Outputs of fMRIPrep

fMRIPrep generates three broad classes of outcomes:

  1. Visual QA (quality assessment) reports: one HTML per subject, that allows the user a thorough visual assessment of the quality of processing and ensures the transparency of fMRIPrep operation.

  2. Pre-processed imaging data which are derivatives of the original anatomical and functional images after various preparation procedures have been applied. For example, INU-corrected versions of the T1-weighted image (per subject), the brain mask, or BOLD images after head-motion correction, slice-timing correction and aligned into the same-subject’s T1w space or into MNI space.

  3. Additional data for subsequent analysis, for instance the transformations between different spaces or the estimated confounds.

fMRIPrep outputs conform to the BIDS Derivatives specification (see BIDS Derivatives RC1).

Visual Reports

fMRIPrep outputs summary reports, written to <output dir>/fmriprep/sub-<subject_label>.html. These reports provide a quick way to make visual inspection of the results easy. Each report is self contained and thus can be easily shared with collaborators (for example via email). View a sample report.

Derivatives of fMRIPrep (preprocessed data)

Preprocessed, or derivative, data are written to <output dir>/fmriprep/sub-<subject_label>/. The BIDS Derivatives RC1 specification describes the naming and metadata conventions we follow.

Anatomical derivatives are placed in each subject’s anat subfolder:


Spatially-standardized derivatives are denoted with a space label, such as MNI152NLin2009cAsym, while derivatives in the original T1w space omit the space- keyword.

Additionally, the following transforms are saved:


If FreeSurfer reconstructions are used, the following surface files are generated:


And the affine translation (and inverse) between the original T1w sampling and FreeSurfer’s conformed space for surface reconstruction (fsnative) is stored in:


Functional derivatives are stored in the func/ subfolder. All derivatives contain task-<task_label> (mandatory) and run-<run_index> (optional), and these will be indicated with [specifiers].


Volumetric output spaces include T1w and MNI152NLin2009cAsym (default).

For each BOLD run processed with fMRIPrep, an accompanying confounds file will be generated. Confounds are saved as a TSV file:


These TSV tables look like the example below, where each row of the file corresponds to one time point found in the corresponding BOLD time series.

csf white_matter  global_signal std_dvars dvars framewise_displacement  t_comp_cor_00 t_comp_cor_01 t_comp_cor_02 t_comp_cor_03 t_comp_cor_04 t_comp_cor_05 a_comp_cor_00 a_comp_cor_01 a_comp_cor_02 a_comp_cor_03 a_comp_cor_04 a_comp_cor_05 non_steady_state_outlier00  trans_x trans_y trans_z rot_x rot_y rot_z aroma_motion_02 aroma_motion_04
682.75275 0.0 491.64752000000004  n/a n/a n/a 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 -0.00017029 -0.0  0.0 0.0 0.0
669.14166 0.0 489.4421  1.168398  17.575331 0.07211929999999998 -0.4506846719 0.1191909139  -0.0945884724 0.1542023065  -0.2302324641 0.0838194238  -0.032426848599999995 0.4284323184  -0.5809158299 0.1382414008  -0.1203486637 0.3783661265  0.0 0.0 0.0207752 0.0463124 -0.000270924  -0.0  0.0 -2.402958171  -0.7574011893
665.3969  0.0 488.03  1.085204  16.323903999999995  0.0348966 0.010819676200000001  0.0651895837  -0.09556632150000001  -0.033148835  -0.4768871111 0.20641088559999998 0.2818768463  0.4303863764  0.41323714850000004 -0.2115232212 -0.0037154909000000004  0.10636180070000001 0.0 0.0 0.0 0.0457372 0.0 -0.0  0.0 -1.341359143  0.1636017242
662.82715 0.0 487.37302 1.01591 15.281561 0.0333937 0.3328022893  -0.2220965269 -0.0912891436 0.2326688125  0.279138129 -0.111878887  0.16901660629999998 0.0550480212  0.1798747037  -0.25383302620000003  0.1646403629  0.3953613889  0.0 0.010164  -0.0103568  0.0424513 0.0 -0.0  0.00019174  -0.1554834655 0.6451987913

If FreeSurfer reconstructions are used, the (aparc+)aseg segmentations are aligned to the subject’s T1w space and resampled to the BOLD grid, and the BOLD series are resampled to the midthickness surface mesh:


Surface output spaces include fsnative (full density subject-specific mesh), fsaverage and the down-sampled meshes fsaverage6 (41k vertices) and fsaverage5 (10k vertices, default).

If CIFTI outputs are requested (with the --cifti-outputs argument), the BOLD series are also saved as dtseries.nii CIFTI2 files:


CIFTI is a container format that holds both volumetric (regularly sampled in a grid) and surface (sampled on a triangular mesh) samples. Sub-cortical time series are volumetric (supported spaces: MNI152NLin2009cAsym), while cortical time series are sampled on the surface (supported spaces: fsaverage5, fsaverage6)

Finally, if ICA-AROMA is used, the MELODIC mixing matrix and the components classified as noise are saved:


FreeSurfer Derivatives

A FreeSurfer subjects directory is created in <output_dir>/freesurfer.


Copies of the fsaverage subjects distributed with the running version of FreeSurfer are copied into this subjects directory, if any functional data are sampled to those subject spaces.


The BOLD signal measured with fMRI is a mixture of fluctuations of both neuronal and non-neuronal origin. Neuronal signals are measured indirectly as changes in the local concentration of oxygenated hemoglobin. Non-neuronal fluctuations in fMRI data may appear as a result of head motion, scanner noise, or physiological fluctuations (related to cardiac or respiratory effects). For a detailed review of the possible sources of noise in the BOLD signal, refer to [Greve2013].

Confounds (or nuisance regressors) are variables representing fluctuations with a potential non-neuronal origin. Such non-neuronal fluctuations may drive spurious results in fMRI data analysis, including standard activation GLM and functional connectivity analyses. It is possible to minimize confounding effects of non-neuronal signals by including them as nuisance regressors in the GLM design matrix or regressing them out from the fMRI data - a procedure known as denoising. There is currently no consensus on an optimal denoising strategy in the fMRI community. Rather, different strategies have been proposed, which achieve different compromises between how much of the non-neuronal fluctuations are effectively removed, and how much of neuronal fluctuations are damaged in the process. The fMRIPrep pipeline generates a large array of possible confounds. Note, fMRIPrep does not perform any denoising itself and it is up to the user to perform this step.

The most well established confounding variables in neuroimaging are the six head motion parameters (three rotations and three translations) - the common output of the head motion correction (also known as realignment) of popular fMRI preprocessing software such as SPM or FSL. One of the biggest advantages of fMRIPrep is the automatic calculation of multiple potential confounding variables beyond the standard head motion parameters.

Confounding variables calculated in fMRIPrep are stored separately for each subject, session and run in TSV files - one column for each confound variable. Such tabular files may include over 100 columns of potential confound regressors.


Do not include all columns of confounds_regressors.tsv table into your design matrix or denoising procedure. Filter the table first, to include only the confounds you want to remove from your fMRI signal. The choice of confounding variables may depend on the analysis you want to perform, and may be not straightforward as no gold standard procedure exists. For detailed description of various denoising strategies and their performance, see [Parkes2018] and [Ciric2017].

Confound regressors description

Basic confouds

  • trans_x, trans_y, trans_z, rot_x, rot_y, rot_z - the 6 rigid-body motion parameters (3 translations and 3 rotation) estimated relative to a reference image;

  • csf - the average signal within anatomically-derived eroded CSF mask;

  • white_matter - the average signal within the anatomically-derived eroded WM masks;

  • global_signal - the average signal within the brain mask.

Parameter expansion of basic confounds

Only accounting for the standard six motion parameters may not be sufficient to remove all variance related to head motion from the fMRI signal. Thus, Friston et al. [Friston1996] and Satterthwaite et al. [Satterthwaite2013] proposed 24-motion-parameter expansion, with a goal of removing from fMRI signal as much of the motion-related variance as possible. To make this technique more accessible, fMRIPrep automaticaly calculates motion parameter expansion [Satterthwaite2013], providing timeseries corresponding to first temporal derivatives of six motion parameters, together with their quadratic terms, resulting in the total 24 head motion parameters (6 standard motion parameters + 6 temporal derivatives of six motion parameters + 12 quadratic terms of 6 motion parameters and their 6 temporal derivatives). Additionally, fMRIPrep returns temporal derivatives and quadratic terms for the csf, white_matter and global_signal to enable applying 36-parameter denoising strategy proposed by Satterthwaite et al. [Satterthwaite2013].

Derivatives and quadratic terms are stored under column names with suffixes: _derivative1 and powers _power2. These were calculated for head motion estimates (trans_ and rot_) and compartment signals (white_matter, csf, and global_signal).

Confounds for outlier detection

  • framewise_displacement - is a quantification of the estimated bulk-head motion calculated using formula proposed by [Power2012];

  • dvars - the derivative of RMS variance over voxels (or DVARS) [Power2012];

  • std_dvars - standardized DVARS;

  • non_steady_state_outlier_XX - columns indicate non-steady state volumes with a single 1 value and 0 elsewhere (i.e., there is one non_steady_state_outlier_XX column per outlier/volume).

All these confounds can be used to detect potential outlier time points - frames with high motion or spikes. Detected outliers can be further removed from time series using methods such as: volume censoring - entirely discarding problematic time points [Power2012], regressing signal from outlier points in denoising procedure, or including outlier points in the subsequent first-level analysis when building the design matrix. Averaged value of confound (for example, mean framewise_displacement) can be added as a regressor in group level analysis [Yan2013].

Spike regressors for outlier censoring can also be generated from within fMRIPrep using the command line options --fd-spike-threshold and --dvars-spike-threshold (default: FD > 0.5 mm or DVARS > 1.5). Spike regressors are stored in separate motion_outlier_XX columns.

ICA-AROMA confounds

  • aroma_motion_XX - the motion-related components identified by ICA -AROMA (if enabled with a flag --use-aroma) .


If you are already using ICA-AROMA cleaned data (~desc-smoothAROMAnonaggr_bold.nii.gz), do not include ICA-AROMA confounds during your design specification or denoising procedure.

CompCor confounds

CompCor is a component-based noise correlation method. In the method, principal components are calculated within an ROI that is unlikely to include signal related to neuronal activity, such as CSF and WM masks. Signals extracted from CompCor components can be further regressed out from the fMRI data with a denoising procedure [Behzadi2007].

  • a_comp_cor_XX - additional noise components are calculated using anatomical CompCor;

  • t_comp_cor_XX - additional noise components are calculated using temporal CompCor.

Four separate CompCor decompositions are performed to compute noise components: one temporal decomposition (t_comp_cor_XX) and three anatomical decompositions (a_comp_cor_XX) across three different noise ROIs: an eroded white matter mask, an eroded CSF mask, and a combined mask derived from the union of these.


Only a subset of these CompCor decompositions should be used for further denoising. The original Behzadi aCompCor implementation [Behzadi2007] can be applied using components from the combined masks, while the more recent Muschelli implementation [Muschelli2014] can be applied using the WM and CSF masks. To determine the provenance of each component, consult the metadata file (see below).

Each confounds data file will also have a corresponding metadata file (~desc-confounds_regressors.json). Metadata files contain additional information about columns in the confounds TSV file:

  "a_comp_cor_00": {
    "CumulativeVarianceExplained": 0.1081970825,
    "Mask": "combined",
    "Method": "aCompCor",
    "Retained": true,
    "SingularValue": 25.8270209974,
    "VarianceExplained": 0.1081970825
  "dropped_0": {
    "CumulativeVarianceExplained": 0.5965809597,
    "Mask": "combined",
    "Method": "aCompCor",
    "Retained": false,
    "SingularValue": 20.7955177198,
    "VarianceExplained": 0.0701465624

For CompCor decompositions, entries include:

  • Method: anatomical or temporal CompCor.

  • Mask: denotes the ROI where the decomposition that generated the component was performed: CSF, WM, or combined for anatomical CompCor.

  • SingularValue: singular value of the component.

  • VarianceExplained: the fraction of variance explained by the component across the decomposition ROI mask.

  • CumulativeVarianceExplained: the total fraction of variance explained by this particular component and all preceding components.

  • Retained: Indicates whether the component was saved in desc-confounds_regressors.tsv for use in denoising. Entries that are not saved in the data file for denoising are still stored in metadata with the dropped prefix.

Confounds and “carpet”-plot on the visual reports

Some of the estimated confounds, as well as a “carpet” visualization of the BOLD time series [Power2016]. This plot is included for each run within the corresponding visual report. An example of these plots follows:


The figure shows on top several confounds estimated for the BOLD series: global signals (‘GlobalSignal’, ‘WM’, ‘GM’), standardized DVARS (‘stdDVARS’), and framewise-displacement (‘FramewiseDisplacement’). At the bottom, a ‘carpetplot’ summarizing the BOLD series. The colormap on the left-side of the carpetplot denotes signals located in cortical gray matter regions (blue), subcortical gray matter (orange), cerebellum (green) and the union of white-matter and CSF compartments (red).

Noise components computed during each CompCor decomposition are evaluated according to the fraction of variance that they explain across the nuisance ROI. This is used by fMRIPrep to determine whether each component should be saved for use in denoising operations: a component is saved if it contributes to explaining the top 50 percent of variance in the nuisance ROI. fMRIPrep can be configured to save all components instead using the command line option --return-all-components. fMRIPrep reports include a plot of the cumulative variance explained by each component, ordered by descending singular value.


The figure displays the cumulative variance explained by components for each of four CompCor decompositions (left to right: anatomical CSF mask, anatomical white matter mask, anatomical combined mask, temporal). The number of components is plotted on the abscissa and the cumulative variance explained on the ordinate. Dotted lines indicate the minimum number of components necessary to explain 50%, 70%, and 90% of the variance in the nuisance mask. By default, only the components that explain the top 50% of the variance are saved.

Also included is a plot of correlations among confound regressors. This can be used to guide selection of a confound model or to assess the extent to which tissue-specific regressors correlate with global signal.


The left-hand panel shows the matrix of correlations among selected confound time series as a heatmap. Note the zero-correlation blocks near the diagonal; these correspond to each CompCor decomposition. The right-hand panel displays the correlation of selected confound time series with the mean global signal computed across the whole brain; the regressors shown are those with greatest correlation with the global signal. This information can be used to diagnose partial volume effects.

See implementation on init_bold_confs_wf.



Behzadi Y, Restom K, Liau J, Liu TT, A component-based noise correction method (CompCor) for BOLD and perfusion-based fMRI. NeuroImage. 2007. doi: 10.1016/j.neuroimage.2007.04.042


Ciric R, Wolf DH, Power JD, Roalf DR, Baum GL, Ruparel K, Shinohara RT, Elliott MA, Eickhoff SB, Davatzikos C., Gur RC, Gur RE, Bassett DS, Satterthwaite TD. Benchmarking of participant-level confound regression strategies for the control of motion artifact in studies of functional connectivity. Neuroimage. 2017. doi: 10.1016/j.neuroimage.2017.03.020


Greve DN, Brown GG, Mueller BA, Glover G, Liu TT, A Survey of the Sources of Noise in fMRI Psychometrika. 2013. doi: 10.1007/s11336-013-9344-2


Friston KJ1, Williams S, Howard R, Frackowiak RS, Turner R, Movement‐Related effects in fMRI time‐series. Magnetic Resonance in Medicine. 1996. doi: 10.1002/mrm.191035031


Muschelli J, Nebel MB, Caffo BS, Barber AD, Pekar JJ, Mostofsky SH, Reduction of motion-related artifacts in resting state fMRI using aCompCor. NeuroImage. 2014. doi: 10.1016/j.neuroimage.2014.03.028


Parkes L, Fulcher B, Yücel M, Fornito A, An evaluation of the efficacy, reliability, and sensitivity of motion correction strategies for resting-state functional MRI. NeuroImage. 2018. doi: 10.1016/j.neuroimage.2017.12.073


Power JD, Barnes KA, Snyder AZ, Schlaggar BL, Petersen, SA, Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion. NeuroImage. 2012. doi: 10.1016/j.neuroimage.2011.10.018


Power JD, A simple but useful way to assess fMRI scan qualities. NeuroImage. 2016. doi: 10.1016/j.neuroimage.2016.08.009


Satterthwaite TD, Elliott MA, Gerraty RT, Ruparel K, Loughead J, Calkins ME, Eickhoff SB, Hakonarson H, Gur RC, Gur RE, Wolf DH, An improved framework for confound regression and filtering for control of motion artifact in the preprocessing of resting-state functional connectivity data. NeuroImage. 2013. doi: 10.1016/j.neuroimage.2012.08.052


Yan CG, Cheung B, Kelly C, Colcombe S, Craddock RC, Di Martino A, Li Q, Zuo XN, Castellanos FX, Milham MP, A comprehensive assessment of regional variation in the impact of head micromovements on functional connectomics. NeuroImage. 2013. doi: 10.1016/j.neuroimage.2013.03.004