta and alpha) just like what happens
during execution.
fMRI shows overlapping activation for observed and executed. We also see
repetition suppression.
mirror neurons in ventral premotor/ posterior ifg (monkey F5) and inferior parietal
cortex (monkey PF) and superior temporal sulcus (for biological motion and action
observation).
Cross-modal repetition suppression is debated based on the model proposed earlier.
Mirror neurons do not adapt in response to observation of repeated actions.
Do we perceive speech by retrieving the motor plan necessary to emit the speech we
are listening to? Mirror neurons also appear to be important for social learning.
Empathy might interact with the core imitation network.
Mirror neurons might simulate the facial expression of seeing someone smiling,
without actually making the person smile…then pass info to insula then pass to
limbic to then feel the emotion….empathy.
activity of motor neurons in inferior frontal cortex, insula and amygdala correlates
with subject’s tendencies to empathize and their social competence.
Autism…defecit in mirroring? less mirror areas active during imitation of facial
emotional expressions.
Higher recover in stroke patients after action observation.
hearing speech and making speech activates same areas at the border of primary
motor and premotor. TMSing this region shows defecits in speech perception (even
though it is far from auditory/wernicke’s).
functional connectivity during speech perception show connectivity between
superior temporal and that same premotor/motor region.
Model: superior temporal implements acoustic analysis while premotor implements
the mirrored phoneme production necessary to repeat. This production gets
compared in superior temporal with the actual hearing. The error signal (if one)
would be sent back to premotor which would generate a corrected phoneme to be
used for categorization.