1
|
|
|
import torch |
2
|
|
|
import torchio as tio |
3
|
|
|
import numpy as np |
4
|
|
|
from ...utils import TorchioTestCase |
5
|
|
|
|
6
|
|
|
|
7
|
|
|
class TestRescaleIntensity(TorchioTestCase): |
8
|
|
|
"""Tests for :class:`tio.RescaleIntensity` class.""" |
9
|
|
|
|
10
|
|
|
def test_rescale_to_same_intentisy(self): |
11
|
|
|
min_t1 = float(self.sample_subject.t1.data.min()) |
12
|
|
|
max_t1 = float(self.sample_subject.t1.data.max()) |
13
|
|
|
transform = tio.RescaleIntensity(out_min_max=(min_t1, max_t1)) |
14
|
|
|
transformed = transform(self.sample_subject) |
15
|
|
|
assert np.allclose( |
16
|
|
|
transformed.t1.data, |
17
|
|
|
self.sample_subject.t1.data, |
18
|
|
|
rtol=0, |
19
|
|
|
atol=1e-05, |
20
|
|
|
) |
21
|
|
|
|
22
|
|
|
def test_min_max(self): |
23
|
|
|
transform = tio.RescaleIntensity(out_min_max=(0, 1)) |
24
|
|
|
transformed = transform(self.sample_subject) |
25
|
|
|
self.assertEqual(transformed.t1.data.min(), 0) |
26
|
|
|
self.assertEqual(transformed.t1.data.max(), 1) |
27
|
|
|
|
28
|
|
|
def test_percentiles(self): |
29
|
|
|
low_quantile = np.percentile(self.sample_subject.t1.data, 5) |
30
|
|
|
high_quantile = np.percentile(self.sample_subject.t1.data, 95) |
31
|
|
|
low_indices = (self.sample_subject.t1.data < low_quantile).nonzero( |
32
|
|
|
as_tuple=True) |
33
|
|
|
high_indices = (self.sample_subject.t1.data > high_quantile).nonzero( |
34
|
|
|
as_tuple=True) |
35
|
|
|
rescale = tio.RescaleIntensity(out_min_max=(0, 1), percentiles=(5, 95)) |
36
|
|
|
transformed = rescale(self.sample_subject) |
37
|
|
|
assert (transformed.t1.data[low_indices] == 0).all() |
38
|
|
|
assert (transformed.t1.data[high_indices] == 1).all() |
39
|
|
|
|
40
|
|
|
def test_masking_using_label(self): |
41
|
|
|
transform = tio.RescaleIntensity( |
42
|
|
|
out_min_max=(0, 1), percentiles=(5, 95), masking_method='label') |
43
|
|
|
transformed = transform(self.sample_subject) |
44
|
|
|
mask = self.sample_subject.label.data > 0 |
45
|
|
|
low_quantile = np.percentile(self.sample_subject.t1.data[mask], 5) |
46
|
|
|
high_quantile = np.percentile(self.sample_subject.t1.data[mask], 95) |
47
|
|
|
low_indices = (self.sample_subject.t1.data < low_quantile).nonzero( |
48
|
|
|
as_tuple=True) |
49
|
|
|
high_indices = (self.sample_subject.t1.data > high_quantile).nonzero( |
50
|
|
|
as_tuple=True) |
51
|
|
|
self.assertEqual(transformed.t1.data.min(), 0) |
52
|
|
|
self.assertEqual(transformed.t1.data.max(), 1) |
53
|
|
|
assert (transformed.t1.data[low_indices] == 0).all() |
54
|
|
|
assert (transformed.t1.data[high_indices] == 1).all() |
55
|
|
|
|
56
|
|
|
def test_ct(self): |
57
|
|
|
ct_max = 1500 |
58
|
|
|
ct_min = -2000 |
59
|
|
|
ct_range = ct_max - ct_min |
60
|
|
|
tensor = torch.rand(1, 30, 30, 30) * ct_range + ct_min |
61
|
|
|
ct = tio.ScalarImage(tensor=tensor) |
62
|
|
|
ct_air = -1000 |
63
|
|
|
ct_bone = 1000 |
64
|
|
|
rescale = tio.RescaleIntensity( |
65
|
|
|
out_min_max=(-1, 1), |
66
|
|
|
in_min_max=(ct_air, ct_bone), |
67
|
|
|
) |
68
|
|
|
rescaled = rescale(ct) |
69
|
|
|
assert rescaled.data.min() < -1 |
70
|
|
|
assert rescaled.data.max() > 1 |
71
|
|
|
|
72
|
|
|
def test_out_min_higher_than_out_max(self): |
73
|
|
|
with self.assertRaises(ValueError): |
74
|
|
|
tio.RescaleIntensity(out_min_max=(1, 0)) |
75
|
|
|
|
76
|
|
|
def test_too_many_values_for_out_min_max(self): |
77
|
|
|
with self.assertRaises(ValueError): |
78
|
|
|
tio.RescaleIntensity(out_min_max=(1, 2, 3)) |
79
|
|
|
|
80
|
|
|
def test_wrong_out_min_max_type(self): |
81
|
|
|
with self.assertRaises(ValueError): |
82
|
|
|
tio.RescaleIntensity(out_min_max='wrong') |
83
|
|
|
|
84
|
|
|
def test_min_percentile_higher_than_max_percentile(self): |
85
|
|
|
with self.assertRaises(ValueError): |
86
|
|
|
tio.RescaleIntensity(out_min_max=(0, 1), percentiles=(1, 0)) |
87
|
|
|
|
88
|
|
|
def test_too_many_values_for_percentiles(self): |
89
|
|
|
with self.assertRaises(ValueError): |
90
|
|
|
tio.RescaleIntensity(out_min_max=(0, 1), percentiles=(1, 2, 3)) |
91
|
|
|
|
92
|
|
|
def test_wrong_percentiles_type(self): |
93
|
|
|
with self.assertRaises(ValueError): |
94
|
|
|
tio.RescaleIntensity(out_min_max=(0, 1), percentiles='wrong') |
95
|
|
|
|