Overview

fMRIPrep

Dependency	Purpose
FSL (FLIRT, FNIRT, MCFLIRT, SUSAN, FAST)	Motion correction, skull stripping, bias field correction, nonlinear warping
ANTs (Advanced Normalization Tools)	Spatial normalization (T1→MNI), bias field correction
FreeSurfer	Cortical surface reconstruction (optional, but recommended)
AFNI (3dTshift, 3dSkullStrip)	Slice-timing correction, alternative skull stripping
Workflows from Nilearn & Nipype	Pipeline orchestration and image processing
NumPy, SciPy, NiBabel, pandas	Data processing
Matplotlib, seaborn	Quality control (QC) reporting
TemplateFlow	Standardized brain templates

FSL::feat

Outputs :
- First-Level Analysis ( Single-Subject )
  - GLM Parameter Estimates ( cope1.nii.gz, varcope1.nii.gz )
  - T-Statistics & Z-Statistics ( tstat1.nii.gz, zstat1.nii.gz )
  - Design Matrix ( design.mat , design.con , design.fsf )
  - Motion Outliers Report ( motion_outliers.txt )
  - Contrast Maps ( e.g., zstat1.nii.gz for statistical significance )
  - Registration Outputs ( e.g., example_func2highres.mat for aligning functional to anatomical )
- Higher-Level Analysis ( Group-Level )
  - Group statistics ( mixed-effects modeling )
  - Cluster-corrected results
  - Thresholded statistical maps

fMRIPrep Outputs Needed for Nilearn

fMRIPrep Output	Used for
`sub-_task-_bold_space-MNI152NLin2009cAsym_preproc.nii.gz`	Preprocessed functional images
`sub-*_desc-brain_mask.nii.gz`	Brain mask for extracting signal
`sub-*_desc-confounds_timeseries.tsv`	Motion parameters, CompCor regressors
`sub-*_space-MNI152NLin2009cAsym_res-2_desc-preproc_bold.nii.gz`	Smoothed images (if needed)
`sub-*_from-T1w_to-MNI152_desc-linear_xfm.h5`	Transformation matrices (for custom ROI masking)

Nilearn Equivalent of FEAT Steps

FEAT Step	Nilearn Alternative
Motion Correction	Already handled by fMRIPrep
Spatial Normalization	Already handled by fMRIPrep
Confound Regression	`nilearn.image.clean_img()`
GLM First-Level	`nilearn.glm.first_level.FirstLevelModel()`
Statistical Maps (t-stats, z-stats)	`.compute_contrast()` in `FirstLevelModel`
ROI Extraction	`nilearn.input_data.NiftiMasker`
MVPA/Decoding	`nilearn.decoding.SearchLight`

Per Person

Option 1 : fMRIPrep ➡️ FSL::feat
1. Run fMRIPrep: Prepares cleaned functional images with confound regressors.
2. Import into FEAT: Use *_preproc_bold.nii.gz as input for FEAT.
3. Apply Confound Regressors: Use desc-confounds_timeseries.tsv in FEAT model.
4. Run First-Level Analysis: Apply GLM to analyze task-based activity.
5. Higher-Level Analysis: Combine across subjects.
Option 2 : fMRIPrep ➡️ nilearn
1. Run fMRIPrep: Prepares standardized, preprocessed fMRI data.
2. Load in Nilearn: Use nibabel and nilearn.image to load preprocessed outputs.
3. Apply Masking & Confound Removal: Use nilearn.input_data.NiftiMasker and regress out motion, CompCor signals, etc.
4. Perform Statistical Modeling: Use nilearn.glm.first_level.FirstLevelModel instead of FSL FEAT.
5. Extract ROIs & Perform Decoding: Use nilearn.decoding for MVPA, or nilearn.connectome for network analyses.
Option 3 : FSL::feat ➡️ nilearn
1. Run FEAT (First-Level Analysis): Process individual subject-level fMRI data.
2. Extract Statistical Maps: Use stats/cope1.nii.gz, stats/zstat1.nii.gz, etc.
3. Load in Nilearn: Use nibabel and nilearn.image.load_img() to read statistical maps.
4. Perform Further Analysis: Apply machine learning, MVPA, or advanced modeling using nilearn.decoding or nilearn.glm.
5. Visualization & Interpretation: Use nilearn.plotting for displaying activation maps.
Option 4 : FSL::feat ➡️ FSL:feat
1. Run FEAT (First-Level Analysis): Process individual subject fMRI data.
2. Extract First-Level Outputs: Obtain contrast maps (cope.nii.gz), variance maps (varcope.nii.gz), and statistical results.
3. Register to Standard Space: Use example_func2standard.mat to align functional data.
4. Set Up Group-Level Model: Create a design.mat and design.con file for group-level GLM analysis.
5. Run Higher-Level FEAT Analysis: Apply mixed-effects modeling to combine subject data.
6. Correct for Multiple Comparisons: Use cluster-based thresholding or TFCE correction.
7. Generate Final Statistical Maps: Export thresh_zstat.nii.gz for interpretation.

Group Level

FSL::feat Group-Level

Collect First-Level Outputs: Gather cope.nii.gz and varcope.nii.gz files from each subject.
Prepare a Group Design Matrix: Create a design.mat and design.con for higher-level analysis.
Run Higher-Level FEAT: Perform mixed-effects analysis to model group differences.
Apply Statistical Thresholding: Use FSL's thresholding options (cluster-based correction, TFCE, etc.).
Interpret Results: Review thresh_zstat.nii.gz for significant group-level activations.

Nilearn Group-Level

Load Individual Subject Maps: Use nilearn.image.load_img() to import first-level statistical maps.
Apply Standardization: Use nilearn.image.mean_img() or nilearn.image.math_img() for normalization.
Perform Group-Level GLM: Use nilearn.glm.second_level.SecondLevelModel() for a random-effects model.
Define Contrasts: Specify conditions or groups for comparison.
Run Statistical Analysis: Compute group contrasts and extract statistical maps.
Multiple Comparison Correction: Use nilearn.stats.threshold_stats_img() for FDR correction.
Visualize Results: Use nilearn.plotting.plot_stat_map() to render group-level activation.