-
Notifications
You must be signed in to change notification settings - Fork 0
/
preprocess.py
executable file
·181 lines (134 loc) · 6.27 KB
/
preprocess.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
import SimpleITK as sitk
import numpy as np
from scipy.spatial.transform import Rotation as R
from dltk.io.preprocessing import whitening
"""
img: simpleitk input image
angle: radian angle to rotate around the z axis
size: voxel size for resampled data
"""
def rotate_image(img, angle, size=[64, 64, 64], is_label=False):
rotation_center = (0, 0, 0)
rotation = sitk.VersorTransform(R.from_euler('Z', angle).as_quat(), rotation_center)
rigid_versor = sitk.VersorRigid3DTransform()
rigid_versor.SetRotation(rotation.GetVersor())
rigid_versor.SetCenter(rotation_center)
out_origin, out_size, out_spacing = get_output_parameters(img, rigid_versor, size)
resample_filter = sitk.ResampleImageFilter()
resample_filter.SetTransform(rigid_versor)
if is_label:
resample_filter.SetInterpolator(sitk.sitkNearestNeighbor)
else:
resample_filter.SetInterpolator(sitk.sitkBSpline)
resample_filter.SetSize(size)
resample_filter.SetOutputOrigin(out_origin)
resample_filter.SetOutputSpacing(out_spacing)
resample_filter.SetOutputDirection(img.GetDirection())
if is_label:
resample_filter.SetOutputPixelType(sitk.sitkUInt8)
else:
resample_filter.SetOutputPixelType(sitk.sitkFloat32)
resample_filter.SetDefaultPixelValue(0.0)
output_img = resample_filter.Execute(img)
if is_label:
return sitk.GetArrayFromImage(output_img)
else:
return whitening(sitk.GetArrayFromImage(output_img))
"""
img: simpleitk input image
axes: 1 for no flip, -1 for a flip of array of (int, 3)
size: voxel size for resampled data
"""
def flip_image(img, axes=[1, -1, 1], size=[64, 64, 64], is_label=False):
out_origin, out_size, out_spacing = get_output_parameters(img, sitk.Transform(3, sitk.sitkIdentity), size)
rotation_center = (0, 0, 0)
rotation = sitk.VersorTransform(np.array([0., 0., 0., 1.]), rotation_center)
rigid_versor = sitk.VersorRigid3DTransform()
rigid_versor.SetRotation(rotation.GetVersor())
rigid_versor.SetCenter(rotation_center)
rigid_versor.SetMatrix([axes[0], 0, 0, 0, axes[1], 0, 0, 0, axes[2]])
resample_filter = sitk.ResampleImageFilter()
resample_filter.SetTransform(rigid_versor)
if is_label:
resample_filter.SetInterpolator(sitk.sitkNearestNeighbor)
else:
resample_filter.SetInterpolator(sitk.sitkBSpline)
resample_filter.SetSize(size)
resample_filter.SetOutputOrigin(img.GetOrigin())
resample_filter.SetOutputSpacing(out_spacing)
resample_filter.SetOutputDirection(img.GetDirection())
if is_label:
resample_filter.SetOutputPixelType(sitk.sitkUInt8)
else:
resample_filter.SetOutputPixelType(sitk.sitkFloat32)
resample_filter.SetDefaultPixelValue(0.0)
output_img = resample_filter.Execute(img)
if is_label:
return sitk.GetArrayFromImage(output_img)
else:
return whitening(sitk.GetArrayFromImage(output_img))
"""
given an image and a transform, provide the transformed bounds
returns: output_origin, size and spacing based on a given transform
output_origin : the origin of the image given a transform
output_spacing: the spacing given the size input set at 64 voxels as a default.
output_size : the size given the input image spacing
"""
def get_output_parameters(image, transform, size=[64, 64, 64]):
# origin and maximum of the transformed image.
x0, y0, z0 = image.GetOrigin()
x1, y1, z1 = image.TransformIndexToPhysicalPoint(image.GetSize())
trans_pts = []
for x in (x0, x1):
for y in (y0, y1):
for z in (z0, z1):
trans_pt = transform.GetInverse().TransformPoint((x, y, z))
trans_pts.append(trans_pt)
min_arr = np.array(trans_pts).min(axis=0)
max_arr = np.array(trans_pts).max(axis=0)
output_origin = min_arr
output_size = np.round(((max_arr - min_arr) / image.GetSpacing())).astype(int)
output_spacing = ((max_arr - min_arr) / size).astype(float)
# print(output_size)
return output_origin, output_size.tolist(), output_spacing.tolist()
"""
Pre-process and augment data (if defined)
Returns a list of all pre-processed/augmented data volumes as tuples
"""
def preprocess(volume_list, augment_data=False):
preprocessed_volumes = []
for volume_tuple in volume_list:
bmode, pd, label = load_volumes(volume_tuple)
preprocessed_volumes.append((bmode, pd, label))
if augment_data:
preprocessed_volumes += augment(volume_tuple)
return preprocessed_volumes
"""
Augments tuple of volumes (BMode, PD, Label) and returns a list of all augmented volumes (as tuples)
TO-DO: Currently just returns an array of the same volumes as a tuple array but should insert logic here
NOTE: It should not return the input volumes in the return array since its already added to the full volume list
ONLY append the augmentations
"""
def augment(volume_tuple):
bmode, pd, label = sitk.ReadImage(volume_tuple[0], sitk.sitkFloat32), sitk.ReadImage(volume_tuple[1], sitk.sitkFloat32), sitk.ReadImage(volume_tuple[2], sitk.sitkUInt8)
augmented_tuples = []
size = [64, 64, 64]
# initial go - -20 to +20 degrees (5 deg increment) no zero
# angles = array([-0.34906585, -0.26179939, -0.17453293, -0.08726646, 0.08726646, 0.17453293, 0.26179939, 0.34906585])
# now with more angles (-40 + 40) in 4 degree increments... to get to ~27 we augmented with prior
angles = np.linspace(-np.pi / 18, np.pi / 18, 11)
angles = angles[angles != 0]
for rad in angles:
augmented_tuples.append((rotate_image(bmode, rad, size), rotate_image(pd, rad, size),
rotate_image(label, rad, size, True)))
axes_flip = [-1, 1, 1], [1, -1, 1], [-1, -1, 1]
for a in axes_flip:
augmented_tuples.append(
(flip_image(bmode, a, size), flip_image(pd, a, size), flip_image(label, a, size, True)))
return augmented_tuples
def load_volumes(volume_tuple):
bmode, pd, label = sitk.ReadImage(volume_tuple[0], sitk.sitkFloat32), sitk.ReadImage(volume_tuple[1], sitk.sitkFloat32), sitk.ReadImage(volume_tuple[2], sitk.sitkUInt8)
bmode_vol = sitk.GetArrayFromImage(bmode)
pd_vol = sitk.GetArrayFromImage(pd)
label_vol = sitk.GetArrayFromImage(label)
return whitening(bmode_vol), whitening(pd_vol), label_vol