|
1
|
|
|
import torch |
|
2
|
|
|
from torchio.data.image import ScalarImage |
|
3
|
|
|
from ....data.subject import Subject |
|
4
|
|
|
from ...intensity_transform import IntensityTransform |
|
5
|
|
|
from typing import Optional |
|
6
|
|
|
|
|
7
|
|
|
|
|
8
|
|
|
class SlabProjection(IntensityTransform): |
|
9
|
|
|
"""Project intensities along a given axis, possibly with sliding slabs. |
|
10
|
|
|
|
|
11
|
|
|
Args: |
|
12
|
|
|
axis: Index for the axis dimension to project across. For 3D images, |
|
13
|
|
|
possible values are 0, 1, or 2, for the 1st, 2nd, or 3rd spatial |
|
14
|
|
|
dimension (ignoring the channel dimension). |
|
15
|
|
|
slab_thickness: Thickness of slab projections. In other words, the |
|
16
|
|
|
number of voxels in the ``axis`` dimension to project across. |
|
17
|
|
|
If ``None``, the projection will be done across the entire span of |
|
18
|
|
|
the ``axis`` dimension (i.e. ``axis`` dimension will be reduced to |
|
19
|
|
|
1). |
|
20
|
|
|
stride: Number of voxels to stride along the ``axis`` dimension between |
|
21
|
|
|
slab projections. Default is 1. |
|
22
|
|
|
projection_type: Type of intensity projection. Possible inputs are |
|
23
|
|
|
``'max'`` (the default), ``'min'``, ``'mean'``, ``'median'``, or |
|
24
|
|
|
``'percentile'``. If ``'percentile'`` is used, the ``percentile`` |
|
25
|
|
|
argument must also be supplied. |
|
26
|
|
|
percentile: Percetile to use for intensity projections. This argument |
|
27
|
|
|
is required if ``projection_type`` is ``'percentile'`` and is |
|
28
|
|
|
silently ignored otherwise. |
|
29
|
|
|
full_slabs_only: Boolean. Should projections be done only for slabs |
|
30
|
|
|
that are ``slab_thickness`` thick? Default is ``True``. |
|
31
|
|
|
If ``False``, some slabs may not be ``slab_thickness`` thick |
|
32
|
|
|
depending on the size of the image, slab thickness, and stride. |
|
33
|
|
|
|
|
34
|
|
|
Example: |
|
35
|
|
|
>>> import torchio as tio |
|
36
|
|
|
>>> ct = tio.datasets.Slicer('CTChest').CT_chest |
|
37
|
|
|
>>> axial_mip = tio.SlabProjection("S", slab_thickness=20) |
|
38
|
|
|
>>> ct_t = axial_mip(ct) |
|
39
|
|
|
>>> ct_t.plot() |
|
40
|
|
|
|
|
41
|
|
|
.. plot:: |
|
42
|
|
|
|
|
43
|
|
|
import torchio as tio |
|
44
|
|
|
sub = tio.datasets.Slicer('CTChest') |
|
45
|
|
|
ct = sub.CT_chest |
|
46
|
|
|
axial_mip = tio.SlabProjection("S", slab_thickness=20) |
|
47
|
|
|
ct_mip = axial_mip(ct) |
|
48
|
|
|
sub.add_image(ct_mip, 'CT_MIP') |
|
49
|
|
|
sub = tio.Clamp(-1000, 1000)(sub) |
|
50
|
|
|
sub.plot() |
|
51
|
|
|
|
|
52
|
|
|
""" |
|
53
|
|
|
def __init__( |
|
54
|
|
|
self, |
|
55
|
|
|
axis: str, |
|
56
|
|
|
slab_thickness: Optional[int] = None, |
|
57
|
|
|
stride: int = 1, |
|
58
|
|
|
projection_type: str = 'max', |
|
59
|
|
|
percentile: Optional[float] = None, |
|
60
|
|
|
full_slabs_only: bool = True, |
|
61
|
|
|
**kwargs |
|
62
|
|
|
): |
|
63
|
|
|
super().__init__(**kwargs) |
|
64
|
|
|
self.args_names = ( |
|
65
|
|
|
'axis', 'slab_thickness', 'stride', |
|
66
|
|
|
'projection_type', 'percentile', 'full_slabs_only' |
|
67
|
|
|
) |
|
68
|
|
|
self.axis = axis |
|
69
|
|
|
self.slab_thickness = slab_thickness |
|
70
|
|
|
self.stride = stride |
|
71
|
|
|
self.projection_type = projection_type |
|
72
|
|
|
self.percentile = self.validate_percentile(percentile) |
|
73
|
|
|
self.full_slabs_only = full_slabs_only |
|
74
|
|
|
self.projection_fun = self.get_projection_function() |
|
75
|
|
|
|
|
76
|
|
|
def validate_percentile(self, percentile): |
|
77
|
|
|
if not self.projection_type == 'percentile': |
|
78
|
|
|
return percentile |
|
79
|
|
|
message = ( |
|
80
|
|
|
"For projection_type='percentile', `percentile` must be a scalar" |
|
81
|
|
|
f' value in the range [0, 1], not {percentile}.' |
|
82
|
|
|
) |
|
83
|
|
|
if percentile is None: |
|
84
|
|
|
raise ValueError(message) |
|
85
|
|
|
elif 0 <= percentile <= 100: |
|
86
|
|
|
return percentile / 100 |
|
87
|
|
|
else: |
|
88
|
|
|
raise ValueError(message) |
|
89
|
|
|
|
|
90
|
|
|
def get_projection_function(self): |
|
91
|
|
|
if self.projection_type == 'max': |
|
92
|
|
|
projection_fun = torch.amax |
|
93
|
|
|
elif self.projection_type == 'min': |
|
94
|
|
|
projection_fun = torch.amin |
|
95
|
|
|
elif self.projection_type == 'mean': |
|
96
|
|
|
projection_fun = torch.mean |
|
97
|
|
|
elif self.projection_type == 'median': |
|
98
|
|
|
projection_fun = torch.median |
|
99
|
|
|
elif self.projection_type == 'percentile': |
|
100
|
|
|
projection_fun = torch.quantile |
|
101
|
|
|
else: |
|
102
|
|
|
message = ( |
|
103
|
|
|
'`projection_type` must be one of "max", "min", "mean",' |
|
104
|
|
|
' "median", or "percentile".' |
|
105
|
|
|
) |
|
106
|
|
|
raise ValueError(message) |
|
107
|
|
|
return projection_fun |
|
108
|
|
|
|
|
109
|
|
|
def get_num_slabs(self): |
|
110
|
|
|
if self.full_slabs_only: |
|
111
|
|
|
start_index = 0 |
|
112
|
|
|
num_slabs = 0 |
|
113
|
|
|
while start_index + self.slab_thickness <= self.axis_span: |
|
114
|
|
|
num_slabs += 1 |
|
115
|
|
|
start_index += self.stride |
|
116
|
|
|
else: |
|
117
|
|
|
num_slabs = torch.ceil(torch.tensor(self.axis_span) / self.stride) |
|
118
|
|
|
num_slabs = int(num_slabs.item()) |
|
119
|
|
|
return num_slabs |
|
120
|
|
|
|
|
121
|
|
|
def apply_transform(self, subject: Subject) -> Subject: |
|
122
|
|
|
for image in self.get_images(subject): |
|
123
|
|
|
self.apply_projection(image) |
|
124
|
|
|
return subject |
|
125
|
|
|
|
|
126
|
|
|
def apply_projection(self, image: ScalarImage) -> None: |
|
127
|
|
|
self.axis_index = image.axis_name_to_index(self.axis) |
|
128
|
|
|
self.axis_span = image.shape[self.axis_index] |
|
129
|
|
|
if self.slab_thickness is None: |
|
130
|
|
|
self.slab_thickness = self.axis_span |
|
131
|
|
|
elif self.slab_thickness > self.axis_span: |
|
132
|
|
|
self.slab_thickness = self.axis_span |
|
133
|
|
|
image.set_data(self.projection(image.data)) |
|
134
|
|
|
|
|
135
|
|
|
def projection(self, tensor: torch.Tensor) -> torch.Tensor: |
|
136
|
|
|
if self.projection_type in ['mean', 'percentile']: |
|
137
|
|
|
tensor = tensor.to(torch.float) |
|
138
|
|
|
|
|
139
|
|
|
num_slabs = self.get_num_slabs() |
|
140
|
|
|
|
|
141
|
|
|
slabs = [] |
|
142
|
|
|
start_index = 0 |
|
143
|
|
|
end_index = start_index + self.slab_thickness |
|
144
|
|
|
|
|
145
|
|
|
for _ in range(num_slabs): |
|
146
|
|
|
slab_indices = torch.arange(start_index, end_index) |
|
147
|
|
|
slab = tensor.index_select(self.axis_index, slab_indices) |
|
148
|
|
|
if self.projection_type == 'median': |
|
149
|
|
|
projected, _ = self.projection_fun( |
|
150
|
|
|
slab, dim=self.axis_index, keepdim=True) |
|
151
|
|
|
elif self.projection_type == 'percentile': |
|
152
|
|
|
projected = self.projection_fun( |
|
153
|
|
|
slab, q=self.percentile, dim=self.axis_index, |
|
154
|
|
|
keepdim=True) |
|
155
|
|
|
else: |
|
156
|
|
|
projected = self.projection_fun( |
|
157
|
|
|
slab, dim=self.axis_index, keepdim=True) |
|
158
|
|
|
slabs.append(projected) |
|
159
|
|
|
start_index += self.stride |
|
160
|
|
|
end_index = start_index + self.slab_thickness |
|
161
|
|
|
if end_index > self.axis_span: |
|
162
|
|
|
end_index = self.axis_span |
|
163
|
|
|
|
|
164
|
|
|
return torch.cat(slabs, dim=self.axis_index) |
|
165
|
|
|
|
fepegar /
torchio
