Spatial Frequency Preferences in Human Visual Cortex

Billy Broderick

Simoncelli and Winawer labs

Oct 19, 2017

Spatial frequency

to put what I just said onto a set of axes, here I'm plotting these two hypotheses, which I'm going to call the constant and scaling hypotheses, as the peak spatial frequency vs. the eccentricity of the receptive field.

in the constant hypothesis, it doesn't matter where you are in the visual field, the cells prefer the same spatial frequency. in the scaling hypothesis, the preferred spatial frequency drops inversely with the distance from the fovea.

there's some existing data on this, but not as much as you may think that looks at this systematically, and those that I found did not agree with each other.

I would like to analyze how visual cortex responds to different spatial frequencies across the map in the primary visual cortex. in order to do this systematically, we need fMRI, which allows us to efficiently measure the response in V1 across all eccentricities in response to images.

for the experiment, we have people lie in the scanner and stare at these stimuli for a long time. each class of stimuli contains 8 images which differ in their phase but have the same frequency information. the images flash on and off for 300/200 msecs. (both the phase differences and on/off periods are done to reduce effect of adaptation). people are shown these images sequentially while fixating in the center at a digit stream, doing a one-back task as a distractor to maintain fixation and a constant cognitive state. there are 52 of these different stimulus categories, covering a wide range of frequencies, and we intersperse 10 "blank classes" to act as baselines (distractor task still happens).

that's one run, and we have 9 of these, with the stimulus class order randomized across runs

after some analysis, we end up with a number for the response of each voxel in the cortex to each of our stimuli.

Eccentricity bands

and this is more or less what we see

in this top plot, we're plotting it as a function of local spatial frequency and we see the curves are shifted around, with different eccentricities having their peaks at different frequencies.

in the bottom, we're plotting it as a function of the stimulus image, and we see the lines are all basically on top of each other, as the scaling hypothesis predicted.

however, there is this interesting amplitude effect and some divergences from the curve – which, though it's hard to see with the fit line, are basically consistent across eccentricities as well. there's also the possibility of edge effects going on here and I need to decide figure out how to handle that

it can be kind of hard to see where the peaks of the curves are because of this amplitude effect, so I'm going to replot this, taking the peak spatial frequency of each curve and plotting it against the spatial frequency.

the peak spatial frequency falls off inversely with receptive field size. I had to rescale the beginning and end of this curve here, and it doesn't exactly follow it, but it's a hyperbola, for sure.

so in conclusion

Receptive field varies systematically across visual field.
V1 neurons have different spatial frequency tuning.
Does V1 spatial freuqeny tuning differ across visual field?
Preliminary results suggest preferred spatial frequency shrinks inversely proportional to receptive field size.

I'd also like to emphasize, though I've formulated this talk as being all about these two qualitative hypotheses, mainly because of time constraints, that's not the actual point of this project. The ultimate goal is to use this data to get careful measurements of these spatial frequency tunings and use that to create forward models of the visual system, which can predict the responses of voxels across the visual cortex to arbitrary visual stimuli.

Thanks!

Jon Winawer
Eero Simoncelli
Noah Benson
Eline Kupers
Winawer and Simoncelli labs

existing data conflicts.

I looked across all the papers that I could find that measured this systematically (if you know, please tell me!) and this is what I found. these are all done in humans with fMRI, but I would love to see similar efforts in physiology as well. these studies generally used different types of stimuli and were aware of each other, citing each other, but they cited each other approvingly, saying "this generally matches", instead of talking about differences or why they could be.

the basic intuition of lower spatial frequencies in the periphery rather than the fovea, holds. and they basically all agree that the drop-off should be exponential rather than linear (these are approximated / averaged from the original data)

but results at a given eccentricity in V1 span about two octaves, or one if we discount the Kay results, unsure how this varies across subjects, not sure exact form of relationship between, or the effect of different stimuli types

Spatial Frequency Preferences in Human Visual Cortex

Billy Broderick

Simoncelli and Winawer labs

Oct 19, 2017

Spatial frequency

Eccentricity bands

Eccentricity bands

Thanks!

Psychophysical evidence of different spatial frequency channels

What is spatial frequency?

What is spatial frequency?

What is spatial frequency?

How do neurons calculate spatial frequency?

What does this have to do with biological vision?

Vision differs across visual field

Vision differs across visual field