diff --git a/monai/transforms/intensity/array.py b/monai/transforms/intensity/array.py
index a80f94727e..d7b405d096 100644
--- a/monai/transforms/intensity/array.py
+++ b/monai/transforms/intensity/array.py
@@ -1055,7 +1055,7 @@ def _restore(x, sub, div, zeroed_idx=None):
                 div, *_ = convert_to_dst_type(div, x)
                 return x * div + sub
             # nonzero=True: the forward mask is the output's non-zero voxels plus the recorded zeroed ones
-            mask = x != 0
+            mask = (x != 0).contiguous()
             zeroed_idx, *_ = convert_to_dst_type(zeroed_idx, mask, dtype=torch.long)
             mask.view(-1)[zeroed_idx] = True
             vals = x[mask]
diff --git a/tests/transforms/test_normalize_intensity.py b/tests/transforms/test_normalize_intensity.py
index 53fea19d33..49d9491ac0 100644
--- a/tests/transforms/test_normalize_intensity.py
+++ b/tests/transforms/test_normalize_intensity.py
@@ -176,6 +176,23 @@ def test_inverse_nonzero_value_equal_to_mean(self, _, args):
         inv = normalizer.inverse(out)
         assert_allclose(inv, img, type_test=False, rtol=0, atol=0)
 
+    @parameterized.expand([[False], [True]])
+    def test_inverse_nonzero_noncontiguous(self, channel_wise):
+        """Restore transposed inputs, including non-zero voxels normalized to zero.
+
+        Args:
+            channel_wise: whether each channel is normalized independently.
+        """
+        self.addCleanup(set_track_meta, get_track_meta())
+        set_track_meta(True)
+        img = MetaTensor(torch.tensor([[[0.0, 1.0], [2.0, 3.0]], [[0.0, 0.0], [0.0, 2.0]]])).transpose(1, 2)
+        normalizer = NormalizeIntensity(nonzero=True, channel_wise=channel_wise)
+        out = normalizer(img.clone())
+        self.assertFalse(out.is_contiguous())
+        inv = normalizer.inverse(out)
+        assert_allclose(inv, img, type_test=False, rtol=1e-6, atol=1e-6)
+        self.assertEqual(len(inv.applied_operations), 0)
+
     def test_inverse_nonzero_in_compose(self):
         self.addCleanup(set_track_meta, get_track_meta())
         set_track_meta(True)
diff --git a/tests/transforms/test_normalize_intensityd.py b/tests/transforms/test_normalize_intensityd.py
index 50a9c302db..992f8db67c 100644
--- a/tests/transforms/test_normalize_intensityd.py
+++ b/tests/transforms/test_normalize_intensityd.py
@@ -98,6 +98,23 @@ def test_inverse(self, _, args):
         inv = normalizer.inverse(out)
         assert_allclose(inv[key], original, type_test=False, rtol=1e-4, atol=1e-4)
 
+    @parameterized.expand([[False], [True]])
+    def test_inverse_nonzero_noncontiguous(self, channel_wise):
+        """Restore transposed dictionary values without losing their zeroed voxels.
+
+        Args:
+            channel_wise: whether each channel is normalized independently.
+        """
+        self.addCleanup(set_track_meta, get_track_meta())
+        set_track_meta(True)
+        img = MetaTensor(torch.tensor([[[0.0, 1.0], [2.0, 3.0]], [[0.0, 0.0], [0.0, 2.0]]])).transpose(1, 2)
+        normalizer = NormalizeIntensityd(keys="img", nonzero=True, channel_wise=channel_wise)
+        out = normalizer({"img": img.clone()})
+        self.assertFalse(out["img"].is_contiguous())
+        inv = normalizer.inverse(out)
+        assert_allclose(inv["img"], img, type_test=False, rtol=1e-6, atol=1e-6)
+        self.assertEqual(len(inv["img"].applied_operations), 0)
+
     def test_invertd_nonzero(self):
         self.addCleanup(set_track_meta, get_track_meta())
         set_track_meta(True)