Massively univariate analysis of a calculation task from the Localizer dataset

This example shows how to use the Localizer dataset in a basic analysis. A standard Anova is performed (massively univariate F-test) and the resulting Bonferroni-corrected p-values are plotted. We use a calculation task and 20 subjects out of the 94 available.

The Localizer dataset contains many contrasts and subject-related variates. The user can refer to the plot_localizer_mass_univariate_methods.py example to see how to use these.

Note

If you are using Nilearn with a version older than 0.9.0, then you should either upgrade your version or import maskers from the input_data module instead of the maskers module.

That is, you should manually replace in the following example all occurrences of:

from nilearn.maskers import NiftiMasker

with:

from nilearn.input_data import NiftiMasker

from nilearn._utils.helpers import check_matplotlib

check_matplotlib()

import matplotlib.pyplot as plt

import numpy as np

from nilearn import datasets
from nilearn.image import get_data
from nilearn.maskers import NiftiMasker

Load Localizer contrast

n_samples = 20
localizer_dataset = datasets.fetch_localizer_calculation_task(
    n_subjects=n_samples
)
tested_var = np.ones((n_samples, 1))

Mask data

nifti_masker = NiftiMasker(
    smoothing_fwhm=5, memory="nilearn_cache", memory_level=1
)
cmap_filenames = localizer_dataset.cmaps
fmri_masked = nifti_masker.fit_transform(cmap_filenames)

Anova (parametric F-scores)

from sklearn.feature_selection import f_regression

# Center=False is used to not remove intercept
_, pvals_anova = f_regression(fmri_masked, tested_var, center=False)
pvals_anova *= fmri_masked.shape[1]
pvals_anova[np.isnan(pvals_anova)] = 1
pvals_anova[pvals_anova > 1] = 1
neg_log_pvals_anova = -np.log10(pvals_anova)
neg_log_pvals_anova_unmasked = nifti_masker.inverse_transform(
    neg_log_pvals_anova
)

Visualization

from nilearn.plotting import plot_stat_map, show

# Various plotting parameters
plotted_slice = 45
threshold = -np.log10(0.1)  # 10% corrected

masked_pvals = np.ma.masked_less(
    get_data(neg_log_pvals_anova_unmasked), threshold
)

title = (
    "Negative $\\log_{10}$ p-values"
    "\n(Parametric + Bonferroni correction)"
    f"\n{(~masked_pvals.mask).sum()} detections"
)

# Plot Anova p-values
display = plot_stat_map(
    neg_log_pvals_anova_unmasked,
    threshold=threshold,
    display_mode="z",
    cut_coords=[plotted_slice],
    figure=plt.figure(figsize=(5, 6), facecolor="w"),
    cmap="inferno",
    vmin=threshold,
    title=title,
)


show()

Estimated memory usage: 0 MB