nilearn · Remi-Gau · Oct 13, 2023 · Oct 7, 2023 · Oct 7, 2023 · Oct 7, 2023
diff --git a/doc/building_blocks/manual_pipeline.rst b/doc/building_blocks/manual_pipeline.rst
@@ -61,7 +61,7 @@ Loading non image data: experiment description
 -----------------------------------------------
 
 An experiment may need additional information about subjects, sessions or
-experiments. In the Haxby experiment, fMRI data are acquired while
+experiments. In the Haxby experiment, :term:`fMRI` data are acquired while
 presenting different category of pictures to the subject (face, cat, ...)
 and the goal of this experiment is to predict which category is presented
 to the subjects from the brain activation.
@@ -216,7 +216,7 @@ Here we want to see the discriminating weights of some voxels.
 Visualizing results
 ===================
 
-Again the visualization code is simple. We can use an fMRI slice as a
+Again the visualization code is simple. We can use an :term:`fMRI` slice as a
 background and plot the weights. Brighter points have a higher
 discriminating weight.
 

diff --git a/doc/building_blocks/neurovault.rst b/doc/building_blocks/neurovault.rst
@@ -77,7 +77,7 @@ The default values for the ``collection_terms`` and ``image_terms`` parameters
 filter out empty collections, and exclude an image if one of the following is
 true:
 
-   - it is not in MNI space.
+   - it is not in :term:`MNI` space.
    - its metadata field "is_valid" is cleared.
    - it is thresholded.
    - its map type is one of "ROI/mask", "anatomical", or "parcellation".
@@ -146,7 +146,7 @@ Using a filter rather than a dictionary, the first example becomes:
   Even if you specify a filter as a function, the default filters for
   ``image_terms`` and ``collection_terms`` still apply; pass an empty
   dictionary if you want to disable them. Without ``image_terms={}`` in the
-  call above, parcellations, images not in MNI space, etc. would be still be
+  call above, parcellations, images not in :term:`MNI` space, etc. would be still be
   filtered out.
 
 
@@ -186,7 +186,7 @@ Neurosynth annotations
 
 It is also possible to ask Neurosynth to annotate the maps found on
 Neurovault. Neurosynth is a platform for large-scale, automated
-synthesis of fMRI data. It can be used to perform decoding.  You can
+synthesis of :term:`fMRI` data. It can be used to perform decoding.  You can
 learn more about Neurosynth at http://www.neurosynth.org.
 
 Neurosynth was introduced in [2]_.

diff --git a/doc/connectivity/connectome_extraction.rst b/doc/connectivity/connectome_extraction.rst
@@ -26,7 +26,7 @@ Sparse inverse covariance for functional connectomes
 
 Functional connectivity can be obtained by estimating a covariance
 (or correlation) matrix for signals from different brain
-regions decomposed, for example on resting-state or naturalistic-stimuli datasets.
+regions decomposed, for example on :term:`resting-state` or naturalistic-stimuli datasets.
 The same information can be represented as a weighted graph,
 vertices being brain regions, weights on edges being covariances
 (gaussian graphical model). However, coefficients in a covariance matrix

diff --git a/doc/connectivity/functional_connectomes.rst b/doc/connectivity/functional_connectomes.rst
@@ -143,7 +143,7 @@ Probabilistic atlases
 The definition of regions as by a continuous probability map captures
 better our imperfect knowledge of boundaries in brain images (notably
 because of inter-subject registration errors). One example of such an
-atlas well suited to resting-state or naturalistic-stimuli data analysis is
+atlas well suited to :term:`resting-state` or naturalistic-stimuli data analysis is
 the `MSDL atlas
 <https://team.inria.fr/parietal/18-2/spatial_patterns/spatial-patterns-in-resting-state/>`_
 (:func:`nilearn.datasets.fetch_atlas_msdl`).
@@ -214,7 +214,7 @@ the edges capture interactions between them, this graph is a "functional
 connectome".
 
 We can display it with the :func:`nilearn.plotting.plot_connectome`
-function that take the matrix, and coordinates of the nodes in MNI space.
+function that take the matrix, and coordinates of the nodes in :term:`MNI` space.
 In the case of the MSDL atlas
 (:func:`nilearn.datasets.fetch_atlas_msdl`), the CSV file readily comes
 with :term:`MNI` coordinates for each region (see for instance example:

diff --git a/doc/connectivity/parcellating.rst b/doc/connectivity/parcellating.rst
@@ -24,9 +24,9 @@ Data loading: movie-watching data
 
 .. currentmodule:: nilearn.datasets
 
-Clustering is commonly applied to resting-state data, but any brain
+Clustering is commonly applied to :term:`resting-state` data, but any brain
 functional data will give rise of a functional parcellation, capturing
-intrinsic brain architecture in the case of resting-state data.
+intrinsic brain architecture in the case of :term:`resting-state` data.
 In the examples, we use naturalistic stimuli-based movie watching
 brain development data downloaded with the function
 :func:`fetch_development_fmri` (see :ref:`loading_data`).
@@ -79,7 +79,7 @@ Ward's algorithm is a hierarchical clustering algorithm: it
 recursively merges voxels, then clusters that have similar signal
 (parameters, measurements or time courses).
 
-**Caching** In practice the implementation of Ward clustering first
+**Caching** In practice the implementation of :term:`Ward clustering` first
 computes a tree of possible merges, and then, given a requested number of
 clusters, breaks apart the tree at the right level.
 
@@ -92,7 +92,7 @@ used for caching.
 
 .. note::
 
-    The Ward clustering computing 1000 parcels runs typically in about 10
+    The :term:`Ward clustering` computing 1000 parcels runs typically in about 10
     seconds. Admittedly, this is very fast.
 
 .. note::
@@ -105,7 +105,7 @@ used for caching.
 
    * A function :func:`nilearn.regions.connected_label_regions` which can be useful to
      break down connected components into regions. For instance, clusters defined using
-     KMeans whereas it is not necessary for Ward clustering due to its
+     KMeans whereas it is not necessary for :term:`Ward clustering` due to its
      spatial connectivity.
 
 
@@ -117,7 +117,7 @@ Using and visualizing the resulting parcellation
 Visualizing the parcellation
 -----------------------------
 
-The labels of the parcellation are found in the ``labels_img_`` attribute of
+The labels of the :term:`parcellation` are found in the ``labels_img_`` attribute of
 the :class:`nilearn.regions.Parcellations` object after fitting it to the data
 using *ward.fit*. We directly use the result for visualization.
 

diff --git a/doc/connectivity/region_extraction.rst b/doc/connectivity/region_extraction.rst
@@ -22,7 +22,7 @@ Region Extraction for better brain parcellations
 Fetching movie-watching based functional datasets
 =================================================
 
-We use a naturalistic stimuli based movie-watching functional connectivity dataset
+We use a naturalistic stimuli based movie-watching :term:`functional connectivity` dataset
 of 20 subjects, which is already preprocessed, downsampled to 4mm isotropic resolution, and publicly available at
 `<https://osf.io/5hju4/files/>`_. We use utilities
 :func:`fetch_development_fmri` implemented in nilearn for automatic fetching of this

diff --git a/doc/decoding/searchlight.rst b/doc/decoding/searchlight.rst
@@ -211,7 +211,7 @@ equivalent to a permuted F-test by setting the argument
 ``two_sided_test`` to ``True``. In the example above, we do perform a two-sided
 test but add back the sign of the effect at the end using the t-scores obtained
 on the original (non-permuted) data. Thus, we can perform two one-sided tests
-(a given contrast and its opposite) for the price of one single run.
+(a given :term:`contrast` and its opposite) for the price of one single run.
 The example results can be interpreted as follows: viewing faces significantly
 activates the Fusiform Face Area as compared to viewing houses, while viewing
 houses does not reveal significant supplementary activations as compared to
@@ -224,7 +224,7 @@ viewing faces.
     assuming that nothing happens (i.e. under the null hypothesis).
     Therefore, a small *p-value* indicates that there is a small chance
     of getting this data if no real difference existed, so the observed
-    voxel must be significant.
+    :term:`voxel` must be significant.
 
 .. [2]
 

diff --git a/doc/decoding/space_net.rst b/doc/decoding/space_net.rst
@@ -47,7 +47,7 @@ Related example
 
     Empirical comparisons using this method have been removed from
     documentation in version 0.7 to keep its computational cost low. You can
-    easily try SpaceNet instead of FREM in :ref:`mixed gambles study <sphx_glr_auto_examples_02_decoding_plot_mixed_gambles_frem.py>` or :ref:`Haxby study <sphx_glr_auto_examples_02_decoding_plot_haxby_frem.py>`.
+    easily try SpaceNet instead of :term:`FREM` in :ref:`mixed gambles study <sphx_glr_auto_examples_02_decoding_plot_mixed_gambles_frem.py>` or :ref:`Haxby study <sphx_glr_auto_examples_02_decoding_plot_haxby_frem.py>`.
 
 .. seealso::
 

diff --git a/doc/developers/group_sparse_covariance.rst b/doc/developers/group_sparse_covariance.rst
@@ -255,7 +255,7 @@ stopped.
 This technique it is only a way to stop iterating based on the
 estimate value instead of the criterion value. It does *not* ensure a
 given uncertainty on the estimate. This has been tested on synthetic
-and real fMRI data: using two different starting points leads to two
+and real :term:`fMRI` data: using two different starting points leads to two
 estimates that can differ (in max norm) by more than the threshold
 (see next paragraph). However, it has the same property as the duality
 gap criterion: quickly converging cases use fewer iterations than

diff --git a/doc/glm/first_level_model.rst b/doc/glm/first_level_model.rst
@@ -8,21 +8,21 @@ First level models
 
   First level models are, in essence, linear regression models run at the level of a single
   session or single subject. The model is applied on a voxel-wise basis, either on the whole
-  brain or within a region of interest. The timecourse of each voxel is regressed against a
-  predicted BOLD response created by convolving the haemodynamic response function (HRF) with
+  brain or within a region of interest. The timecourse of each :term:`voxel` is regressed against a
+  predicted :term:`BOLD` response created by convolving the haemodynamic response function (HRF) with
   a set of predictors defined within the design matrix.
 
 
 HRF models
 ==========
 
-Nilearn offers a few different HRF models including the commonly used double-gamma SPM model ('spm')
+Nilearn offers a few different :term:`HRF` models including the commonly used double-gamma :term:`SPM` model ('spm')
 and the model shape proposed by G. Glover ('glover'), both allowing the option of adding time and
 dispersion derivatives. The addition of these derivatives allows to better model any uncertainty in
-timing information. In addition, an FIR (finite impulse response, 'fir') model of the HRF is also available.
+timing information. In addition, an :term:`FIR` (finite impulse response, 'fir') model of the :term:`HRF` is also available.
 
 In order to visualize the predicted regressor prior to plugging it into the linear model, use the
-function :func:`nilearn.glm.first_level.compute_regressor`, or explore the HRF plotting
+function :func:`nilearn.glm.first_level.compute_regressor`, or explore the :term:`HRF` plotting
 example :ref:`sphx_glr_auto_examples_04_glm_first_level_plot_hrf.py`.
 
 
@@ -88,7 +88,7 @@ Fitting a first level model
 ===========================
 
 The :class:`nilearn.glm.first_level.FirstLevelModel` class provides the tools to fit the linear model to
-the fMRI data. The :func:`nilearn.glm.first_level.FirstLevelModel.fit()` function takes the fMRI data
+the :term:`fMRI` data. The :func:`nilearn.glm.first_level.FirstLevelModel.fit()` function takes the fMRI data
 and design matrix as input and fits the GLM. Like other Nilearn functions,
 :func:`nilearn.glm.first_level.FirstLevelModel.fit()` accepts file names as input, but can also
 work with :nipy:`NiftiImage objects <nibabel/nibabel_images.html>`. More information about
@@ -102,7 +102,7 @@ input formats is available :ref:`here <loading_data>` ::
 Computing contrasts
 -------------------
 
-To get more interesting results out of the GLM model, contrasts can be computed between regressors of interest.
+To get more interesting results out of the :term:`GLM` model, contrasts can be computed between regressors of interest.
 The :func:`nilearn.glm.first_level.FirstLevelModel.compute_contrast` function can be used for that. First,
 the contrasts of interest must be defined. In the spm_multimodal_fmri dataset referenced above, subjects are
 presented with 'normal' and 'scrambled' faces. The basic contrasts that can be constructed are the main effects