|
1
|
|
|
import numpy as np |
|
2
|
|
|
from torchio.transforms import RescaleIntensity |
|
3
|
|
|
from ...utils import TorchioTestCase |
|
4
|
|
|
|
|
5
|
|
|
|
|
6
|
|
|
class TestRescaleIntensity(TorchioTestCase): |
|
7
|
|
|
"""Tests for :py:class:`RescaleIntensity` class.""" |
|
8
|
|
|
|
|
9
|
|
|
def test_rescale_to_same_intentisy(self): |
|
10
|
|
|
min_t1 = float(self.sample.t1.data.min()) |
|
11
|
|
|
max_t1 = float(self.sample.t1.data.max()) |
|
12
|
|
|
transform = RescaleIntensity(out_min_max=(min_t1, max_t1)) |
|
13
|
|
|
transformed = transform(self.sample) |
|
14
|
|
|
assert np.allclose( |
|
15
|
|
|
transformed.t1.data, self.sample.t1.data, rtol=0, atol=1e-06) |
|
16
|
|
|
|
|
17
|
|
|
def test_min_max(self): |
|
18
|
|
|
transform = RescaleIntensity(out_min_max=(0., 1.)) |
|
19
|
|
|
transformed = transform(self.sample) |
|
20
|
|
|
self.assertEqual(transformed.t1.data.min(), 0.) |
|
21
|
|
|
self.assertEqual(transformed.t1.data.max(), 1.) |
|
22
|
|
|
|
|
23
|
|
|
def test_percentiles(self): |
|
24
|
|
|
low_quantile = np.percentile(self.sample.t1.data, 5) |
|
25
|
|
|
high_quantile = np.percentile(self.sample.t1.data, 95) |
|
26
|
|
|
low_indices = (self.sample.t1.data < low_quantile).nonzero( |
|
27
|
|
|
as_tuple=True) |
|
28
|
|
|
high_indices = (self.sample.t1.data > high_quantile).nonzero( |
|
29
|
|
|
as_tuple=True) |
|
30
|
|
|
transform = RescaleIntensity(out_min_max=(0., 1.), percentiles=(5, 95)) |
|
31
|
|
|
transformed = transform(self.sample) |
|
32
|
|
|
assert (transformed.t1.data[low_indices] == 0.).all() |
|
33
|
|
|
assert (transformed.t1.data[high_indices] == 1.).all() |
|
34
|
|
|
|
|
35
|
|
|
def test_masking_using_label(self): |
|
36
|
|
|
transform = RescaleIntensity( |
|
37
|
|
|
out_min_max=(0., 1.), percentiles=(5, 95), masking_method='label') |
|
38
|
|
|
transformed = transform(self.sample) |
|
39
|
|
|
mask = self.sample.label.data > 0 |
|
40
|
|
|
low_quantile = np.percentile(self.sample.t1.data[mask], 5) |
|
41
|
|
|
high_quantile = np.percentile(self.sample.t1.data[mask], 95) |
|
42
|
|
|
low_indices = (self.sample.t1.data < low_quantile).nonzero( |
|
43
|
|
|
as_tuple=True) |
|
44
|
|
|
high_indices = (self.sample.t1.data > high_quantile).nonzero( |
|
45
|
|
|
as_tuple=True) |
|
46
|
|
|
self.assertEqual(transformed.t1.data.min(), 0.) |
|
47
|
|
|
self.assertEqual(transformed.t1.data.max(), 1.) |
|
48
|
|
|
assert (transformed.t1.data[low_indices] == 0.).all() |
|
49
|
|
|
assert (transformed.t1.data[high_indices] == 1.).all() |
|
50
|
|
|
|
|
51
|
|
|
def test_out_min_higher_than_out_max(self): |
|
52
|
|
|
with self.assertRaises(ValueError): |
|
53
|
|
|
RescaleIntensity(out_min_max=(1., 0.)) |
|
54
|
|
|
|
|
55
|
|
|
def test_too_many_values_for_out_min_max(self): |
|
56
|
|
|
with self.assertRaises(ValueError): |
|
57
|
|
|
RescaleIntensity(out_min_max=(1., 2., 3.)) |
|
58
|
|
|
|
|
59
|
|
|
def test_wrong_out_min_max_type(self): |
|
60
|
|
|
with self.assertRaises(ValueError): |
|
61
|
|
|
RescaleIntensity(out_min_max='wrong') |
|
62
|
|
|
|
|
63
|
|
|
def test_min_percentile_higher_than_max_percentile(self): |
|
64
|
|
|
with self.assertRaises(ValueError): |
|
65
|
|
|
RescaleIntensity(out_min_max=(0., 1.), percentiles=(1., 0.)) |
|
66
|
|
|
|
|
67
|
|
|
def test_too_many_values_for_percentiles(self): |
|
68
|
|
|
with self.assertRaises(ValueError): |
|
69
|
|
|
RescaleIntensity(out_min_max=(0., 1.), percentiles=(1., 2., 3.)) |
|
70
|
|
|
|
|
71
|
|
|
def test_wrong_percentiles_type(self): |
|
72
|
|
|
with self.assertRaises(ValueError): |
|
73
|
|
|
RescaleIntensity(out_min_max=(0., 1.), percentiles='wrong') |
|
74
|
|
|
|
fepegar /
torchio
