fMRI data + stimuli loading for SOC project

This respository contains the code used to generate the stimuli presented in the Winawerlab Second-Order Contrast project experiments from 2015-2016, as well as code to load the data from those experiments in a canonical format with plenty of metadata.

First, make sure to point the soc_fmripath.m and soc_stimpath.m scripts at the appropriate locations to find the fMRI data and the stimuli. These are by default set to point to the winawerlab server, but you can point them at a local copy of the data if you want.

Experiments

This repository pertains to four individual fMRI subject datasets:

• 2015-04-07 - subj004

  • This pilot study was run just before VSS, to test two new image types - "waves" and "noise bars" - alongside two existing image types - "gratings" and "patterns". Only these four stimulus types were run.

• 2015-06-19 - subj022

  • This is the first "full" dataset collected at NYU for the SOC experiment. A variety of images were used: patterns, gratings, noisebars, and waves, at different sparsities, orientations, and contrasts; as well as "cross-modulated" gratings. No retinotopic stimuli were used, becasue the hope was that a separate retinotopy scan for this subject could be used. This dataset is designated the "horizontal" dataset, because the default orientation for gratings was horizontal.

• 2015-10-22 - subj001

  • This is the second "full" dataset collected at NYU for the SOC experiment. The stimuli used are basically identical to those used in June, except this is the "vertical" dataset, in that every stimulus was rotated 90 degrees before display, making the default orientation consist of vertical bars. The gratings-versus-patterns discrepancy is less pronounced at vertical orientations, suggesting some orientation-dependence.

• 2016-02-17 - subj001

  • This small dataset explores spatial frequency tuning, with the intent to inform the design of a multiband model. Pattern stimuli at different spatial frequencies were used. Noise stimuli were generated, but not actually used in the experiment.

Code organization

The subdirectories of this repository are:

• dataload (load_subj scripts)

  • These scripts load in all the data and stimuli from a given run into a "data" struct which contains metadata

• imagereshape

  • Stacks of images are represented in different ways across this code and KNK's code. Stimuli consist of images (X * Y pixels) of different categories (C) where each category can have multiple examples (frames, F), and each individual image can be represented in a multi-band format with several bands (B).

    • cell: A cell array of C categories, each cell containing a matrix of images shaped X pixels * Y pixels * F frames. Used in the file stimulus.mat downloadable from KNK's website, but nowhere else in the code.

  • stack: A matrix X * Y * C * F. Can also have a fifth dimension B for bands. This is the structure that most easily allows slicing and indexing, for visualizing images.

    • flat: A matrix (X * Y) * (C * F) * B bands, where each image has been turned into a vector, and category boundaries have been removed, although band distinctions remain. The "model steps" mostly operate on flattened images for efficiency purposes

  • pixel vector: A matrix (X * Y) * C * F, where each image has been turned into a vector, but category boundaries remain. KNK's model-fitting code expects something like a pixel vector, but with the dimensions flipped to be C * (X * Y) * F, so interfaces with KNK's code will often include a "permute" step

• knkcode

  • All the code in this repository was originally written with a dependency on the knkutils package. For explicitness, the specific dependencies have been pulled out.

• stimulusgen

  • stimulusgen_<date> scripts contain exactly what was run to generate the stimuli shown for a specific experiment. Although the random seed was unfortunately not always captured, the files saved as stimuli-<date>.mat /should/ contain the actual original stimuli used, whereas stimuli-<date>_rerun.mat were regenerated at a later point to test the script. However, if in doubt, you can always go fetch the .mat file from the "Stimulus" folder of the fMRI_CBI folder, which is guaranteed to be the one that was actually shown to the subject.

  • The other functions in this directory are helper functions to generate certain classes of stimuli, bandpass filters, and so on.

• utils

  • These functions capture a few common motifs in masking and indexing.