Breaking an atlas of labels in separated regions#

This example shows how to use nilearn.regions.connected_label_regions to assign each spatially-separated region of the atlas a unique label.

Indeed, often in a given atlas of labels, the same label (number) may be used in different connected regions, for instance a region in each hemisphere. If we want to operate on regions and not networks (for instance in signal extraction), it is useful to assign a different label to each region. We end up with a new atlas that has more labels, but each one points to a single region.

We use the Yeo atlas as an example for labeling regions, nilearn.datasets.fetch_atlas_yeo_2011

The original Yeo atlas#

# First we fetch the Yeo atlas
from nilearn import datasets

atlas_yeo_2011 = datasets.fetch_atlas_yeo_2011()
atlas_yeo = atlas_yeo_2011.thick_7

# Let's now plot it
from nilearn import plotting

plotting.plot_roi(
    atlas_yeo,
    title="Original Yeo atlas",
    cut_coords=(8, -4, 9),
    colorbar=True,
    cmap="Paired",
)
plot extract regions labels image
<nilearn.plotting.displays._slicers.OrthoSlicer object at 0x7fb385cc1910>

The original Yeo atlas has 7 labels, that is indicated in the colorbar. The colorbar also shows the correspondence between the color and the label

Note that these 7 labels correspond actually to networks that comprise several regions. We are going to split them up.

Relabeling the atlas into separated regions#

Now we use the connected_label_regions to break apart the networks of the Yeo atlas into separated regions

/home/remi/github/nilearn/nilearn/env/lib/python3.11/site-packages/nilearn/regions/region_extractor.py:607: UserWarning:

Data array used to create a new image contains 64-bit ints. This is likely due to creating the array with numpy and passing `int` as the `dtype`. Many tools such as FSL and SPM cannot deal with int64 in Nifti images, so for compatibility the data has been converted to int32.

Plotting the new regions

plotting.plot_roi(
    region_labels,
    title="Relabeled Yeo atlas",
    cut_coords=(8, -4, 9),
    colorbar=True,
    cmap="Paired",
)
plot extract regions labels image
<nilearn.plotting.displays._slicers.OrthoSlicer object at 0x7fb3cb2ba950>

Note that the same cluster in original and labeled atlas could have different color, so, you cannot directly compare colors.

However, you can see that the regions in the left and right hemispheres now have different colors. For some regions it is difficult to tell apart visually, as the colors are too close on the colormap (eg in the blue: regions labeled around 3).

Also, we can see that there are many more labels: the colorbar goes up to 49. The 7 networks of the Yeo atlas are now broken up into 49 ROIs.

You can save the new atlas to a nifti file using to_filename method.

from pathlib import Path

output_dir = Path.cwd() / "results" / "plot_extract_regions_labels_image"
output_dir.mkdir(exist_ok=True, parents=True)
print(f"Output will be saved to: {output_dir}")

region_labels.to_filename(output_dir / "relabeled_yeo_atlas.nii.gz")
Output will be saved to: /home/remi/github/nilearn/nilearn/examples/06_manipulating_images/results/plot_extract_regions_labels_image

Different connectivity modes#

Using the parameter connect_diag=False we separate in addition two regions that are connected only along the diagonal.

region_labels_not_diag = connected_label_regions(atlas_yeo, connect_diag=False)

plotting.plot_roi(
    region_labels_not_diag,
    title="Relabeling and connect_diag=False",
    cut_coords=(8, -4, 9),
    colorbar=True,
    cmap="Paired",
)
plot extract regions labels image
<nilearn.plotting.displays._slicers.OrthoSlicer object at 0x7fb37ed71150>

A consequence of using connect_diag=False is that we can get a lot of small regions, around 110 judging from the colorbar.

Hence we suggest use connect_diag=True

Parameter min_size#

In the above, we get around 110 regions, but many of these are very small. We can remove them with the min_size parameter, keeping only the regions larger than 100mm^3.

region_labels_min_size = connected_label_regions(
    atlas_yeo, min_size=100, connect_diag=False
)

plotting.plot_roi(
    region_labels_min_size,
    title="Relabeling and min_size",
    cut_coords=(8, -4, 9),
    colorbar=True,
    cmap="Paired",
)

plotting.show()
plot extract regions labels image

Total running time of the script: (0 minutes 31.301 seconds)

Estimated memory usage: 780 MB

Gallery generated by Sphinx-Gallery