[FIX] change creation of region_names and add test

nilearn/maskers/nifti_labels_masker.py

@@ -503,6 +503,39 @@ def fit(self, imgs=None, y=None):
         msg = f"loading data from {repr}"
         _utils.logger.log(msg=msg, verbose=self.verbose)
         self.labels_img_ = _utils.check_niimg_3d(self.labels_img)
+
+        # create region_id_name_ dictionary
+        self.region_id_name_ = None
+        if self.labels is not None:
+            known_backgrounds = {"background", "Background"}
+            initial_region_ids = [
+                region_id
+                for region_id in np.unique(
+                    _utils.niimg.safe_get_data(self.labels_img_)
+                )
+                if region_id != self.background_label
+            ]
+            initial_region_names = [
+                region_name
+                for region_name in self.labels
+                if region_name not in known_backgrounds
+            ]
+
+            if len(initial_region_ids) != len(initial_region_names):
+                warnings.warn(
+                    "Number of regions in the labels image "
+                    "does not match the number of labels provided.",
+                    stacklevel=3,
+                )
+            # if number of regions in the labels image is more
+            # than the number of labels provided, then we cannot
+            # create region_id_name_ dictionary
+            if len(initial_region_ids) <= len(initial_region_names):
+                self.region_id_name_ = {
+                    region_id: initial_region_names[i]
+                    for i, region_id in enumerate(initial_region_ids)
+                }
+
         if self.mask_img is not None:
             repr = _utils._repr_niimgs(
                 self.mask_img, shorten=(not self.verbose)
@@ -761,20 +794,10 @@ def transform_single_imgs(self, imgs, confounds=None, sample_mask=None):
             self.labels_, tolerant=True, resampling_done=True
         )
 
-        if self.labels is not None:
-
-            # Keep track if background was explicitly passed as a label
-            # background should always be explicitly passed in the labels
-            # to avoid this.
-            lower_case_labels = {x.lower() for x in self.labels}
-            known_backgrounds = {"background"}
-            background_in_labels = any(
-                known_backgrounds.intersection(lower_case_labels)
-            )
-            offset = 1 if background_in_labels else 0
+        if self.region_id_name_ is not None:
 
             self.region_names_ = {
-                key: self.labels[key + offset]
+                key: self.region_id_name_[region_id]
                 for key, region_id in region_ids.items()
                 if region_id != self.background_label
             }

nilearn/maskers/tests/test_nifti_labels_masker.py

@@ -841,6 +841,107 @@ def check_region_names_after_fit(
         assert region_names_after_fit == region_names
 
 
+@pytest.mark.parametrize(
+    "background",
+    [
+        None,
+        "background",
+        "Background",
+    ],
+)
+@pytest.mark.parametrize(
+    "affine_data",
+    [
+        None,  # no resampling
+        np.diag(
+            (4, 4, 4, 4)  # with resampling
+        ),  # region_names_ matches signals after resampling drops labels
+    ],
+)
+@pytest.mark.parametrize(
+    "masking",
+    [
+        False,  # no masking
+        True,  # with masking
+    ],
+)
+@pytest.mark.parametrize(
+    "keep_masked_labels",
+    [
+        False,
+        True,
+    ],
+)
+def test_region_names_ids_match_after_fit(
+    shape_3d_default,
+    affine_eye,
+    background,
+    affine_data,
+    n_regions,
+    masking,
+    keep_masked_labels,
+):
+    """Test that the same region names and ids correspond after fit."""
+    if affine_data is None:
+        # no resampling
+        affine_data = affine_eye
+    fmri_img, _ = generate_random_img(shape_3d_default, affine=affine_data)
+    labels_img = generate_labeled_regions(
+        shape_3d_default[:3],
+        affine=affine_eye,
+        n_regions=n_regions,
+    )
+
+    region_names = generate_labels(n_regions, background=background)
+    region_ids = [region_id for region_id in np.unique(get_data(labels_img))]
+
+    if masking:
+        # create a mask_img with 3 regions
+        labels_data = get_data(labels_img)
+        mask_data = (
+            (labels_data == 1) + (labels_data == 2) + (labels_data == 5)
+        )
+        mask_img = Nifti1Image(mask_data.astype(np.int8), labels_img.affine)
+    else:
+        mask_img = None
+
+    masker = NiftiLabelsMasker(
+        labels_img,
+        labels=region_names,
+        resampling_target="data",
+        mask_img=mask_img,
+        keep_masked_labels=keep_masked_labels,
+    )
+
+    _ = masker.fit().transform(fmri_img)
+
+    check_region_names_ids_match_after_fit(
+        masker, region_names, region_ids, background
+    )
+
+
+def check_region_names_ids_match_after_fit(
+    masker, region_names, region_ids, background
+):
+    """Check the region names and ids correspondence.
+
+    Check that the same region names and ids correspond to each other
+    after fit by comparing with before fit.
+    """
+    # region_ids includes background, so we make
+    # sure that the region_names also include it
+    if not background:
+        region_names.insert(0, "background")
+    # if they don't have the same length, we can't compare them
+    if len(region_names) == len(region_ids):
+        region_id_names = {
+            region_id: region_names[i]
+            for i, region_id in enumerate(region_ids)
+        }
+        for key, region_name in masker.region_names_.items():
+            assert region_id_names[masker.region_ids_[key]] == region_name
+
+
 def generate_labels(n_regions, background=True):
     labels = []
     if background: