[FIX] do not mask view_img background with MNI brain mask (#3312)

fix issue #3120

doc/changes/latest.rst

@@ -27,6 +27,7 @@ Fixes
 - Fix bug introduced due to a fix in the pre-release version of scipy (``1.9.0rc1``) which now enforces that elements of a band-pass filter must meet condition ``Wn[0] < Wn[1]``.
   Now if band-pass elements are equal :func:`~nilearn.signal.butterworth` returns an unfiltered signal with a warning (:gh:`3293` by `Yasmin Mzayek`_).
 - The parameter ``alpha`` is now correctly passed to :func:`~plotting.plot_glass_brain` in :func:`~plotting.plot_connectome` (:gh:`3306` by `Koen Helwegen`_).
+- Fix plotting of background image in :func:`~nilearn.plotting.view_img` when the background is not the MNI template (:gh:`3312` by `Jerome Dockes`_).
 
 Enhancements
 ------------

nilearn/plotting/html_stat_map.py

@@ -14,7 +14,7 @@
 
 from nibabel.affines import apply_affine
 
-from ..image import resample_to_img, new_img_like, reorder_img
+from ..image import resample_to_img, new_img_like, reorder_img, get_data
 from .js_plotting_utils import get_html_template, colorscale
 from ..plotting import cm
 from ..plotting.find_cuts import find_xyz_cut_coords
@@ -26,7 +26,6 @@
 from .._utils.extmath import fast_abs_percentile
 from .._utils.niimg import _safe_get_data
 from ..datasets import load_mni152_template
-from ..masking import compute_brain_mask
 
 
 def _data_to_sprite(data):
@@ -209,19 +208,23 @@ def _load_bg_img(stat_map_img, bg_img='MNI152', black_bg='auto', dim='auto'):
     Returns: bg_img, bg_min, bg_max, black_bg
 
     """
-    if (bg_img is None or bg_img is False) and black_bg == 'auto':
-        black_bg = False
-
-    if bg_img is not None and bg_img is not False:
+    if (bg_img is None or bg_img is False):
+        if black_bg == 'auto':
+            black_bg = False
+        bg_img = new_img_like(
+            stat_map_img, np.ma.masked_all(stat_map_img.shape))
+        bg_min, bg_max = 0, 0
+    else:
         if isinstance(bg_img, str) and bg_img == "MNI152":
             bg_img = load_mni152_template()
-        bg_img, black_bg, bg_min, bg_max = _load_anat(bg_img, dim=dim,
-                                                      black_bg=black_bg)
-    else:
-        bg_img = new_img_like(stat_map_img, np.zeros(stat_map_img.shape),
-                              stat_map_img.affine)
-        bg_min = 0
-        bg_max = 0
+        else:
+            bg_img = check_niimg_3d(bg_img)
+        masked_data = np.ma.masked_inside(
+            _safe_get_data(bg_img, ensure_finite=True),
+            -1e-6, 1e-6, copy=False)
+        bg_img = new_img_like(bg_img, masked_data)
+        bg_img, black_bg, bg_min, bg_max = _load_anat(
+            bg_img, dim=dim, black_bg=black_bg)
     bg_img = reorder_img(bg_img, resample='nearest')
     return bg_img, bg_min, bg_max, black_bg
 
@@ -313,10 +316,7 @@ def _json_view_size(params):
 
 def _get_bg_mask_and_cmap(bg_img, black_bg):
     """Helper function of _json_view_data."""
-    bg_mask = _safe_get_data(compute_brain_mask(bg_img),
-                             ensure_finite=True)
-    bg_mask = np.logical_not(bg_mask).astype(float)
-    bg_mask[bg_mask == 1] = np.nan
+    bg_mask = np.ma.getmaskarray(get_data(bg_img))
     bg_cmap = copy.copy(matplotlib.cm.get_cmap('gray'))
     if black_bg:
         bg_cmap.set_bad('black')

nilearn/plotting/tests/test_html_stat_map.py

@@ -187,6 +187,14 @@ def test_load_bg_img():
     _check_affine(bg_img.affine)
 
 
+def test_get_bg_mask_and_cmap():
+    # non-regression test for issue #3120 (bg image was masked with mni
+    # template mask)
+    img, _ = _simulate_img()
+    mask, cmap = html_stat_map._get_bg_mask_and_cmap(img, False)
+    assert (mask == np.zeros(img.shape, dtype=bool)).all()
+
+
 def test_resample_stat_map():
 
     # Start with simple simulated data