Add transform to ensure shape multiple of N

docs/source/transforms/preprocessing.rst

@@ -59,6 +59,13 @@ Spatial
     :members: _get_six_bounds_parameters
 
 
+:class:`EnsureShapeMultiple`
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: EnsureShapeMultiple
+    :show-inheritance:
+
+
 :class:`Crop`
 ~~~~~~~~~~~~~
 

tests/transforms/preprocessing/test_ensure_shape_multiple.py

@@ -0,0 +1,31 @@
+import torchio as tio
+from ...utils import TorchioTestCase
+
+
+class TestEnsureShapeMultiple(TorchioTestCase):
+
+    def test_bad_method(self):
+        with self.assertRaises(ValueError):
+            tio.EnsureShapeMultiple(1, method='bad')
+
+    def test_pad(self):
+        sample_t1 = self.sample_subject.t1
+        assert sample_t1.shape == (1, 10, 20, 30)
+        transform = tio.EnsureShapeMultiple(4, method='pad')
+        transformed = transform(sample_t1)
+        assert transformed.shape == (1, 12, 20, 32)
+
+    def test_crop(self):
+        sample_t1 = self.sample_subject.t1
+        assert sample_t1.shape == (1, 10, 20, 30)
+        transform = tio.EnsureShapeMultiple(4, method='crop')
+        transformed = transform(sample_t1)
+        assert transformed.shape == (1, 8, 20, 28)
+
+    def test_2d(self):
+        sample_t1 = self.sample_subject.t1
+        sample_2d = sample_t1.data[..., :1]
+        assert sample_2d.shape == (1, 10, 20, 1)
+        transform = tio.EnsureShapeMultiple(4, method='crop')
+        transformed = transform(sample_2d)
+        assert transformed.shape == (1, 8, 20, 1)

tests/transforms/test_transforms.py

@@ -25,6 +25,7 @@ def get_transform(self, channels, is_3d=True, labels=True):
             tio.CropOrPad(cp_args),
             tio.ToCanonical(),
             tio.RandomAnisotropy(downsampling=(1.75, 2), axes=axes_downsample),
+            tio.EnsureShapeMultiple(2, method='crop'),
             tio.Resample((1, 1.1, 1.25)),
             tio.RandomFlip(axes=flip_axes, flip_probability=1),
             tio.RandomMotion(),

torchio/transforms/__init__.py

@@ -33,6 +33,7 @@
 from .preprocessing import ToCanonical
 from .preprocessing import ZNormalization
 from .preprocessing import RescaleIntensity
+from .preprocessing import EnsureShapeMultiple
 from .preprocessing import HistogramStandardization
 from .preprocessing.intensity.histogram_standardization import train as train_histogram
 from .preprocessing.label.remap_labels import RemapLabels
@@ -81,6 +82,7 @@
     'HistogramStandardization',
     'RescaleIntensity',
     'CropOrPad',
+    'EnsureShapeMultiple',
     'train_histogram',
     'RemapLabels',
     'SequentialLabels',

torchio/transforms/preprocessing/__init__.py

@@ -3,6 +3,7 @@
 from .spatial.resample import Resample
 from .spatial.crop_or_pad import CropOrPad
 from .spatial.to_canonical import ToCanonical
+from .spatial.ensure_shape_multiple import EnsureShapeMultiple
 
 from .intensity.rescale import RescaleIntensity
 from .intensity.z_normalization import ZNormalization
@@ -19,8 +20,9 @@
     'Resample',
     'ToCanonical',
     'CropOrPad',
-    'RescaleIntensity',
+    'EnsureShapeMultiple',
     'ZNormalization',
+    'RescaleIntensity',
     'HistogramStandardization',
     'RemapLabels',
     'SequentialLabels',

torchio/transforms/preprocessing/spatial/ensure_shape_multiple.py

@@ -0,0 +1,62 @@
+from typing import Union, Optional
+
+import numpy as np
+
+from ... import SpatialTransform
+from ....utils import to_tuple
+from ....data.subject import Subject
+from ....typing import TypeTripletInt
+from .crop_or_pad import CropOrPad
+
+
+class EnsureShapeMultiple(SpatialTransform):
+    """Crop or pad an image to a shape that is a multiple of :math:`N`.
+
+    Args:
+        target_multiple: Tuple :math:`(w, h, d)`. If a single value :math:`n` is
+            provided, then :math:`w = h = d = n`.
+        method: Either ``'crop'`` or ``'pad'``.
+        **kwargs: See :class:`~torchio.transforms.Transform` for additional
+            keyword arguments.
+
+    Example:
+        >>> import torchio as tio
+        >>> image = tio.datasets.Colin27().t1
+        >>> image.shape
+        (1, 181, 217, 181)
+        >>> transform = tio.EnsureShapeMultiple(8, method='pad')
+        >>> transformed = transform(image)
+        >>> transformed.shape
+        (1, 184, 224, 184)
+        >>> transform = tio.EnsureShapeMultiple(8, method='crop')
+        >>> transformed = transform(image)
+        >>> transformed.shape
+        (1, 176, 216, 176)
+        >>> image_2d = image.data[..., :1]
+        >>> image_2d.shape
+        torch.Size([1, 181, 217, 1])
+        >>> transformed = transform(image_2d)
+        >>> transformed.shape
+        torch.Size([1, 176, 216, 1])
+
+    """
+    def __init__(
+            self,
+            target_multiple: Union[int, TypeTripletInt],
+            *,
+            method: Optional[str] = 'pad',
+            **kwargs
+            ):
+        super().__init__(**kwargs)
+        self.target_multiple = np.array(to_tuple(target_multiple, 3))
+        if method not in ('crop', 'pad'):
+            raise ValueError('Method must be "crop" or "pad"')
+        self.method = method
+
+    def apply_transform(self, subject: Subject) -> Subject:
+        source_shape = np.array(subject.spatial_shape, np.uint16)
+        function = np.floor if self.method == 'crop' else np.ceil
+        integer_ratio = function(source_shape / self.target_multiple)
+        target_shape = integer_ratio * self.target_multiple
+        target_shape = np.maximum(target_shape, 1)
+        return CropOrPad(target_shape.astype(int))(subject)

torchio/transforms/transform.py

@@ -129,13 +129,14 @@ def apply_transform(self, subject: Subject):
 
     def add_transform_to_subject_history(self, subject):
         from .augmentation import RandomTransform
-        from . import Compose, OneOf, CropOrPad
+        from . import Compose, OneOf, CropOrPad, EnsureShapeMultiple
         from .preprocessing.label import SequentialLabels
         call_others = (
             RandomTransform,
             Compose,
             OneOf,
             CropOrPad,
+            EnsureShapeMultiple,
             SequentialLabels,
         )
         if not isinstance(self, call_others):