Notes prepared  by Flavia

                                                107B Lecture from 02/09/2004

 

V1 to V2 projections/ pathways:

 

Blobs (color, brightness)           à thin stripes

Interblobs (orientation)              à thick stripes, interstripes

layer 4B (direction)                   à interstripes, thick stripes

                                                 

------------------------------------------------------------------

magno dLGN à 4cα à 4B (V1) à MT à MSTd

------------------------------------------------------------------

parvo dLGN à 4cβ à layer 3 (V1)à V2 (thin stripes)à V4 à posterior inferotemporal ctx.

------------------------------------------------------------------------------------------------------------

 

Aperture problem continued

 

The problem is figuring out what direction a whole pattern or object is moving in given local measurements of local/component motion (e.g. in V1).

In V1 (layer 4B), only motion perpendicular to the edge of an object can be measured.

Motion perpendicular to a contour cannot be measured at all in V1 – because it can’t be detected.

 

                                    

 

How do we know whether V1 can solve the aperture problem?

à Easy:

 

1)      map direction selectivity of a layer 4B V1 neuron (with a single bar). Say it prefers “down” direction.

2)      Now move a plaid grating in the “down” direction. The neuron does not respond, because it does not detect the downward translation of the entire plaid pattern, and it is getting confused by the individual bars making up the plaid pattern, which appear to be moving down and to the right, and down and to the left.

 

                                                           

                                                               

                                         cell does not respond                                             cell responds     

 

3)      Instead, the cell gives a response when the plaid pattern is moved i)down and to the right and ii) down and to the left. This is because the individual

4)      (= component) bars of the plaid now appear to be moving downward, the cell’s preferred motion direction. Hence the cell only detects the COMPONENT, i.e. LOCAL motion, but not PATTERN motion. It therefore has an aperture problem.

 

 

MT solves the aperture problem for pattern translation

 

In contrast to cells in V1, area MT (= middle temporal) does detect the overall pattern motion. When tested with the first grating above, an MT cell that prefers downward motion responds correctly.

Notice that:

1)      V1 has very narrow tuning curve (indicating a specific preferred motion direction), whereas MT has a broader tuning curve (i.e. cells in MT have a less narrow direction preference, e.g. instead of exactly vertical up, anything roughly up will elicit a response from the cell;

2)      MT neurons have bigger receptive fields than V1 cells – i.e. they see a bigger portion of the visual scene/field – and hence a bigger chunk of the pattern/object is seen by MT.

 

 

How does MT solve the problem for pattern translation?

 

                                                    Here, simply averaging the two thin lines (the local motion vectors) will not result in the actual pattern motion vector (= thick line). V1 cells are fooled by the local contours of the plaid.

 

 

 

                                       vs.           

 

 

MT has its own aperture problem for flowfields:

 

MT neurons are not selective for:

 

We know, because when testing an MT cell with a clockwise versus counterclockwise rotating stimulus, the cell gives the same response (indicating that it might be responding to a subcomponent of the rotation movement – e.g. “up” or “down”;

-         or, the cell might not respond the same to a particular direction of rotation depending on where in the receptive field it is shown. This lack of invariance indicates that the receptive field is only picking up a subcomponent of the rotation, rather than rotation per se.

 

MSTd (medial superior temporal area, dorsal aspect) IS selective for these types of optic flow patterns. We know, because an individual MSTd cell might be selective for, say, counterclockwise rotation, but won’t fire to clockwise rotation, and will respond the same to its preferred stimulus no matter where that stimulus is shown in its receptive field (position-invariant response).

 

How does MSTd solve the aperture problem for more complex motion patterns?

- cells have a template for preferred motion directions in their receptive fields.

 

 

 

KNOW!!!

 

      à lack of selectivity of response, i.e. responds to multiple stimuli that differ                    along some dimension we care about; e.g. responds the same to up and down, or          ccw and cw, or a white shirt in an orange light and a blue shirt in an orange light.

      à shows you what the cell preferentially responds to, e.g vertical up motion.     What the cell is “tuned” to.

      à show tuning curves of a V1 cell, and its responses to a plaid pattern moving in           its preferred direction