Skip to content

fMRI Preprocessing

Jump to bottom
Taylor Salo edited this page Apr 12, 2017 · 2 revisions

Preprocessing fMRI Data

Before analyzing fMRI data, you must preprocess it to remove artifacts and, generally, to warp individual subjects' data into a standard space to compare across subjects.

Typical steps:

  • Slice timing correction
    • For what data should we do this? Does it depend on experimental design and/or acquisition parameters?
    • When in the pipeline should it be done?
  • Motion correction
    • Which distance metric is best?
  • Nuisance signal extraction
    • Examples
      • CSF/WM signal (e.g., aCompCor)
      • DVARS
    • Which nuisance regressors are best? Does regressor choice depend on experimental design and/or acquisition parameters?
  • Coregistration (linearly register functional data to a higher resolution structural image)
  • Segmentation
  • Normalization (linearly or nonlinearly warp subject-space data to some standard space)
  • Spatial smoothing
  • Highpass/Bandpass filtering
    • Should we do this at the same time as modeling? Simult paper.
  • Modeling
    • Task timing information
    • Motion regressors
      • Framewise displacement
      • Derivatives of motion parameters
    • Additional nuisance regressors
      • Censored volumes based on motion or signal change
      • Physio
      • CSF/WM signal
  • Contrast generation
    • Linearly combine beta (and sometimes error) maps for conditions of interest and compare statistically
  • Group-level analysis
Clone this wiki locally