How does the accuracy of machine-learning classifiers depend on fMRI resolution?

High-field (7T) high-resolution (1.2mm) fMRI and eye tracking data from human subjects repeatedly watching 5-minute clips of naturalistic (cinematic) movie content without sound. A high number of repeated stimulus presentations and fMRI acquisitions at two different resolutions (1.2mm and 2mm isotropic) were chosen to investigate the accuracy of machine-learning classifiers and to test its dependence on fMRI resolution. Please find details in below publication Mandelkow et al. 2017, which should be quoted as a reference. All data were acquired at the National Institutes of Health under NINDS protocol 00-N-0082 in 2012-2015.

Effects of spatial fMRI resolution on the classification of naturalistic movies,
H. Mandelkow, J.A. de Zwart, J.H. Duyn, 2017, NeuroImage
https://doi.org/10.1016/j.neuroimage.2017.08.053

Keywords: fMRI; High-resolution; High-field; 7T; Multivariate pattern analysis; MVPA; Naturalistic stimuli; Movies; Eye tracking

---

Data Files

NOTE: This dataset comprises 7 subjects, 4 with >16 movie stimulus repetitions. Of the original 10 subjects three were discarded without analysis due to subsequent changes in the experimental protocol (MRI sequences).

fMRI data

sub-02_ses-05_task-Mx1_acq-Hr_run-2_bold.nii.gz

One run of BOLD fMRI data in Nifti format (.nii.gz).

• sub-## : human subject number ##
• ses-## : experimental session number ## (1-5 separate sessions per subject)
• task-* : stimulus presented (5' movie clip)
  • task-Mx1 : The Matrix I opening scene
  • task-Mx2 : The Matrix II
  • task-Gbu1 : The Good, the Bad, and the Ugly
  • tast-ffMRI : single-slice (fast) fMRI with high temporal resolution without stimulus e.g. for characterizing cardiac artifacts
• acq-Hr / acq-Lr : Higher- / Lower-resolution fMRI at 1.2mm / 2mm isotropic
• run-# : experimental run number within each fMRI session in temporal order

MRI anatomy

sub-02_ses-05_acq-AnaGE_T2w.nii.gz : gradient-echo anatomy (magnitude)

sub-02_ses-05_acq-AnaGEp_T2w.nii.gz : gradient-echo anatomy (phase)

Eye-tracking data

sub-02_ses-05_task-Mx1_acq-Hr_run-2_vpdat.txt

• *_vpset0.txt : calibration before movie start
• *_vpset1.txt : calibration at movie start
• *_vpdat.txt : eye-tracking data
• *_vpset2.txt : calibration at movie end
• *_vpset3.txt : calibration after movie end