5.5.1. Masks

Masks are essentially boolean arrays that are used to constrain the extraction of features to voxels that are meaningful. For example, in an fMRI imaging study, a mask can be used to constrain the extraction of features to voxels that contain a certain ratio of gray matter to white matter / cerebrospinal fluid, ensuring that the features are not extracted from voxels that contain mostly white matter or cerebrospinal fluid, which could add noise to the BOLD signal.

Junifer provides a number of built-in masks, which can be listed using the list_masks(). Some masks are images, while other masks can be computed using nilearn functions.

For markers and steps that accept masks as an argument, the mask can be specified as a string, which will be the name of a built-in mask, or as a dictionary in which the only key is the built-in mask name and the value is a dictionary of keyword arguments to pass to the mask function.

For example, the following is a valid mask specification that specified the GM_prob0.2 mask.

masks: GM_prob0.2

The following is a valid mask specification that specifies the compute_brain_mask mask (function from nilearn), with a threshold of 0.5.

masks:
    compute_brain_mask:
        threshold: 0.5

Furthermore, junifer allows you to combine several masks using nilearn.masking.intersect_masks(). This is done by specifying a list of masks, where each mask is a string or dictionary as described above. For example, the following is a valid mask specification that specifies the intersection of the GM_prob0.2 and compute_brain_mask masks.

masks:
    - GM_prob0.2
    - compute_brain_mask:
        threshold: 0.5

We can also specify the arguments of nilearn.masking.intersect_masks() (threshold and connected). The following example combines the same masks as the previous one, but computing the full intersection.

masks:
    - GM_prob0.2
    - compute_brain_mask:
        threshold: 0.5
    - threshold: 1  # intersection

Alternatively, we can also compute the union, even if the voxels do not form a connected component:

masks:
    - GM_prob0.2
    - compute_brain_mask:
        threshold: 0.5
    - threshold: 0  # union
    - connected: False  # keep disconnected components