|
1
|
|
|
import tempfile |
|
2
|
|
|
from pathlib import Path |
|
3
|
|
|
|
|
4
|
|
|
import torch |
|
5
|
|
|
import pytest |
|
6
|
|
|
import numpy as np |
|
7
|
|
|
import SimpleITK as sitk |
|
8
|
|
|
|
|
9
|
|
|
from ..utils import TorchioTestCase |
|
10
|
|
|
from torchio.data import io, ScalarImage |
|
11
|
|
|
from torchio.data.io import ensure_4d, nib_to_sitk, sitk_to_nib |
|
12
|
|
|
|
|
13
|
|
|
|
|
14
|
|
|
class TestIO(TorchioTestCase): |
|
15
|
|
|
"""Tests for `io` module.""" |
|
16
|
|
|
def setUp(self): |
|
17
|
|
|
super().setUp() |
|
18
|
|
|
self.nii_path = self.get_image_path('read_image') |
|
19
|
|
|
self.dicom_dir = self.get_tests_data_dir() / 'dicom' |
|
20
|
|
|
self.dicom_path = self.dicom_dir / 'IMG0001.dcm' |
|
21
|
|
|
string = ( |
|
22
|
|
|
'1.5 0.18088 -0.124887 0.65072 ' |
|
23
|
|
|
'-0.20025 0.965639 -0.165653 -11.6452 ' |
|
24
|
|
|
'0.0906326 0.18661 0.978245 11.4002 ' |
|
25
|
|
|
'0 0 0 1 ' |
|
26
|
|
|
) |
|
27
|
|
|
tensor = torch.from_numpy(np.fromstring(string, sep=' ').reshape(4, 4)) |
|
28
|
|
|
self.matrix = tensor |
|
29
|
|
|
|
|
30
|
|
|
def test_read_image(self): |
|
31
|
|
|
# I need to find something readable by nib but not sitk |
|
32
|
|
|
io.read_image(self.nii_path) |
|
33
|
|
|
|
|
34
|
|
|
def test_save_rgb(self): |
|
35
|
|
|
im = ScalarImage(tensor=torch.rand(1, 4, 5, 1)) |
|
36
|
|
|
with self.assertWarns(RuntimeWarning): |
|
37
|
|
|
im.save(self.dir / 'test.jpg') |
|
38
|
|
|
|
|
39
|
|
|
def test_read_dicom_file(self): |
|
40
|
|
|
tensor, _ = io.read_image(self.dicom_path) |
|
41
|
|
|
self.assertEqual(tuple(tensor.shape), (1, 88, 128, 1)) |
|
42
|
|
|
|
|
43
|
|
|
def test_read_dicom_dir(self): |
|
44
|
|
|
tensor, _ = io.read_image(self.dicom_dir) |
|
45
|
|
|
self.assertEqual(tuple(tensor.shape), (1, 88, 128, 17)) |
|
46
|
|
|
|
|
47
|
|
|
def test_dicom_dir_missing(self): |
|
48
|
|
|
with self.assertRaises(FileNotFoundError): |
|
49
|
|
|
io._read_dicom('missing') |
|
50
|
|
|
|
|
51
|
|
|
def test_dicom_dir_no_files(self): |
|
52
|
|
|
empty = self.dir / 'empty' |
|
53
|
|
|
empty.mkdir() |
|
54
|
|
|
with self.assertRaises(FileNotFoundError): |
|
55
|
|
|
io._read_dicom(empty) |
|
56
|
|
|
|
|
57
|
|
|
def write_read_matrix(self, suffix): |
|
58
|
|
|
out_path = self.dir / f'matrix{suffix}' |
|
59
|
|
|
io.write_matrix(self.matrix, out_path) |
|
60
|
|
|
matrix = io.read_matrix(out_path) |
|
61
|
|
|
assert torch.allclose(matrix, self.matrix) |
|
62
|
|
|
|
|
63
|
|
|
def test_matrix_itk(self): |
|
64
|
|
|
self.write_read_matrix('.tfm') |
|
65
|
|
|
self.write_read_matrix('.h5') |
|
66
|
|
|
|
|
67
|
|
|
def test_matrix_txt(self): |
|
68
|
|
|
self.write_read_matrix('.txt') |
|
69
|
|
|
|
|
70
|
|
|
def test_ensure_4d_5d(self): |
|
71
|
|
|
tensor = torch.rand(3, 4, 5, 1, 2) |
|
72
|
|
|
assert ensure_4d(tensor).shape == (2, 3, 4, 5) |
|
73
|
|
|
|
|
74
|
|
|
def test_ensure_4d_5d_t_gt_1(self): |
|
75
|
|
|
tensor = torch.rand(3, 4, 5, 2, 2) |
|
76
|
|
|
with self.assertRaises(ValueError): |
|
77
|
|
|
ensure_4d(tensor) |
|
78
|
|
|
|
|
79
|
|
|
def test_ensure_4d_2d(self): |
|
80
|
|
|
tensor = torch.rand(4, 5) |
|
81
|
|
|
assert ensure_4d(tensor).shape == (1, 4, 5, 1) |
|
82
|
|
|
|
|
83
|
|
|
def test_ensure_4d_2d_3dims_rgb_first(self): |
|
84
|
|
|
tensor = torch.rand(3, 4, 5) |
|
85
|
|
|
assert ensure_4d(tensor).shape == (3, 4, 5, 1) |
|
86
|
|
|
|
|
87
|
|
|
def test_ensure_4d_2d_3dims_rgb_last(self): |
|
88
|
|
|
tensor = torch.rand(4, 5, 3) |
|
89
|
|
|
assert ensure_4d(tensor).shape == (3, 4, 5, 1) |
|
90
|
|
|
|
|
91
|
|
|
def test_ensure_4d_3d(self): |
|
92
|
|
|
tensor = torch.rand(4, 5, 6) |
|
93
|
|
|
assert ensure_4d(tensor).shape == (1, 4, 5, 6) |
|
94
|
|
|
|
|
95
|
|
|
def test_ensure_4d_2_spatial_dims(self): |
|
96
|
|
|
tensor = torch.rand(4, 5, 6) |
|
97
|
|
|
assert ensure_4d(tensor, num_spatial_dims=2).shape == (4, 5, 6, 1) |
|
98
|
|
|
|
|
99
|
|
|
def test_ensure_4d_3_spatial_dims(self): |
|
100
|
|
|
tensor = torch.rand(4, 5, 6) |
|
101
|
|
|
assert ensure_4d(tensor, num_spatial_dims=3).shape == (1, 4, 5, 6) |
|
102
|
|
|
|
|
103
|
|
|
def test_ensure_4d_nd_not_supported(self): |
|
104
|
|
|
tensor = torch.rand(1, 2, 3, 4, 5) |
|
105
|
|
|
with self.assertRaises(ValueError): |
|
106
|
|
|
ensure_4d(tensor) |
|
107
|
|
|
|
|
108
|
|
|
def test_sitk_to_nib(self): |
|
109
|
|
|
data = np.random.rand(10, 12) |
|
110
|
|
|
image = sitk.GetImageFromArray(data) |
|
111
|
|
|
tensor, affine = sitk_to_nib(image) |
|
112
|
|
|
self.assertAlmostEqual(data.sum(), tensor.sum()) |
|
113
|
|
|
|
|
114
|
|
|
|
|
115
|
|
|
# This doesn't work as a method of the class |
|
116
|
|
|
libs = 'sitk', 'nibabel' |
|
117
|
|
|
parameters = [] |
|
118
|
|
|
for save_lib in libs: |
|
119
|
|
|
for load_lib in libs: |
|
120
|
|
|
for dims in 2, 3, 4: |
|
121
|
|
|
parameters.append((save_lib, load_lib, dims)) |
|
122
|
|
|
|
|
123
|
|
|
|
|
124
|
|
|
@pytest.mark.parametrize(('save_lib', 'load_lib', 'dims'), parameters) |
|
125
|
|
|
def test_write_nd_with_a_read_it_with_b(save_lib, load_lib, dims): |
|
126
|
|
|
shape = [1, 4, 5, 6] |
|
127
|
|
|
if dims == 2: |
|
128
|
|
|
shape[-1] = 1 |
|
129
|
|
|
elif dims == 4: |
|
130
|
|
|
shape[0] = 2 |
|
131
|
|
|
tensor = torch.randn(*shape) |
|
132
|
|
|
affine = np.eye(4) |
|
133
|
|
|
tempdir = Path(tempfile.gettempdir()) / '.torchio_tests' |
|
134
|
|
|
tempdir.mkdir(exist_ok=True) |
|
135
|
|
|
path = tempdir / 'test_io.nii' |
|
136
|
|
|
save_function = getattr(io, f'_write_{save_lib}') |
|
137
|
|
|
load_function = getattr(io, f'_read_{save_lib}') |
|
138
|
|
|
save_function(tensor, affine, path) |
|
139
|
|
|
loaded_tensor, loaded_affine = load_function(path) |
|
140
|
|
|
TorchioTestCase.assertTensorEqual( |
|
141
|
|
|
tensor.squeeze(), loaded_tensor.squeeze(), |
|
142
|
|
|
f'Save lib: {save_lib}; load lib: {load_lib}; dims: {dims}' |
|
143
|
|
|
) |
|
144
|
|
|
TorchioTestCase.assertTensorEqual(affine, loaded_affine) |
|
145
|
|
|
|
|
146
|
|
|
|
|
147
|
|
|
class TestNibabelToSimpleITK(TorchioTestCase): |
|
148
|
|
|
|
|
149
|
|
|
def setUp(self): |
|
150
|
|
|
super().setUp() |
|
151
|
|
|
self.affine = np.eye(4) |
|
152
|
|
|
|
|
153
|
|
|
def test_wrong_num_dims(self): |
|
154
|
|
|
with self.assertRaises(ValueError): |
|
155
|
|
|
nib_to_sitk(np.random.rand(10, 10), self.affine) |
|
156
|
|
|
|
|
157
|
|
|
def test_2d_single(self): |
|
158
|
|
|
data = np.random.rand(1, 10, 12, 1) |
|
159
|
|
|
image = nib_to_sitk(data, self.affine) |
|
160
|
|
|
assert image.GetDimension() == 2 |
|
161
|
|
|
assert image.GetSize() == (10, 12) |
|
162
|
|
|
assert image.GetNumberOfComponentsPerPixel() == 1 |
|
163
|
|
|
|
|
164
|
|
|
def test_2d_multi(self): |
|
165
|
|
|
data = np.random.rand(5, 10, 12, 1) |
|
166
|
|
|
image = nib_to_sitk(data, self.affine) |
|
167
|
|
|
assert image.GetDimension() == 2 |
|
168
|
|
|
assert image.GetSize() == (10, 12) |
|
169
|
|
|
assert image.GetNumberOfComponentsPerPixel() == 5 |
|
170
|
|
|
|
|
171
|
|
|
def test_2d_3d_single(self): |
|
172
|
|
|
data = np.random.rand(1, 10, 12, 1) |
|
173
|
|
|
image = nib_to_sitk(data, self.affine, force_3d=True) |
|
174
|
|
|
assert image.GetDimension() == 3 |
|
175
|
|
|
assert image.GetSize() == (10, 12, 1) |
|
176
|
|
|
assert image.GetNumberOfComponentsPerPixel() == 1 |
|
177
|
|
|
|
|
178
|
|
|
def test_2d_3d_multi(self): |
|
179
|
|
|
data = np.random.rand(5, 10, 12, 1) |
|
180
|
|
|
image = nib_to_sitk(data, self.affine, force_3d=True) |
|
181
|
|
|
assert image.GetDimension() == 3 |
|
182
|
|
|
assert image.GetSize() == (10, 12, 1) |
|
183
|
|
|
assert image.GetNumberOfComponentsPerPixel() == 5 |
|
184
|
|
|
|
|
185
|
|
|
def test_3d_single(self): |
|
186
|
|
|
data = np.random.rand(1, 8, 10, 12) |
|
187
|
|
|
image = nib_to_sitk(data, self.affine) |
|
188
|
|
|
assert image.GetDimension() == 3 |
|
189
|
|
|
assert image.GetSize() == (8, 10, 12) |
|
190
|
|
|
assert image.GetNumberOfComponentsPerPixel() == 1 |
|
191
|
|
|
|
|
192
|
|
|
def test_3d_multi(self): |
|
193
|
|
|
data = np.random.rand(5, 8, 10, 12) |
|
194
|
|
|
image = nib_to_sitk(data, self.affine) |
|
195
|
|
|
assert image.GetDimension() == 3 |
|
196
|
|
|
assert image.GetSize() == (8, 10, 12) |
|
197
|
|
|
assert image.GetNumberOfComponentsPerPixel() == 5 |
|
198
|
|
|
|
fepegar /
torchio
