Fix/intensity remap range (#9155)

IntensityRemap is supposed to keep an images intensities in the same
[min, max] range but it rescaled with * max + min so the output ended up
in [min, min + max] instead. For example a CT scan in [-1000, 3000] came
out in [-1000, 2000] and a image in [10, 20] came out in [10, 30]. The
fix is one line, multiply by (max - min) instead and a second commit
makes RandIntensityRemap use its own seeded generator instead of numpys
global one so setting the seed actually makes the output repeatable.
These are the first tests these two transforms have ever had, without
the fixes 6 of 9 fail and with them all 9 pass.

---------

Signed-off-by: devteamaegis <devteam.aegis@gmail.com>
Co-authored-by: Eric Kerfoot <17726042+ericspod@users.noreply.github.com>

Author: devteamaegis

monai/transforms/intensity/array.py

@@ -2740,7 +2740,7 @@ def __call__(self, img: torch.Tensor) -> torch.Tensor:
         grid = torch.arange(len(noise)) / len(noise)
         noise += self.slope * grid
         # rescale
-        noise = (noise - noise.min()) / (noise.max() - noise.min()) * img_.max() + img_.min()
+        noise = (noise - noise.min()) / (noise.max() - noise.min()) * (img_.max() - img_.min()) + img_.min()
 
         # intensity remapping function
         index_img = torch.bucketize(img_, torch.tensor(vals_to_sample))
@@ -2787,17 +2787,18 @@ def __call__(self, img: torch.Tensor) -> torch.Tensor:
         img = convert_to_tensor(img, track_meta=get_track_meta())
         if self._do_transform:
             if self.channel_wise:
-                img = torch.stack(
-                    [
-                        IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))(img[i])
-                        for i in range(len(img))
-                    ]
-                )
+                img = torch.stack([self._remap(img[i]) for i in range(len(img))])
             else:
-                img = IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))(img)
+                img = self._remap(img)
 
         return img
 
+    def _remap(self, img: torch.Tensor) -> torch.Tensor:
+        # share this transform's random state so that `set_random_state` also controls the remapping curve
+        remap = IntensityRemap(self.kernel_size, self.R.choice([-self.slope, self.slope]))
+        remap.set_random_state(state=self.R)
+        return remap(img)
+
 
 class ForegroundMask(Transform):
     """

tests/transforms/test_intensity_remap.py

@@ -0,0 +1,67 @@
+# Copyright (c) MONAI Consortium
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#     http://www.apache.org/licenses/LICENSE-2.0
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+import unittest
+
+import torch
+from parameterized import parameterized
+
+from monai.transforms import IntensityRemap, RandIntensityRemap
+from tests.test_utils import assert_allclose
+
+# input ranges with a negative, zero and positive minimum
+TEST_RANGES = [(-1000.0, 3000.0), (0.0, 255.0), (10.0, 20.0)]
+
+
+class TestIntensityRemap(unittest.TestCase):
+    @parameterized.expand(TEST_RANGES)
+    def test_output_range(self, low, high):
+        img = torch.linspace(low, high, 500).reshape(1, 10, 50)
+        remap = IntensityRemap(kernel_size=10, slope=0.7)
+        remap.set_random_state(seed=0)
+        result = remap(img)
+        self.assertEqual(result.shape, img.shape)
+        assert_allclose(result.min(), low, type_test=False, atol=1e-3, rtol=1e-5)
+        assert_allclose(result.max(), high, type_test=False, atol=1e-3, rtol=1e-5)
+
+
+class TestRandIntensityRemap(unittest.TestCase):
+    @parameterized.expand(TEST_RANGES)
+    def test_output_range(self, low, high):
+        img = torch.stack([torch.linspace(low, high, 500).reshape(10, 50)] * 2)
+        remap = RandIntensityRemap(prob=1.0, kernel_size=10, channel_wise=True)
+        remap.set_random_state(seed=0)
+        result = remap(img)
+        self.assertEqual(result.shape, img.shape)
+        for c in range(img.shape[0]):
+            assert_allclose(result[c].min(), low, type_test=False, atol=1e-3, rtol=1e-5)
+            assert_allclose(result[c].max(), high, type_test=False, atol=1e-3, rtol=1e-5)
+
+    @parameterized.expand([[True], [False]])
+    def test_set_random_state(self, channel_wise):
+        img = torch.stack([torch.linspace(0.0, 100.0, 500).reshape(10, 50)] * 2)
+        results = []
+        for _ in range(2):
+            remap = RandIntensityRemap(prob=1.0, kernel_size=10, channel_wise=channel_wise)
+            remap.set_random_state(seed=0)
+            results.append(remap(img))
+        assert_allclose(results[0], results[1], type_test=False)
+
+    def test_prob_zero(self):
+        img = torch.linspace(-1.0, 1.0, 500).reshape(1, 10, 50)
+        remap = RandIntensityRemap(prob=0.0)
+        assert_allclose(remap(img), img, type_test=False)
+
+
+if __name__ == "__main__":
+    unittest.main()