bugfix resample only if type is not contour or contourf (#2656)

* bugfix resample only if type is not contour or contourf

* fix case where img is a string

* nearest interpolation for binary images

* Add explanations in _map_show

* Add warning in resampling when data is binary and interpolation not nearest

* Add tests

nilearn/_utils/niimg.py

@@ -146,6 +146,29 @@ def load_niimg(niimg, dtype=None):
     return niimg
 
 
+def _is_binary_niimg(niimg):
+    """Returns whether a given niimg is binary or not.
+
+    Parameters
+    ----------
+    niimg: Niimg-like object
+        See http://nilearn.github.io/manipulating_images/input_output.html
+        Image to test.
+
+    Returns
+    -------
+    is_binary: Boolean
+        True if binary, False otherwise.
+
+    """
+    niimg = load_niimg(niimg)
+    data = _safe_get_data(niimg, ensure_finite=True)
+    unique_values = np.unique(data)
+    if len(unique_values) != 2:
+        return False
+    return sorted(list(unique_values)) == [0,1]
+
+
 def copy_img(img):
     """Copy an image to a nibabel.Nifti1Image.
 

nilearn/image/resampling.py

@@ -267,6 +267,14 @@ def _resample_one_img(data, A, b, target_shape,
         data = _extrapolate_out_mask(data, np.logical_not(not_finite),
                                      iterations=2)[0]
 
+    # If data is binary and interpolation is continuous or linear,
+    # warn the user as this might be unintentional
+    if sorted(list(np.unique(data))) == [0,1] and interpolation_order != 0:
+        warnings.warn("Resampling binary images with continuous or "
+                      "linear interpolation. This might lead to "
+                      "unexpected results. You might consider using "
+                      "nearest interpolation instead.")
+
     # Suppresses warnings in https://github.com/nilearn/nilearn/issues/1363
     with warnings.catch_warnings():
         if LooseVersion(scipy.__version__) >= LooseVersion('0.18'):

nilearn/image/tests/test_resampling.py

@@ -13,6 +13,7 @@
 
 from nibabel import Nifti1Image
 
+from nilearn import _utils
 from nilearn.image.resampling import resample_img, resample_to_img, reorder_img
 from nilearn.image.resampling import from_matrix_vector, coord_transform
 from nilearn.image.resampling import get_bounds
@@ -249,6 +250,27 @@ def test_resampling_error_checks():
                      interpolation="an_invalid_interpolation"
                      )
 
+    # Resampling a binary image with continuous or
+    # linear interpolation should raise a warning.
+    data_binary = rng.randint(4, size=(1, 4, 4))
+    data_binary[data_binary>0] = 1
+    assert sorted(list(np.unique(data_binary))) == [0,1]
+
+    rot = rotation(0, np.pi / 4)
+    img_binary = Nifti1Image(data_binary, np.eye(4))
+    assert _utils.niimg._is_binary_niimg(img_binary)
+
+    with pytest.warns(Warning, match="Resampling binary images with"):
+        rot_img = resample_img(img_binary,
+                               target_affine=rot,
+                               interpolation='continuous')
+
+    with pytest.warns(Warning, match="Resampling binary images with"):
+        rot_img = resample_img(img_binary,
+                               target_affine=rot,
+                               interpolation='linear')
+
+
     # Noop
     target_shape = shape[:3]
 

nilearn/plotting/displays.py

@@ -28,7 +28,7 @@
 from .edge_detect import _edge_map
 from .find_cuts import find_xyz_cut_coords, find_cut_slices
 from .. import _utils
-from ..image import new_img_like
+from ..image import new_img_like, load_img
 from ..image.resampling import (get_bounds, reorder_img, coord_transform,
                                 get_mask_bounds)
 from nilearn.image import get_data
@@ -834,7 +834,17 @@ def add_contours(self, img, threshold=1e-6, filled=False, **kwargs):
     def _map_show(self, img, type='imshow',
                   resampling_interpolation='continuous',
                   threshold=None, **kwargs):
-        img = reorder_img(img, resample=resampling_interpolation)
+        # In the special case where the affine of img is not diagonal,
+        # the function `reorder_img` will trigger a resampling
+        # of the provided image with a continuous interpolation
+        # since this is the default value here. In the special
+        # case where this image is binary, such as when this function
+        # is called from `add_contours`, continuous interpolation
+        # does not make sense and we turn to nearest interpolation instead.
+        if _utils.niimg._is_binary_niimg(img):
+            img = reorder_img(img, resample='nearest')
+        else:
+            img = reorder_img(img, resample=resampling_interpolation)
         threshold = float(threshold) if threshold is not None else None
 
         if threshold is not None:

nilearn/plotting/tests/test_img_plotting.py

@@ -14,6 +14,7 @@
 
 from scipy import sparse
 
+from nilearn import _utils
 from nilearn.image.resampling import coord_transform, reorder_img
 from nilearn._utils import data_gen
 from nilearn.image import get_data
@@ -509,6 +510,7 @@ def test_plot_img_with_resampling(testdata_3d):
                        [0.,  0.,  1.,  0.],
                        [0.,  0.,  0.,  1.]])
     img = nibabel.Nifti1Image(data, affine)
+    assert not _utils.niimg._is_binary_niimg(img)
     display = plot_img(img)
     display.add_overlay(img)
     display.add_contours(img, contours=2, linewidth=4,
@@ -518,6 +520,21 @@ def test_plot_img_with_resampling(testdata_3d):
     # Save execution time and memory
     plt.close()
 
+def test_plot_binary_img_with_resampling(testdata_3d):
+    data = get_data(testdata_3d['img'])
+    data[data > 0] = 1
+    data[data < 0] = 0
+    affine = np.array([[1., -1.,  0.,  0.],
+                       [1.,  1.,  0.,  0.],
+                       [0.,  0.,  1.,  0.],
+                       [0.,  0.,  0.,  1.]])
+    img = nibabel.Nifti1Image(data, affine)
+    assert _utils.niimg._is_binary_niimg(img)
+    display = plot_img(img)
+    display.add_overlay(img)
+    display.add_contours(img)
+    plt.close()
+
 
 def test_plot_noncurrent_axes():
     """Regression test for Issue #450"""