|
1
|
|
|
import copy |
|
2
|
|
|
import shutil |
|
3
|
|
|
import random |
|
4
|
|
|
import tempfile |
|
5
|
|
|
import unittest |
|
6
|
|
|
from pathlib import Path |
|
7
|
|
|
from random import shuffle |
|
8
|
|
|
|
|
9
|
|
|
import torch |
|
10
|
|
|
import numpy as np |
|
11
|
|
|
from numpy.testing import assert_array_equal, assert_array_almost_equal |
|
12
|
|
|
import torchio as tio |
|
13
|
|
|
|
|
14
|
|
|
|
|
15
|
|
|
class TorchioTestCase(unittest.TestCase): |
|
16
|
|
|
|
|
17
|
|
|
def setUp(self): |
|
18
|
|
|
"""Set up test fixtures, if any.""" |
|
19
|
|
|
self.dir = Path(tempfile.gettempdir()) / '.torchio_tests' |
|
20
|
|
|
self.dir.mkdir(exist_ok=True) |
|
21
|
|
|
random.seed(42) |
|
22
|
|
|
np.random.seed(42) |
|
23
|
|
|
|
|
24
|
|
|
registration_matrix = np.array([ |
|
25
|
|
|
[1, 0, 0, 10], |
|
26
|
|
|
[0, 1, 0, 0], |
|
27
|
|
|
[0, 0, 1.2, 0], |
|
28
|
|
|
[0, 0, 0, 1] |
|
29
|
|
|
]) |
|
30
|
|
|
|
|
31
|
|
|
subject_a = tio.Subject( |
|
32
|
|
|
t1=tio.ScalarImage(self.get_image_path('t1_a')), |
|
33
|
|
|
) |
|
34
|
|
|
subject_b = tio.Subject( |
|
35
|
|
|
t1=tio.ScalarImage(self.get_image_path('t1_b')), |
|
36
|
|
|
label=tio.LabelMap(self.get_image_path('label_b', binary=True)), |
|
37
|
|
|
) |
|
38
|
|
|
subject_c = tio.Subject( |
|
39
|
|
|
label=tio.LabelMap(self.get_image_path('label_c', binary=True)), |
|
40
|
|
|
) |
|
41
|
|
|
subject_d = tio.Subject( |
|
42
|
|
|
t1=tio.ScalarImage( |
|
43
|
|
|
self.get_image_path('t1_d'), |
|
44
|
|
|
pre_affine=registration_matrix, |
|
45
|
|
|
), |
|
46
|
|
|
t2=tio.ScalarImage(self.get_image_path('t2_d')), |
|
47
|
|
|
label=tio.LabelMap(self.get_image_path('label_d', binary=True)), |
|
48
|
|
|
) |
|
49
|
|
|
subject_a4 = tio.Subject( |
|
50
|
|
|
t1=tio.ScalarImage(self.get_image_path('t1_a'), components=2), |
|
51
|
|
|
) |
|
52
|
|
|
self.subjects_list = [ |
|
53
|
|
|
subject_a, |
|
54
|
|
|
subject_a4, |
|
55
|
|
|
subject_b, |
|
56
|
|
|
subject_c, |
|
57
|
|
|
subject_d, |
|
58
|
|
|
] |
|
59
|
|
|
self.dataset = tio.SubjectsDataset(self.subjects_list) |
|
60
|
|
|
self.sample_subject = self.dataset[-1] # subject_d |
|
61
|
|
|
|
|
62
|
|
|
def make_2d(self, subject): |
|
63
|
|
|
subject = copy.deepcopy(subject) |
|
64
|
|
|
for image in subject.get_images(intensity_only=False): |
|
65
|
|
|
image.data = image.data[..., :1] |
|
66
|
|
|
return subject |
|
67
|
|
|
|
|
68
|
|
|
def make_multichannel(self, subject): |
|
69
|
|
|
subject = copy.deepcopy(subject) |
|
70
|
|
|
for image in subject.get_images(intensity_only=False): |
|
71
|
|
|
image.data = torch.cat(4 * (image.data,)) |
|
72
|
|
|
return subject |
|
73
|
|
|
|
|
74
|
|
|
def flip_affine_x(self, subject): |
|
75
|
|
|
subject = copy.deepcopy(subject) |
|
76
|
|
|
for image in subject.get_images(intensity_only=False): |
|
77
|
|
|
image.affine = np.diag((-1, 1, 1, 1)) @ image.affine |
|
78
|
|
|
return subject |
|
79
|
|
|
|
|
80
|
|
|
def get_inconsistent_shape_subject(self): |
|
81
|
|
|
"""Return a subject containing images of different shape.""" |
|
82
|
|
|
subject = tio.Subject( |
|
83
|
|
|
t1=tio.ScalarImage(self.get_image_path('t1_inc')), |
|
84
|
|
|
t2=tio.ScalarImage( |
|
85
|
|
|
self.get_image_path('t2_inc', shape=(10, 20, 31))), |
|
86
|
|
|
label=tio.LabelMap( |
|
87
|
|
|
self.get_image_path( |
|
88
|
|
|
'label_inc', |
|
89
|
|
|
shape=(8, 17, 25), |
|
90
|
|
|
binary=True, |
|
91
|
|
|
), |
|
92
|
|
|
), |
|
93
|
|
|
label2=tio.LabelMap( |
|
94
|
|
|
self.get_image_path( |
|
95
|
|
|
'label2_inc', |
|
96
|
|
|
shape=(18, 17, 25), |
|
97
|
|
|
binary=True, |
|
98
|
|
|
), |
|
99
|
|
|
), |
|
100
|
|
|
) |
|
101
|
|
|
return subject |
|
102
|
|
|
|
|
103
|
|
|
def get_reference_image_and_path(self): |
|
104
|
|
|
"""Return a reference image and its path""" |
|
105
|
|
|
path = self.get_image_path('ref', shape=(10, 20, 31), spacing=(1, 1, 2)) |
|
106
|
|
|
image = tio.ScalarImage(path) |
|
107
|
|
|
return image, path |
|
108
|
|
|
|
|
109
|
|
|
def get_subject_with_partial_volume_label_map(self, components=1): |
|
110
|
|
|
"""Return a subject with a partial-volume label map.""" |
|
111
|
|
|
return tio.Subject( |
|
112
|
|
|
t1=tio.ScalarImage( |
|
113
|
|
|
self.get_image_path('t1_d'), |
|
114
|
|
|
), |
|
115
|
|
|
label=tio.LabelMap( |
|
116
|
|
|
self.get_image_path( |
|
117
|
|
|
'label_d2', binary=False, components=components |
|
118
|
|
|
) |
|
119
|
|
|
), |
|
120
|
|
|
) |
|
121
|
|
|
|
|
122
|
|
|
def tearDown(self): |
|
123
|
|
|
"""Tear down test fixtures, if any.""" |
|
124
|
|
|
shutil.rmtree(self.dir) |
|
125
|
|
|
|
|
126
|
|
|
def get_ixi_tiny(self): |
|
127
|
|
|
root_dir = Path(tempfile.gettempdir()) / 'torchio' / 'ixi_tiny' |
|
128
|
|
|
return tio.datasets.IXITiny(root_dir, download=True) |
|
129
|
|
|
|
|
130
|
|
|
def get_image_path( |
|
131
|
|
|
self, |
|
132
|
|
|
stem, |
|
133
|
|
|
binary=False, |
|
134
|
|
|
shape=(10, 20, 30), |
|
135
|
|
|
spacing=(1, 1, 1), |
|
136
|
|
|
components=1, |
|
137
|
|
|
add_nans=False, |
|
138
|
|
|
suffix=None, |
|
139
|
|
|
force_binary_foreground=True, |
|
140
|
|
|
): |
|
141
|
|
|
shape = (*shape, 1) if len(shape) == 2 else shape |
|
142
|
|
|
data = np.random.rand(components, *shape) |
|
143
|
|
|
if binary: |
|
144
|
|
|
data = (data > 0.5).astype(np.uint8) |
|
145
|
|
|
if not data.sum() and force_binary_foreground: |
|
146
|
|
|
data[..., 0] = 1 |
|
147
|
|
|
elif self.flip_coin(): # cast some images |
|
148
|
|
|
data *= 100 |
|
149
|
|
|
dtype = np.uint8 if self.flip_coin() else np.uint16 |
|
150
|
|
|
data = data.astype(dtype) |
|
151
|
|
|
if add_nans: |
|
152
|
|
|
data[:] = np.nan |
|
153
|
|
|
affine = np.diag((*spacing, 1)) |
|
154
|
|
|
if suffix is None: |
|
155
|
|
|
suffix = random.choice(('.nii.gz', '.nii', '.nrrd', '.img', '.mnc')) |
|
156
|
|
|
path = self.dir / f'{stem}{suffix}' |
|
157
|
|
|
if self.flip_coin(): |
|
158
|
|
|
path = str(path) |
|
159
|
|
|
image = tio.ScalarImage( |
|
160
|
|
|
tensor=data, |
|
161
|
|
|
affine=affine, |
|
162
|
|
|
check_nans=not add_nans, |
|
163
|
|
|
) |
|
164
|
|
|
image.save(path) |
|
165
|
|
|
return path |
|
166
|
|
|
|
|
167
|
|
|
def flip_coin(self): |
|
168
|
|
|
return np.random.rand() > 0.5 |
|
169
|
|
|
|
|
170
|
|
|
def get_tests_data_dir(self): |
|
171
|
|
|
return Path(__file__).parent / 'image_data' |
|
172
|
|
|
|
|
173
|
|
|
def assertTensorNotEqual(self, *args, **kwargs): # noqa: N802 |
|
174
|
|
|
message_kwarg = dict(msg=args[2]) if len(args) == 3 else {} |
|
175
|
|
|
with self.assertRaises(AssertionError, **message_kwarg): |
|
176
|
|
|
self.assertTensorEqual(*args, **kwargs) |
|
177
|
|
|
|
|
178
|
|
|
@staticmethod |
|
179
|
|
|
def assertTensorEqual(*args, **kwargs): # noqa: N802 |
|
180
|
|
|
assert_array_equal(*args, **kwargs) |
|
181
|
|
|
|
|
182
|
|
|
@staticmethod |
|
183
|
|
|
def assertTensorAlmostEqual(*args, **kwargs): # noqa: N802 |
|
184
|
|
|
assert_array_almost_equal(*args, **kwargs) |
|
185
|
|
|
|
|
186
|
|
|
def get_large_composed_transform(self): |
|
187
|
|
|
all_classes = get_all_random_transforms() |
|
188
|
|
|
shuffle(all_classes) |
|
189
|
|
|
transforms = [t() for t in all_classes] |
|
190
|
|
|
# Hack as default patch size for RandomSwap is 15 and sample_subject |
|
191
|
|
|
# is (10, 20, 30) |
|
192
|
|
|
for tr in transforms: |
|
193
|
|
|
if tr.name == 'RandomSwap': |
|
194
|
|
|
tr.patch_size = np.array((10, 10, 10)) |
|
195
|
|
|
return tio.Compose(transforms) |
|
196
|
|
|
|
|
197
|
|
|
|
|
198
|
|
|
def get_all_random_transforms(): |
|
199
|
|
|
transforms_names = [ |
|
200
|
|
|
name |
|
201
|
|
|
for name in dir(tio.transforms) |
|
202
|
|
|
if name.startswith('Random') |
|
203
|
|
|
] |
|
204
|
|
|
classes = [getattr(tio.transforms, name) for name in transforms_names] |
|
205
|
|
|
return classes |
|
206
|
|
|
|