-
Notifications
You must be signed in to change notification settings - Fork 239
/
image.py
516 lines (447 loc) · 17.9 KB
/
image.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
import warnings
from pathlib import Path
from typing import Any, Dict, Tuple, Optional, Union, Sequence, List
import torch
import humanize
import numpy as np
import nibabel as nib
import SimpleITK as sitk
from ..utils import (
nib_to_sitk,
get_rotation_and_spacing_from_affine,
get_stem,
ensure_4d,
)
from ..torchio import (
TypeData,
TypePath,
TypeTripletInt,
TypeTripletFloat,
DATA,
TYPE,
AFFINE,
PATH,
STEM,
INTENSITY,
LABEL,
)
from .io import read_image, write_image
PROTECTED_KEYS = DATA, AFFINE, TYPE, PATH, STEM
class Image(dict):
r"""TorchIO image.
For information about medical image orientation, check out `NiBabel docs`_,
the `3D Slicer wiki`_, `Graham Wideman's website`_, `FSL docs`_ or
`SimpleITK docs`_.
Args:
path: Path to a file or sequence of paths to files that can be read by
:mod:`SimpleITK` or :mod:`nibabel`, or to a directory containing
DICOM files. If :py:attr:`tensor` is given, the data in
:py:attr:`path` will not be read.
If a sequence of paths is given, data
will be concatenated on the channel dimension so spatial
dimensions must match.
type: Type of image, such as :attr:`torchio.INTENSITY` or
:attr:`torchio.LABEL`. This will be used by the transforms to
decide whether to apply an operation, or which interpolation to use
when resampling. For example, `preprocessing`_ and `augmentation`_
intensity transforms will only be applied to images with type
:attr:`torchio.INTENSITY`. Spatial transforms will be applied to
all types, and nearest neighbor interpolation is always used to
resample images with type :attr:`torchio.LABEL`.
The type :attr:`torchio.SAMPLING_MAP` may be used with instances of
:py:class:`~torchio.data.sampler.weighted.WeightedSampler`.
tensor: If :py:attr:`path` is not given, :attr:`tensor` must be a 4D
:py:class:`torch.Tensor` or NumPy array with dimensions
:math:`(C, W, H, D)`.
affine: If :attr:`path` is not given, :attr:`affine` must be a
:math:`4 \times 4` NumPy array. If ``None``, :attr:`affine` is an
identity matrix.
check_nans: If ``True``, issues a warning if NaNs are found
in the image. If ``False``, images will not be checked for the
presence of NaNs.
**kwargs: Items that will be added to the image dictionary, e.g.
acquisition parameters.
TorchIO images are `lazy loaders`_, i.e. the data is only loaded from disk
when needed.
Example:
>>> import torchio
>>> image = torchio.ScalarImage('t1.nii.gz') # subclass of Image
>>> image # not loaded yet
ScalarImage(path: t1.nii.gz; type: intensity)
>>> times_two = 2 * image.data # data is loaded and cached here
>>> image
ScalarImage(shape: (1, 256, 256, 176); spacing: (1.00, 1.00, 1.00); orientation: PIR+; memory: 44.0 MiB; type: intensity)
>>> image.save('doubled_image.nii.gz')
.. _lazy loaders: https://en.wikipedia.org/wiki/Lazy_loading
.. _preprocessing: https://torchio.readthedocs.io/transforms/preprocessing.html#intensity
.. _augmentation: https://torchio.readthedocs.io/transforms/augmentation.html#intensity
.. _NiBabel docs: https://nipy.org/nibabel/image_orientation.html
.. _3D Slicer wiki: https://www.slicer.org/wiki/Coordinate_systems
.. _FSL docs: https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Orientation%20Explained
.. _SimpleITK docs: https://simpleitk.readthedocs.io/en/master/fundamentalConcepts.html
.. _Graham Wideman's website: http://www.grahamwideman.com/gw/brain/orientation/orientterms.htm
"""
def __init__(
self,
path: Union[TypePath, Sequence[TypePath], None] = None,
type: str = None,
tensor: Optional[TypeData] = None,
affine: Optional[TypeData] = None,
check_nans: bool = False, # removed by ITK by default
channels_last: bool = False,
**kwargs: Dict[str, Any],
):
self.check_nans = check_nans
self.channels_last = channels_last
if type is None:
warnings.warn(
'Not specifying the image type is deprecated and will be'
' mandatory in the future. You can probably use ScalarImage or'
' LabelMap instead'
)
type = INTENSITY
if path is None and tensor is None:
raise ValueError('A value for path or tensor must be given')
self._loaded = False
tensor = self.parse_tensor(tensor)
affine = self.parse_affine(affine)
if tensor is not None:
self[DATA] = tensor
self[AFFINE] = affine
self._loaded = True
for key in PROTECTED_KEYS:
if key in kwargs:
message = f'Key "{key}" is reserved. Use a different one'
raise ValueError(message)
super().__init__(**kwargs)
self.path = self._parse_path(path)
self[PATH] = '' if self.path is None else str(self.path)
self[STEM] = '' if self.path is None else get_stem(self.path)
self[TYPE] = type
def __repr__(self):
properties = []
if self._loaded:
properties.extend([
f'shape: {self.shape}',
f'spacing: {self.get_spacing_string()}',
f'orientation: {"".join(self.orientation)}+',
f'memory: {humanize.naturalsize(self.memory, binary=True)}',
])
else:
properties.append(f'path: "{self.path}"')
properties.append(f'type: {self.type}')
properties = '; '.join(properties)
string = f'{self.__class__.__name__}({properties})'
return string
def __getitem__(self, item):
if item in (DATA, AFFINE):
if item not in self:
self.load()
return super().__getitem__(item)
def __array__(self):
return self[DATA].numpy()
def __copy__(self):
kwargs = dict(
tensor=self.data,
affine=self.affine,
type=self.type,
path=self.path,
)
for key, value in self.items():
if key in PROTECTED_KEYS: continue
kwargs[key] = value # should I copy? deepcopy?
return self.__class__(**kwargs)
@property
def data(self):
return self[DATA]
@property
def tensor(self):
return self.data
@property
def affine(self):
return self[AFFINE]
@property
def type(self):
return self[TYPE]
@property
def shape(self) -> Tuple[int, int, int, int]:
return tuple(self.data.shape)
@property
def spatial_shape(self) -> TypeTripletInt:
return self.shape[1:]
def check_is_2d(self):
if not self.is_2d():
message = f'Image is not 2D. Spatial shape: {self.spatial_shape}'
raise RuntimeError(message)
@property
def height(self) -> int:
self.check_is_2d()
return self.spatial_shape[1]
@property
def width(self) -> int:
self.check_is_2d()
return self.spatial_shape[0]
@property
def orientation(self):
return nib.aff2axcodes(self.affine)
@property
def spacing(self):
_, spacing = get_rotation_and_spacing_from_affine(self.affine)
return tuple(spacing)
@property
def memory(self):
return np.prod(self.shape) * 4 # float32, i.e. 4 bytes per voxel
def axis_name_to_index(self, axis: str):
"""Convert an axis name to an axis index.
Args:
axis: Possible inputs are ``'Left'``, ``'Right'``, ``'Anterior'``,
``'Posterior'``, ``'Inferior'``, ``'Superior'``. Lower-case versions
and first letters are also valid, as only the first letter will be
used.
.. note:: If you are working with animals, you should probably use
``'Superior'``, ``'Inferior'``, ``'Anterior'`` and ``'Posterior'``
for ``'Dorsal'``, ``'Ventral'``, ``'Rostral'`` and ``'Caudal'``,
respectively.
.. note:: If your images are 2D, you can use ``'Top'``, ``'Bottom'``,
``'Left'`` and ``'Right'``.
"""
# Top and bottom are used for the vertical 2D axis as the use of
# Height vs Horizontal might be ambiguous
if not isinstance(axis, str):
raise ValueError('Axis must be a string')
axis = axis[0].upper()
# Generally, TorchIO tensors are (C, W, H, D)
if axis in 'TB': # Top, Bottom
return -2
else:
try:
index = self.orientation.index(axis)
except ValueError:
index = self.orientation.index(self.flip_axis(axis))
# Return negative indices so that it does not matter whether we
# refer to spatial dimensions or not
index = -3 + index
return index
# flake8: noqa: E701
@staticmethod
def flip_axis(axis):
if axis == 'R': return 'L'
elif axis == 'L': return 'R'
elif axis == 'A': return 'P'
elif axis == 'P': return 'A'
elif axis == 'I': return 'S'
elif axis == 'S': return 'I'
else:
values = ', '.join('LRPAISTB')
message = f'Axis not understood. Please use one of: {values}'
raise ValueError(message)
def get_spacing_string(self):
strings = [f'{n:.2f}' for n in self.spacing]
string = f'({", ".join(strings)})'
return string
def get_bounds(self):
"""Get image bounds in mm."""
first_index = 3 * (-0.5,)
last_index = np.array(self.spatial_shape) - 0.5
first_point = nib.affines.apply_affine(self.affine, first_index)
last_point = nib.affines.apply_affine(self.affine, last_index)
array = np.array((first_point, last_point))
bounds_x, bounds_y, bounds_z = array.T.tolist()
return bounds_x, bounds_y, bounds_z
@staticmethod
def _parse_single_path(
path: TypePath
) -> Path:
try:
path = Path(path).expanduser()
except TypeError:
message = (
f'Expected type str or Path but found {path} with '
f'{type(path)} instead'
)
raise TypeError(message)
except RuntimeError:
message = (
f'Conversion to path not possible for variable: {path}'
)
raise RuntimeError(message)
if not (path.is_file() or path.is_dir()): # might be a dir with DICOM
raise FileNotFoundError(f'File not found: {path}')
return path
def _parse_path(
self,
path: Union[TypePath, Sequence[TypePath]]
) -> Union[Path, List[Path]]:
if path is None:
return None
if isinstance(path, (str, Path)):
return self._parse_single_path(path)
else:
return [self._parse_single_path(p) for p in path]
def parse_tensor(self, tensor: TypeData) -> torch.Tensor:
if tensor is None:
return None
if isinstance(tensor, np.ndarray):
tensor = torch.from_numpy(tensor.astype(np.float32))
elif isinstance(tensor, torch.Tensor):
tensor = tensor.float()
if tensor.ndim != 4:
raise ValueError('Input tensor must be 4D')
if self.check_nans and torch.isnan(tensor).any():
warnings.warn(f'NaNs found in tensor')
return tensor
def parse_tensor_shape(self, tensor: torch.Tensor) -> torch.Tensor:
return ensure_4d(tensor)
@staticmethod
def parse_affine(affine: np.ndarray) -> np.ndarray:
if affine is None:
return np.eye(4)
if not isinstance(affine, np.ndarray):
raise TypeError(f'Affine must be a NumPy array, not {type(affine)}')
if affine.shape != (4, 4):
raise ValueError(f'Affine shape must be (4, 4), not {affine.shape}')
return affine
def load(self) -> None:
r"""Load the image from disk.
Returns:
Tuple containing a 4D tensor of size :math:`(C, W, H, D)` and a 2D
:math:`4 \times 4` affine matrix to convert voxel indices to world
coordinates.
"""
if self._loaded:
return
paths = self.path if isinstance(self.path, list) else [self.path]
tensor, affine = self.read_and_check(paths[0])
tensors = [tensor]
for path in paths[1:]:
new_tensor, new_affine = self.read_and_check(path)
if not np.array_equal(affine, new_affine):
message = (
'Files have different affine matrices.'
f'\nMatrix of {paths[0]}:'
f'\n{affine}'
f'\nMatrix of {path}:'
f'\n{new_affine}'
)
warnings.warn(message, RuntimeWarning)
if not tensor.shape[1:] == new_tensor.shape[1:]:
message = (
f'Files shape do not match, found {tensor.shape}'
f'and {new_tensor.shape}'
)
RuntimeError(message)
tensors.append(new_tensor)
tensor = torch.cat(tensors)
self[DATA] = tensor
self[AFFINE] = affine
self._loaded = True
def read_and_check(self, path):
tensor, affine = read_image(path)
tensor = self.parse_tensor_shape(tensor)
if self.channels_last:
tensor = tensor.permute(3, 0, 1, 2)
if self.check_nans and torch.isnan(tensor).any():
warnings.warn(f'NaNs found in file "{path}"')
return tensor, affine
def save(self, path: TypePath, squeeze: bool = True):
"""Save image to disk.
Args:
path: String or instance of :py:class:`pathlib.Path`.
squeeze: If ``True``, the singleton dimensions will be removed
before saving.
"""
write_image(
self[DATA],
self[AFFINE],
path,
squeeze=squeeze,
)
def is_2d(self) -> bool:
return self.shape[-1] == 1
def numpy(self) -> np.ndarray:
"""Get a NumPy array containing the image data."""
return np.asarray(self)
def as_sitk(self, **kwargs) -> sitk.Image:
"""Get the image as an instance of :py:class:`sitk.Image`."""
return nib_to_sitk(self[DATA], self[AFFINE], **kwargs)
def get_center(self, lps: bool = False) -> TypeTripletFloat:
"""Get image center in RAS+ or LPS+ coordinates.
Args:
lps: If ``True``, the coordinates will be in LPS+ orientation, i.e.
the first dimension grows towards the left, etc. Otherwise, the
coordinates will be in RAS+ orientation.
"""
size = np.array(self.spatial_shape)
center_index = (size - 1) / 2
r, a, s = nib.affines.apply_affine(self.affine, center_index)
if lps:
return (-r, -a, s)
else:
return (r, a, s)
def set_check_nans(self, check_nans: bool):
self.check_nans = check_nans
def crop(self, index_ini: TypeTripletInt, index_fin: TypeTripletInt):
new_origin = nib.affines.apply_affine(self.affine, index_ini)
new_affine = self.affine.copy()
new_affine[:3, 3] = new_origin
i0, j0, k0 = index_ini
i1, j1, k1 = index_fin
patch = self.data[:, i0:i1, j0:j1, k0:k1].clone()
kwargs = dict(
tensor=patch,
affine=new_affine,
type=self.type,
path=self.path,
)
for key, value in self.items():
if key in PROTECTED_KEYS: continue
kwargs[key] = value # should I copy? deepcopy?
return self.__class__(**kwargs)
class ScalarImage(Image):
"""Alias for :py:class:`~torchio.Image` of type :py:attr:`torchio.INTENSITY`.
Example:
>>> import torch
>>> import torchio
>>> # Loading from a file
>>> t1_image = torchio.ScalarImage('t1.nii.gz')
>>> dmri = torchio.ScalarImage(tensor=torch.rand(32, 128, 128, 88))
>>> image = torchio.ScalarImage('safe_image.nrrd', check_nans=False)
>>> data, affine = image.data, image.affine
>>> affine.shape
(4, 4)
>>> image.data is image[torchio.DATA]
True
>>> image.data is image.tensor
True
>>> type(image.data)
torch.Tensor
See :py:class:`~torchio.Image` for more information.
Raises:
ValueError: A :py:attr:`type` is used for instantiation.
"""
def __init__(self, *args, **kwargs):
if 'type' in kwargs and kwargs['type'] != INTENSITY:
raise ValueError('Type of ScalarImage is always torchio.INTENSITY')
kwargs.update({'type': INTENSITY})
super().__init__(*args, **kwargs)
class LabelMap(Image):
"""Alias for :py:class:`~torchio.Image` of type :py:attr:`torchio.LABEL`.
Example:
>>> import torch
>>> import torchio
>>> labels = torchio.LabelMap(tensor=torch.rand(128, 128, 68) > 0.5)
>>> labels = torchio.LabelMap('t1_seg.nii.gz') # loading from a file
>>> tpm = torchio.LabelMap( # loading from files
... 'gray_matter.nii.gz',
... 'white_matter.nii.gz',
... 'csf.nii.gz',
... )
See :py:class:`~torchio.data.image.Image` for more information.
Raises:
ValueError: If a value for :py:attr:`type` is given.
"""
def __init__(self, *args, **kwargs):
if 'type' in kwargs and kwargs['type'] != LABEL:
raise ValueError('Type of LabelMap is always torchio.LABEL')
kwargs.update({'type': LABEL})
super().__init__(*args, **kwargs)