|
1
|
|
|
from pathlib import Path |
|
2
|
|
|
from typing import Optional |
|
3
|
|
|
|
|
4
|
|
|
from ..typing import TypePath |
|
5
|
|
|
from ..transforms import Transform |
|
6
|
|
|
from ..download import download_and_extract_archive |
|
7
|
|
|
from .. import SubjectsDataset, Subject, ScalarImage, LabelMap |
|
8
|
|
|
|
|
9
|
|
|
|
|
10
|
|
|
class EPISURG(SubjectsDataset): |
|
11
|
|
|
""" |
|
12
|
|
|
`EPISURG`_ is a clinical dataset of :math:`T_1`-weighted MRI from 430 |
|
13
|
|
|
epileptic patients who underwent resective brain surgery at the National |
|
14
|
|
|
Hospital of Neurology and Neurosurgery (Queen Square, London, United |
|
15
|
|
|
Kingdom) between 1990 and 2018. |
|
16
|
|
|
|
|
17
|
|
|
The dataset comprises 430 postoperative MRI. The corresponding preoperative |
|
18
|
|
|
MRI is present for 268 subjects. |
|
19
|
|
|
|
|
20
|
|
|
Three human raters segmented the resection cavity on partially overlapping |
|
21
|
|
|
subsets of EPISURG. |
|
22
|
|
|
|
|
23
|
|
|
If you use this dataset for your research, you agree with the *Data use |
|
24
|
|
|
agreement* presented at the EPISURG entry on the `UCL Research Data |
|
25
|
|
|
Repository <EPISURG>`_ and you must cite the corresponding publications. |
|
26
|
|
|
|
|
27
|
|
|
.. _EPISURG: https://doi.org/10.5522/04/9996158.v1 |
|
28
|
|
|
|
|
29
|
|
|
Args: |
|
30
|
|
|
root: Root directory to which the dataset will be downloaded. |
|
31
|
|
|
transform: An instance of |
|
32
|
|
|
:class:`~torchio.transforms.transform.Transform`. |
|
33
|
|
|
download: If set to ``True``, will download the data into :attr:`root`. |
|
34
|
|
|
|
|
35
|
|
|
.. warning:: The size of this dataset is multiple GB. |
|
36
|
|
|
If you set :attr:`download` to ``True``, it will take some time |
|
37
|
|
|
to be downloaded if it is not already present. |
|
38
|
|
|
""" |
|
39
|
|
|
|
|
40
|
|
|
data_url = 'https://s3-eu-west-1.amazonaws.com/pstorage-ucl-2748466690/26153588/EPISURG.zip' # noqa: E501 |
|
41
|
|
|
md5 = '5ec5831a2c6fbfdc8489ba2910a6504b' |
|
42
|
|
|
|
|
43
|
|
|
def __init__( |
|
44
|
|
|
self, |
|
45
|
|
|
root: TypePath, |
|
46
|
|
|
transform: Optional[Transform] = None, |
|
47
|
|
|
download: bool = False, |
|
48
|
|
|
**kwargs, |
|
49
|
|
|
): |
|
50
|
|
|
root = Path(root).expanduser().absolute() |
|
51
|
|
|
if download: |
|
52
|
|
|
self._download(root) |
|
53
|
|
|
subjects_list = self._get_subjects_list(root) |
|
54
|
|
|
self.kwargs = kwargs |
|
55
|
|
|
super().__init__(subjects_list, transform=transform, **kwargs) |
|
56
|
|
|
|
|
57
|
|
|
@staticmethod |
|
58
|
|
|
def _check_exists(root, modalities): |
|
59
|
|
|
for modality in modalities: |
|
60
|
|
|
modality_dir = root / modality |
|
61
|
|
|
if not modality_dir.is_dir(): |
|
62
|
|
|
exists = False |
|
63
|
|
|
break |
|
64
|
|
|
else: |
|
65
|
|
|
exists = True |
|
66
|
|
|
return exists |
|
67
|
|
|
|
|
68
|
|
|
@staticmethod |
|
69
|
|
|
def _get_subjects_list(root): |
|
70
|
|
|
subjects_dir = root / 'EPISURG' / 'subjects' |
|
71
|
|
|
subjects = [] |
|
72
|
|
|
for subject_dir in sorted(subjects_dir.glob('sub-*')): |
|
73
|
|
|
subject_id = subject_dir.name[-4:] |
|
74
|
|
|
images_dict = {'subject_id': subject_id} |
|
75
|
|
|
preop_dir = subject_dir / 'preop' |
|
76
|
|
|
preop_paths = list(preop_dir.glob('*preop*')) |
|
77
|
|
|
assert len(preop_paths) <= 1 |
|
78
|
|
|
if preop_paths: |
|
79
|
|
|
images_dict['preop_mri'] = ScalarImage(preop_paths[0]) |
|
80
|
|
|
postop_dir = subject_dir / 'postop' |
|
81
|
|
|
postop_path = list(postop_dir.glob('*postop-t1mri*'))[0] |
|
82
|
|
|
images_dict['postop_mri'] = ScalarImage(postop_path) |
|
83
|
|
|
for seg_path in postop_dir.glob('*seg*'): |
|
84
|
|
|
seg_id = seg_path.name[-8] |
|
85
|
|
|
images_dict[f'seg_{seg_id}'] = LabelMap(seg_path) |
|
86
|
|
|
subjects.append(Subject(**images_dict)) |
|
87
|
|
|
return subjects |
|
88
|
|
|
|
|
89
|
|
|
def _download(self, root): |
|
90
|
|
|
"""Download the EPISURG data if it does not exist already.""" |
|
91
|
|
|
if (root / 'EPISURG').is_dir(): |
|
92
|
|
|
return |
|
93
|
|
|
root.mkdir(exist_ok=True, parents=True) |
|
94
|
|
|
download_and_extract_archive( |
|
95
|
|
|
self.data_url, |
|
96
|
|
|
download_root=root, |
|
97
|
|
|
md5=self.md5, |
|
98
|
|
|
) |
|
99
|
|
|
(root / 'EPISURG.zip').unlink() # cleanup |
|
100
|
|
|
|
|
101
|
|
|
def _glob_subjects(self, string): |
|
102
|
|
|
subjects = [] |
|
103
|
|
|
for subject in self._subjects: |
|
104
|
|
|
for image_name in subject: |
|
105
|
|
|
if string in image_name: |
|
106
|
|
|
subjects.append(subject) |
|
107
|
|
|
break |
|
108
|
|
|
return subjects |
|
109
|
|
|
|
|
110
|
|
|
def _get_labeled_subjects(self): |
|
111
|
|
|
return self._glob_subjects('seg') |
|
112
|
|
|
|
|
113
|
|
|
def _get_paired_subjects(self): |
|
114
|
|
|
return self._glob_subjects('preop') |
|
115
|
|
|
|
|
116
|
|
|
def _get_subset(self, subjects): |
|
117
|
|
|
dataset = SubjectsDataset( |
|
118
|
|
|
subjects, |
|
119
|
|
|
transform=self._transform, |
|
120
|
|
|
**(self.kwargs), |
|
121
|
|
|
) |
|
122
|
|
|
return dataset |
|
123
|
|
|
|
|
124
|
|
|
def get_labeled(self) -> SubjectsDataset: |
|
125
|
|
|
"""Get dataset from subjects with manual annotations.""" |
|
126
|
|
|
return self._get_subset(self._get_labeled_subjects()) |
|
127
|
|
|
|
|
128
|
|
|
def get_unlabeled(self) -> SubjectsDataset: |
|
129
|
|
|
"""Get dataset from subjects without manual annotations.""" |
|
130
|
|
|
subjects = [ |
|
131
|
|
|
s for s in self._subjects |
|
132
|
|
|
if s not in self._get_labeled_subjects() |
|
133
|
|
|
] |
|
134
|
|
|
return self._get_subset(subjects) |
|
135
|
|
|
|
|
136
|
|
|
def get_paired(self) -> SubjectsDataset: |
|
137
|
|
|
"""Get dataset from subjects with pre- and post-op MRI.""" |
|
138
|
|
|
return self._get_subset(self._get_paired_subjects()) |
|
139
|
|
|
|
fepegar /
torchio
