FIX failing load_confound test related to standardisation (#4411)

* ADD linear trend to the simulated image

* add latest entry

doc/changes/latest.rst

@@ -11,6 +11,8 @@ NEW
 Fixes
 -----
 
+- :bdg-dark:`Code` Fix failing test in ``test_nilearn_standardize`` on MacOS 14 by adding trend in simulated data (:gh:`4411` by `Hao-Ting Wang`_).
+
 Enhancements
 ------------
 

nilearn/interfaces/fmriprep/tests/test_load_confounds.py

@@ -21,9 +21,10 @@
     get_legal_confound,
 )
 from nilearn.maskers import NiftiMasker
+from nilearn.tests.test_signal import generate_trends
 
 
-def _simu_img(tmp_path, demean):
+def _simu_img(tmp_path, trend, demean):
     """Simulate an nifti image based on confound file \
     with some parts confounds and some parts noise."""
     file_nii, _ = create_tmp_filepath(tmp_path, copy_confounds=True)
@@ -68,6 +69,11 @@ def _simu_img(tmp_path, demean):
     # Shift the mean to non-zero
     vol = vol + 10
 
+    # add a linear trend to the data
+    if trend:
+        signal_trend = generate_trends(n_features=nx * ny * 2 * nz, length=nt)
+        vol += signal_trend.reshape(nx, ny, 2 * nz, nt)
+
     # create an nifti image with the data, and corresponding mask
     img = Nifti1Image(vol, np.eye(4))
     mask_conf = Nifti1Image(vol_conf, np.eye(4))
@@ -106,7 +112,7 @@ def _handle_non_steady(confounds):
 def _regression(confounds, tmp_path):
     """Perform simple regression with NiftiMasker."""
     # Simulate data
-    img, mask_conf, _, _, _ = _simu_img(tmp_path, demean=False)
+    img, mask_conf, _, _, _ = _simu_img(tmp_path, trend=False, demean=False)
     confounds = _handle_non_steady(confounds)
     # Do the regression
     masker = NiftiMasker(mask_img=mask_conf, standardize=True)
@@ -206,7 +212,7 @@ def test_nilearn_standardize_false(tmp_path):
 
     # Simulate data; set demean to False as standardize_confounds=True
     (img, mask_conf, mask_rand, confounds, sample_mask) = _simu_img(
-        tmp_path, demean=False
+        tmp_path, trend=False, demean=False
     )
 
     # Check that most variance is removed
@@ -248,7 +254,7 @@ def test_nilearn_standardize(
     """Test confounds removal with logical parameters for processing signal."""
     # demean is set to False to let signal.clean handle everything
     (img, mask_conf, mask_rand, confounds, mask) = _simu_img(
-        tmp_path, demean=False
+        tmp_path, trend=True, demean=False
     )
     # We now load the time series with vs without confounds
     # in voxels composed of pure confounds