N8 - Visual Streams And Binding

Front 1

What are the two visual streams?

Back 1

Ventral (occipitotemporal)"what" stream - identification of objects

Dorsal (occipitoparietal) "where" stream - spatial releationships including motion, and for guidance of movements

First identified through lesion studies e.g. Mishkin et al 1983

Although not just serial from V1→ end of stream, but lots of jumping, backwards, recurrent connections, between paths.

Front 2

What's the difference between P cells and M cells

Back 2

Parvocellular and Magnocellular cells in LGN

Parvocellular = small → responsible for form/colour→ feeds into what?

Magnocellular = big → responsible for form/motion→ feeds into where?

Front 3

What is Marr's theory of object recognition in the ventral stream? (1982)

Back 3

Presumes a hierarchy of visual processing → increasingly complex representations i.e.

1. 2D Sketch = 2D description of input - high contrast, uses gestalt grouping processing principles to process what is an object
2. 2.5D Sketch = Still viewpoint dependent, but processing depth, orientation, shading texture to start to get to 3D
3. 3D model = viewpoint invariant 3D representation of objects and relative position

Front 4

What's the progression in the ventral stream?

Back 4

RFs get larger as you go furher back → more representation of both visual fields in both hemispheres∴ trigger features actually get much less location specific, more about complex/specific triggers. Also cares less about variability in light, angle, occlusions→ very good at this compared to machines

• V1 responds to lines orientations (Hubel & Wiesel 1959)
  
• V2 responds to edges (Van der Heydt et al 1984 - used illusory edges)
• V4 responds to conjunctions of features e.g. different slants of a line (Hinkle & Connor 2002)
• IT responds specifically to faces, even in infants! (Rodman et al 1993).

Front 5

What else might be special about IT cell responses?

Back 5

• Respond to faces (innate, expert?)
• Show object constancy - if lesion IT → can't recognise rewarded object as being same if in diff shadow/orientation (Weiskrantz & Saunders 1984)
• Show more mnemonic effects i.e. different responding with multiple presentations - some habituate, others enhance (Miller & Desimone 1994)
• Cell A and B respond to stim A and B. If A and B presentation is paired, then the activity of A and B becomes more similar. If unpaired - diverged = memory? (Messinger et al 2001)
  

Front 6

What are the two types of visual agnostic patients (unable to recognise objects) - how can they help us understand how object processing works

Back 6

Identified by Lissaier (1980)

Apperceptive = Early stages of object recognition - either to do with inability to group edges, bind features or normalise views

Associative = can do perceptual tasks, but can't recognise stuff....

Front 7

What are the specific problems in Apperceptive Agnosia patients?

Back 7

• Patient DF - can't recognise or copy line drawings of objects, bad at gestalt grouping e.g. colinearity. Problem with Occipital → same area that responds well to colinear lines in healthy patients. (Milner et al 1991)
• Patient HJA - couldn't bind or integrate object features (but could copy/draw objects from memory). Can discriminate objects when they're separate, but not when they're superimposed/occluded. Definitely binding problem. Still can't recognise objects. (Humphreys & Riddoch 1987, Giersch et al 2000)
• Patients with right Posterior cortex damage - can't match drawings from diff views, but can categorise by semantics. (Warrington & Taylor 1978, Warrington & James 1988). Warrington then proposed left/right distinction between semantic/perception - but probably too simplistic.
• Although... right fusiform seems to respond more to perceptual manip, left fusiform responds more to semantic manip, so may be correct (Simons et al 2003)

Front 8

What are the specific problems in Associative Agnosia patients?

Back 8

• Some can't distinguish real from non-real objects i.e. trouble distinguishing real from nonreal. (Riddoch & Humphreys 2001)
  
• Patient JB can't match semantically similar pictures, although fine with words (Riddoch & Humphreys 1987)
• Neuroimaging suggests Mid fusiform = about structural decisions e.g. real/notreal, whereas semantic categorisation is more in anterior IT i.e. kinda hierarchical.

Front 9

What is the Grandmother Cell Hypothesis (Quiroga et al 2005)

Back 9

Found a cell that responds to different pictures of Jennifer Anniston, but not when with Brad Pitt → One cell per object/person?!

Unlikely∵ cells die regularly BUT could just be very specific object recognition i.e. diff features recognised and they together fire one cell - kinda like one cell/person, but more flexible...

Front 10

What does the Bruce and Young Model of face recognition state

Back 10

• Similar to Riddoch/Humphreys model i.e. structure, then semantic distinction, with separate expression/speech processing
• Dissociation between recognising familiar matching unfamiliar face→ difference in recognition and structural combinations (Malone et al 1982)
• Name generations only really happen if you don't have any other info generally i.e. unlikely to know name if you don't know anything else when you see their face. (Young et al 1985 diary study, Flude et al 1878 famous faces study). Although can build up identity without face e.g. authors/directors....
• Serial nature of network can be seen in speed of recognition i.e. fastest at familiar or not, then occupation, then name (Young et al 1986)

Front 11

Are faces special?

Back 11

Is the IT an innate face module, or are we just v specialised for recognising faces

Pro

• Stimulation of IT neurons→ biased to report noisy stim as faces→ could just be specialised though... (Afraz et al 2006)
• Prosopagnosics (right posterior cortex damage) can't recognise faces, but can recognise by voice. Also can be expert in other stuff e.g. coins, cars, sheep (Da Renzi et al 1986, 1991)
• Neuroimaging suggests specialised fusiform face area (Kanwisher et al 1999)

Anti

• Better at recognising faces within race (Tanaka et al 2003)
  
• Inversion effect study - i.e. all people are much worse at recognition memory of faces if they're inverted. Is it because of expertise - dog experts show same effect on dogs, novices are fine (Diamond & Carey 1986)
• Other Neuroimaging sugests the fusiform face area is just responding for expertise (Gauthier et al 1999)

Front 12

What are the various inputs for the parts of the dorsal stream?

Back 12

MT is first part - gets input from V1, V2, V4, specific for Motion

Parietal lobe in general receives lots of inputs e.g. processed/polymodal from association cortices, afferent moto inputs, limbic inputs

Front 13

What are the LIP / VIP neurons?

Back 13

both dorsal stream -

LIP responses are retinotopic but modulated by eye position, attention, behavioural relevance e.g. can respond to stimulus in the RF that they're about to saccade to even when the stim is removed during saccade i.e. knows planned moves (efference copy?) (Duhamel et al 1992) PLUS eye centred

VIP neurons prefer moving stim - selective for speed/orientation, polymodal PLUS head centred

This may mean that for small head movements, the LIP knows where stim is and can move eyes to get it back into RF, but if movement of gaze, VIP still has an idea of where stuff was (because head centred) and can direct movement back. Generally both are able to direct attention

Front 14

What are the problems in the way that the dorsal stream has been studied

Back 14

Animal lesion studies - homology of the regions between animals and humans isn't wholly clear... especially when you consider humans more complex planned motor actions. e.g. Macaque parietal v different

Human lesions aren't much help either because broad

fMRI is tentative - may be that TMS will be useful in future/fMRI getting better?

Front 15

What is the Ungerleider Mishkin (1982) theory of the streams?

Back 15

Largely came from parietal lesions in animals→ parietal lesions→ contralateral neglect/extinction (if less severe). Neglect can even be in mental imagery e.g. of Duomo in milan (Bisiach & Luzzatti 1978). Fine at ventral stream processing on both sides i.e. objects on neglect side can still prime for word/non-word decision even though not attended (McGinchley-Berroth et al 1993)

Similarly Balints syndrome can be caused by bilateral parietal lesions = simultagnosia, so can't perceive more than one object∴ no spatial relationships, although if it can be framed as a single item e.g. a face, not two circles → can perceive it (Shalev & Humphreys 2002)

Neuroimaging finds more dorsal activation in spatial comparison space task, more ventral stream for facial recogntion, although V1 in both (Haxby et al 1991). Activation in neglect patient GK is fine for early/ventral stream i.e. identical for bilaterally presented (where it isn't perceived) as it is to unilateral presentation (where it is perceived).

Front 16

What is the Milner & Goodale (1974) Theory of visual streams

Back 16

Used mainly data from neuropsychological studies - suggested that it's more about perception/action rather than objects in space. i.e. using the objects in that space.

Dissociation between impaired action and impaired recognition:

• optic ataxia = superior parietal lesion→ impaired touching tip of finger (Rizzo & Vecera 2002), can't orient hand for slot/reach correctly - failed integration of visual/motor information
• visual agnosia = ventral lesion. Impaired object recognition, copying/recognising objects, discriminating edges, matching letterboxes with diff orientations BUT fine at orienting T shaped object to a slot
  

Ventral lesioned patient (DF) can't match irregular wooden shapes, but can pick them up at the correct grasp points, whereas dorsal lesioned patient (RF) can match the shapes, but can't pick them up. (Goodale et al 1994)

Illusions even show the action/perception distinction→ 3D setup of Muller Lyer - two identical lines perceived as diff sizes + estimate length with fingers, but as you grasp that's corrected nad becomes accurate (Hart et al 1999)

Front 17

How cna we resolve the differences between the ungerlieder/mishkin model and Milner/Goodale model

Back 17

Creem & Proffitt (2001) suggest ataxia may be more superior lesions, neglect more inferior lesions→ superior about action, inferior about spatial processing. BUT not actually that easy to separate the lesions e.g. neglect also seems to be lateralised i.e. hemisphere specific unlike ataxia + no neuroimaging support

Glover (2004) suggested planning-control model where inferior parietal selects targets for action/plan it, superior parietal monitors accuracy/makes adjustments based on spatial info. e.g. in 3D Muller lyer or 3D Ebbinghaus, the fingers close in ∵ adjusting

Front 18

How are features of stimuli bound together - given features are processed kinda separately e.g. colour, shape, motion→ single image

Back 18

May be that they all converge on a single grandmother-like cell (probs not because not enough cells/they die) or form a cell assembly i.e. briefly act as a closed system→ then activates other systems (Hebb 1949)

Front 19

What could be used to bind cell assemblies?

Back 19

• Spatial location i.e. spatially selective spotlight bind the unconsciously separate features. Indeed patients with neglect and those who have divided attention get illusory conjunctions ∵ attention → binding? Based on Treisman's feature integration theory.
• Firing rate i.e. two neurons with increased firing rate → encoding same object, but what if encoding more than one object? In this model only one set of neurons could be active, so only one cell assembly possible. Plus incompatible with rate codes
• Binding by synchronous firing - coactivation of firing in neurons coding same thing. Diff objects→ diff patterns. May actually lead to greater firing rates higher up, because synchronous firing is more likely to active postsynapses.

Front 20

What's the evidence for synchronous firing?

Back 20

Cat visual cortex - Single bar moving across both cells receptive fields→ synchronous firing, two bars movingg across RFs so it looks like a single bar→ synchronous firing BUT if diff directions, so clearly diff object → no synchrony (Singer et al 1990, 1993, 1995, Engel et al 2001)

Similarly single bars moving over multiple RFs → synchrony between them all, but if 2 bars at diff orientation → splits population so one set prefers bar 1 and syncs amongst selves, same for other bar. (Singer 1995, Engels 1992)

Front 21

How might neurons detect synchrony?

Back 21

Temporal summation in postsynapse w/ tight window (few ms) (Softky & Koch 1993)

Front 22

How are populations likely to form stable cell assemblies?

Back 22

Synaptic plasticity? May be too slow?

May be driven by 40Hz oscillations generated by thalamus and cortex→ permit/aid synchrony. W/o them→ no synchrony. May be initiated by reticular formation→ stim of reticular formation→ stimulation. Reticular formation important in sleep, so it makes sense that synchrony is diff between wake/sleep which it is....

Probably relies on corticocortical connections → if kitten given divergent squint→ disrupts intracortical connections in visual cortex→ no synchrony between groups on different eyes (Konig et al 1993)

Front 23

What does Treisman say about attention and how it influences binding? What's the evidence for this?

Back 23

Treisman suggests that not everything is bound and that the spatial attention spotlight binds things.

Based on popout i.e. if you have to bind features→ much slower to find something - needs to use spotlight to scan, whereas if just looking for a single feature → very quick popout.

Posner et al (1984) looked at valid/invalid cueing followed by a response. Found invalid cue→ longer ∵ has to disengage attention, move, reengage.

Can actually separate i.e.

• Disengage relies on posterior parietal cortex - invalid cue takes a lot longer if ipsilateral side lesioned (Posner et al 1984)
  
• Movement relies on superior colliculus - lesion slows controlat saccade onset/completion (Wurton & Goldberg 1972)
• Engage depends on pulvinar nucleus (thalamus) - lesioned monkeys slower to get to valid cue/target ∵ can't engage with it, but lesioned actually faster than regular in invalid cue range ∵ invalid cue doesn't engage attention

Front 24

What is Desimone and Duncan's biased competition model?

Back 24

Top down attentional bias starts to change the activity, then neurons have their own competition based on inputs and ultimately some stim fire neurons more effectively→ attended to.

Evidence for top down control

• If humans instructed to look for motion→ higher blood flow to motion area V5. Instructed to attend colour→ higher blood flow to V4. Even works if no relevant stim there.
  
• fMRI shows increased activity in FFA (fusiform face) over PPA (parahippocampal place) when asked to attend face in face/house superimposed stim
  

Evidence for bottom up competition

• Masking effect where cells with a certain RF respond only to green bars, only blue bars, but still have a reduced response if both are presented in the RF. (Response greater if told to attend blue = top down) (Luck et al 1997)
  
• Similar effect in V2 cells where horizontal bars are preferred, but response reduced by vertical bar UNLESS attention paid to horizontal bar→ horizontal outcompetes → more response (Reynolds and Desimone 1999)

Interestingly, top down control has same effect on activity of cells as doubling stimulus salience.

No effects in V1, just in higher cortex. May be that RFs in V1 small so unlikely to even have two stim in one V1 RF.

How does this relate to binding? Hotly debated→ some believe that attention mods the binding (Luck, Fries), others argue that binding gives you the groups that can then be attended to (Singer)

Front 25

How do the biased competition model and spatial spotlight models compare on how they think about attention?

Back 25

Both incorporate bottom up and top down attention i.e. voluntary and automatic controls

e.g. latPFC and frontal eye circuits for top down and LIP for bottom up (Buschman and Miller 2007)

May be that diff freq of synchrony i.e. top down = low freq sync, bottom up = high freq sync