Ventral Visual Pathway 2012 Kravitz Ungerleider Mishkin

Nov 23, 2013 07:26 PM

Review in TICS  from  Lab of Brain and Cog  NIMH;  key paper;

1st author Dwight Kravitz

2nd author  Kadharbatcha  Saleen;

senior authors  Leslie Ungerleider and Mortimer Mishkin...

NIMH;

abstract:  it is a RECURRENT, Occipitotemp Network containing

neural reps of Object Quality both utilized and constrained by at least 6 distinct

cortical and subcortical systems.  Each system has its own special behav, cognit and affective function.

the expanded framework contrasts with (classic)  depiction of ventral vis pathway as largely serial

staged hierarchy culminating in singular object reps and this more parsimonioulsy incorporates

attention,  context and feedback effects;

we argue that ventral vision pathway is a recurrent,  highly interactive occiptemp network linking early visual areas and aIT (anterior inferotemporal)  along multiple routes.

includes at least 6 subcortical and cortical regions and support many difft forms of object quality proc.

bidriecitonal:

classic view: feedfwd orig  in layers 2,3 and terminate in layer 4,

whereas feedbk originate in layers 5,6 and terminate in layer 1.

Now a wealth of evidence =>  ventral vision is a complex recurrent network.

V1 goes not just to V2 but also to V3 direclty and to MT.

thur  visual info can go from V1 to TEO in 2 steps vis V2 or V4 or MT.

there are also  unidirec, non-recipr projections from Late stages to early stages.

Also feedfwd connex go across the board  (not just to layer 4)  terminate in all 6 layers.

incorp masking, attn and context.

page 27  3 aspects of our (current) modified framework

1)  ventral pathway actually  a complex network of feedfwd and feedback proections

2)  strong link btwn retinotopic and intrinsic network such that even

hi-level object rep is constrained by retina position.

3) at least 6 distinct subcortical and cortical strucs

1) unidirec occip-temp-  neostriatal pathway support SR assocs.

2) difft areas within IT support occip-temp to amygal pathway  and proc emotion salient stim.

3)  cortico-subcortic striatum path  originates in aIT and supports stim valence.

also 3 paths  cortico-cortical

1)  to med temporal:  support  LT memory

2) to orbitofrontal  support  object-reward assoc and

3) to VL PFC support object working memory.

page 29  Occip temp network: intrinsic connect.

Box 2:  functions of anterior IT cortex:

what visual functions, if any, are subserved by aIT in humans.

(persistent problem is lack of good fMRI signals from here;

but it IS implicated in semantic processing.

causes profound defects in monkeys but not in humans (perhaps cuz lesions are unilateral);

cites studies that clearly show that human aIT stores visual info and rich connex to MTL

(medial temporal lobe).

page 31:  strong evidence that lesions along the central route do NOT cause profound effects in later regions

as would be expected with a single hierarchical, serial system.

In monkeys, many bilat lesions and disconnections studies FAIL to silence aIT neurons =>

multiple parallel routes.

page 32:  strong support for an increase in complexity of reps as one goes anteriorly to aIT but not

of the necessity or suffic or a central route.

Even high level visual object reps are somewhat position-dependent:

contra vs ipsi,  upper vs lower and central vs peripheral.

Human neuroimaging =>  strong contralateral retinotopic bias that extends far into posterior IT!!

(and in monkeys even to aIT...

Object identiy is much more accurate in approp. contralateral hemifield in

EBA  extrastraite body area when in typical field  (R body in L field);

also  widespread eccentricity bias thruut  occipitotemp network:  thru early visual areas and even in

IT regions.  in monkeys  IT shows strong foveal bias.

In huamns the PPA  (parahippoc  place areas) containing scene selective info 

from the peripheral space of scenes;  whereas IT is aligned with foveally biased rgions for

faces, wrods, etc.;  also small objects are lateral and large objects medial;

retinotopy may be a genl organizing princip. of the ventral visual path.

CLUSTERING  body parts, faces, scenes, objects, tools, written workds, color.

clusters could emerge either in phlogeny or in development.

strongest version is that clusters emerge exclusively thru experience.

p 36:  simplest acct of clustering is that it relfects large scale connectivity.

eg distinct clusters for faces and for scenes ... mosre similar in monozyg than dizyg twins.

STS (super temp sulcus) receives strong input from MT/ MST complex: shows strong sensit to

expression adn other dynamic aspects of faces.  By contrast, OFA (occip face area and FFA)

are more sensitive to stabel aspects of face.

So, retinotopic biases may be propagated and be impt in constraining tuning.

Recurrent Processing:  page 36:  see col 2:  top-down goals  (attn and goals) contrib to the

state of entire network (even) before stim onset!

Visual attention is known to affect proc. of perceptions in occiptotemp network.

Dense bidireciton connections along central route.

Change blindness often occurs when change does not alter the gist of the scene or

a current relevant feature..  Undetected (even large) changes evoke less activity than detected changes.

Same for backward masking (in time)... would not affect a strictly feedfwd system.

shows import of recurrent connex even for initial basic visual object perception.

page 37:  3 separate paths for cortical-subcortical output paths:

Occipitotempr-neostriatal path:  OT-neostraitum.  habit formation, procedural learning.

in humans  typically implicit and not verablizable.  visual habit formation

to caudate  (differ btwn head of caudate:  coding for explicit rules avs  tail of caudate (complex visual stim);

Occipitotemp  to ventral striatum  (to nucl accumbens)

ventral striatum receives from orbitofrontal cortex, OFC,  medial frontal, HC, amygd...

participates in assigning value (+ or -)  to partic stim.:  anticip  reward and to magnit of reward.

(indep of the action that monkey takes to get the reward... but they do encode the

genl effort or cost required to get the reward.

In human fMRI:  ventral striatum  activated by affective valence in both social and non-social domains.

area also genl activ by unexpected or novel stim.  area has dense limbic and medPFC connex;

region may integrate output of reward based systems for purpose of motivation or prioritzing actions.

p 39:  Occiptiotemp-amyg paths:  amygdal receives most of its input from aIT but projects back to

almost every area in the OT net  (unlike neostriatum which projects back to NONE!)

These provide nec input for visual-depend. emotional regulation for which amygdala is critical.

basolat nuclei of amygd have many neurons that are visual.. eg for face ident and expression.

Amygd  stignals current reward value of a stimulus.

lesions of amygd reduce the response to fearful faces but not the slectivty of the region.

amygdala may direct attention to potential danger and other emotionally intense events.

Cortico-cortical output paths:

Occipitotemp to MTL path used to encode LT memory of object quality.

prob also object info for partic landmarks, palces views of enviro.

page 40  proposed that  HC is optimized to maximize distinctions btwn similar patterns

enabling differential encdoing of similar events for selective retrieval.

lesions of HC =>  impaired recllection of specific events  (leaving familiar events intact)

so  PRh  (perirhinal... may be used to subserve familiar events (unlike HC which empasizes new events).

page 41 col 2:  In humans the lOFC (lateral)  strongly activate when the expected value of a stim choice

is updated.

By contrast,  mOFC  more involved in fine-grained comparisons btwn reward values.

In huamns,  mOFC  signals the current level of tward.

p 41:  OT-  to vlPFC..

the entire  lateral PFC is involved with maintenance and manip of  task-relevant info repres in

posterior cortices.  (attention,  workingmemory, switching task set.  

But complete removal of the area doe not impair basic perception or categorizaiton.

Concluding Remarks p 42....

we propose an expanded neural framework for object quality:

recurrent net;  visual info toat least 6 cortical and subcort areas each mediating

difft forms of learning and memory;

each distinct area performs specialized proc of distinct aspects that are

eventually syntesized into uniform repres in aIT and in MTL.

p43 in contast to the occiptoparietal network  much of the ventral network does not receive

feedback from output targets.  This lack of direct feedback might enable ventral areas to

more faithfully reflect stats of visual experience

Ventral Visual Pathway 2012 Kravitz Ungerleider Mishkin

Attachments