Adjudicating between face-coding models with individual-face fMRI responses

This is a human fMRI dataset that investigates coding of individual faces in the
visual cortex of healthy human volunteers. We also include behavioral data from
a similarity judgment task, and computational models that can be used to fit
both data modalities.

See the associated reference (Carlin & Kriegeskorte, in press, PLOS CB) for
details about the experimental protocol. In this README we focus on technical
details that may be useful for re-analysing the data.

The behavioral and fMRI distance matrices as well as the computational modeling
efforts have been shared previously at OSF (https://osf.io/5g9rv) and zenodo
(https://doi.org/10.5281/zenodo.242666), so this may be an easier way to go if
you don't want to re-run the entire fMRI preprocessing pipeline.

DICOM TO BIDS CONVERSION
DCM files were converted to nifti using dcm2niix v1.0.20170130
(https://github.com/rordenlab/dcm2niix), and dcm2bids
(https://github.com/jooh/Dcm2Bids). Anatomical images were de-faced using
pydeface (https://github.com/poldracklab/pydeface).

DATA ANALYSIS SETUP
We analysed data using Matlab R2013A, SPM, FSL, and various custom software
developed in Matlab. The following packages (and their associated
dependencies) are necessary to get the included analysis code to run:

* Automatic Analysis 5 (AA, https://github.com/rhodricusack/automaticanalysis)
* pilab (https://github.com/jooh/pilab)
* pilab_aa (https://github.com/jooh/pilab_aa)
* facedistid_analysis (https://github.com/jooh/facedistid_analysis - an up to
date copy is available in the code folder)

In general, the AA pipeline generates all fMRI results and figures (Figs 1-2,
S3-S4 Figs in the manuscript). We then extracted fMRI distance matrices from
cortical regions of interests for further computational modeling (remaining
figures in the manuscript).

KEY FILES
facedist_aa_frombids.m The master function for running the AA
fMRI analysis pipeline
facedist_aa_frombids_tasklist.xml Specifies which AA modules to run - note
that the roiroot flag specifies an
absolute path that will need
updating for your file system
facedist_doit_facepairs The master function for running the
behavioral similarity judgment
analysis
facedist_doit_modeling The master function for running the
computational model fits (you will
need to run through
facedist_aa_frombids and
facedist_doit_facepairs first to
generate intermediate results)
derivatives/aa/aap_prov.png Nice visualisation of fMRI result
provenance
derivatives/rois ROI masks for fROI analysis (if you want
to re-define ROIs from the localiser
data you can do so using
https://github.com/jooh/roitools/blob/master/spm2roi.m)
misc/data_perceptual_judgment_task.mat data from behavioral similarity judgment
task
misc/stimuli_mri.mat video stimuli used during MRI scanning
misc/stimuli_perceptual_judgments.mat video stimuli used during behavioral
task

A NOTE ON REPRODUCIBILITY
If you run the above pipeline you will obtain results that are very similar to
those in the manuscript (which, again, are publicly available on OSF/Zenodo),
but not identical. This is because of the following differences with regard to
the analysis in the paper:

* The paper analysis used SPM8, not 12
* The paper analysis used SPM dicom conversion, not dcm2niix
* The paper analysis included a super-extraneous conversion of the floating
point precision on the niftis during preprocessing, which did nothing but
blow up file size it turns out.
* The paper analysis used a bastardised version of AA4, not 5, which probably
introduces lots of subtle differences in the preprocessing parameters (for
this AA version, see
https://github.com/jooh/automaticanalysis/tree/v4-master)
* Lots of other dependencies (including pilab) continued to be developed and
improved.

Note that in particular, the ROI masks were generated using the old analysis, so
the results could definitely be improved by re-running ROI definition, if
someone has a few days to spare... But again, discrepancies are very small and
do not qualitatively change any conclusions made in the paper. Exact
reproducibility in neuroimaging is hard. If you want to inspect the AA analysis
that is reported in the paper, please get in touch and we will see if there is a
way to convince the MRC to let you have access to non-anonymous data.

REFERENCE

Carlin, J.D & Kriegeskorte, N. (in press). Adjudicating between face-coding models
with individual-face fMRI responses. PLOS Computational Biology. See
BioRXiv for a preprint (2017, original version 2015): https://doi.org/10.1101/029603

CONTACT

Johan Carlin, MRC CBU, Cambridge, UK. [e-mail]

### Comments added by Openfmri Curators ###
===========================================

General Comments
----------------

Defacing
--------
Pydeface was used on all anatomical images to ensure deindentification of subjects. The code
can be found at https://github.com/poldracklab/pydeface

Quality Control
---------------
Mriqc was run on the dataset. Results are located in derivatives/mriqc. Learn more about it here: https://mriqc.readthedocs.io/en/latest/

Where to discuss the dataset
----------------------------
1) www.openfmri.org/dataset/ds000232/ See the comments section at the bottom of the dataset
page.
2) www.neurostars.org Please tag any discussion topics with the tags openfmri and ds000232.
3) Send an email to [e-mail]. Please include the accession number in
your email.

Known Issues
------------
N/A

Bids-validator Output
---------------------
1: Not all subjects/sessions/runs have the same scanning parameters. (code: 39 - INCONSISTENT_PARAMETERS)
/sub-02/ses-01/anat/sub-02_ses-01_T1w.nii.gz
/sub-02/ses-01/func/sub-02_ses-01_task-localizer_run-01_bold.nii.gz
/sub-02/ses-01/func/sub-02_ses-01_task-localizer_run-02_bold.nii.gz
/sub-02/ses-01/func/sub-02_ses-01_task-main_run-01_bold.nii.gz
/sub-02/ses-01/func/sub-02_ses-01_task-main_run-02_bold.nii.gz
/sub-10/ses-01/anat/sub-10_ses-01_T1w.nii.gz

Summary: Available Tasks: Available Modalities:
1120 Files, 34.38GB localizer T1w
10 - Subjects main bold
4 - Sessions