6  Multiple comparison correction

In whole-brain analysis

Reminder: group analysis

Family-wise error rate

Image credit: unknown

If the data is not smoothed, equivalent to bonferroni

Other correction types

Cluster threshold

False discovery rate

Cluster-level inference

Uncorrected

# interactive plot (you can browse the activations)
from nilearn import plotting

# Use subject's anatomy as background
bg_img = '/home/jovyan/gambling/bids/derivatives/fmriprep/sub-001/ses-1/anat/sub-001_ses-1_acq-mprage_desc-preproc_T1w.nii.gz'
plotting.view_img(zmap, threshold=1.96, vmax=10, 
    bg_img=bg_img,
    cut_coords=[0, 0, 0],
    width_view=600,
    title=contrast_string)

Cluster-thresholded

from nilearn.glm import threshold_stats_img

thresholded_map1, threshold1 = threshold_stats_img(
    zmap,
    alpha=0.05,
    height_control="fpr",
    cluster_threshold = 100,
    two_sided=True,
)

# Use subject's anatomy as background
bg_img = '/home/jovyan/gambling/bids/derivatives/fmriprep/sub-001/ses-1/anat/sub-001_ses-1_acq-mprage_desc-preproc_T1w.nii.gz'
plotting.view_img(thresholded_map1, threshold=threshold1, vmax=10, 
    bg_img=bg_img,
    cut_coords=[0, 0, 0],
    width_view=600,
    title=contrast_string)

FDR-thresholded

from nilearn.glm import threshold_stats_img

thresholded_map1, threshold1 = threshold_stats_img(
    zmap,
    alpha=0.05,
    height_control="fdr",
    two_sided=True,
)

# Use subject's anatomy as background
bg_img = '/home/jovyan/gambling/bids/derivatives/fmriprep/sub-001/ses-1/anat/sub-001_ses-1_acq-mprage_desc-preproc_T1w.nii.gz'
plotting.view_img(thresholded_map1, threshold=threshold1, vmax=10, 
    bg_img=bg_img,
    cut_coords=[0, 0, 0],
    width_view=600,
    title=contrast_string)

FWE-thresholded

from nilearn.glm import threshold_stats_img

thresholded_map1, threshold1 = threshold_stats_img(
    zmap,
    alpha=0.05,
    height_control="bonferroni",
    two_sided=True,
)

# Use subject's anatomy as background
bg_img = '/home/jovyan/gambling/bids/derivatives/fmriprep/sub-001/ses-1/anat/sub-001_ses-1_acq-mprage_desc-preproc_T1w.nii.gz'
plotting.view_img(thresholded_map1, threshold=threshold1, vmax=10, 
    bg_img=bg_img,
    cut_coords=[0, 0, 0],
    width_view=600,
    title=contrast_string)

Non-parametric inference

Permutation-based inference

(Nichols and Holmes 2001)

For the multiple comparison case, we use the same logic, but instead of single voxel we consider the maximum of each statistical map.

References

Nichols, Thomas E., and Andrew P. Holmes. 2001. “Nonparametric Permutation Tests for Functional Neuroimaging: A Primer with Examples.” Human Brain Mapping 15 (1): 1–25. https://doi.org/10.1002/hbm.1058.