Cortical surface-based searchlight decoding#

This is a demo for surface-based searchlight decoding, as described in Chen et al.[1].

Load Haxby dataset#

import pandas as pd

from nilearn import datasets

# We fetch 2nd subject from haxby datasets (which is default)
haxby_dataset = datasets.fetch_haxby()

fmri_filename = haxby_dataset.func[0]
labels = pd.read_csv(haxby_dataset.session_target[0], sep=" ")
y = labels["labels"]
run = labels["chunks"]

Restrict to faces and houses#

Surface BOLD response#

from sklearn import neighbors

from nilearn import datasets, surface

# Fetch a coarse surface of the left hemisphere only for speed
fsaverage = datasets.fetch_surf_fsaverage(mesh="fsaverage5")
hemi = "left"

# Average voxels 5 mm close to the 3d pial surface
radius = 5.0
pial_mesh = fsaverage[f"pial_{hemi}"]
X = surface.vol_to_surf(fmri_img, pial_mesh, radius=radius).T

# To define the :term:`BOLD` responses
# to be included within each searchlight "sphere"
# we define an adjacency matrix based on the inflated surface vertices such
# that nearby surfaces are concatenated within the same searchlight.

infl_mesh = fsaverage[f"infl_{hemi}"]
coords, _ = surface.load_surf_mesh(infl_mesh)
radius = 3.0
nn = neighbors.NearestNeighbors(radius=radius)
adjacency =

Searchlight computation#

from sklearn.linear_model import RidgeClassifier
from sklearn.model_selection import KFold
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler

from nilearn.decoding.searchlight import search_light

# Simple linear estimator preceded by a normalization step
estimator = make_pipeline(StandardScaler(), RidgeClassifier(alpha=10.0))

# Define cross-validation scheme
cv = KFold(n_splits=3, shuffle=False)

# Cross-validated search light
scores = search_light(X, y, estimator, adjacency, cv=cv, n_jobs=2)


from nilearn import plotting

chance = 0.5
    scores - chance,
    title="Accuracy map, left hemisphere",
Accuracy map, left hemisphere


Total running time of the script: (1 minutes 11.876 seconds)

Estimated memory usage: 913 MB

Gallery generated by Sphinx-Gallery