|
1
|
|
|
import csv |
|
2
|
|
|
import warnings |
|
3
|
|
|
from pathlib import Path |
|
4
|
|
|
from typing import List, Sequence |
|
5
|
|
|
|
|
6
|
|
|
from ..typing import TypePath |
|
7
|
|
|
from .. import SubjectsDataset, Subject, ScalarImage |
|
8
|
|
|
|
|
9
|
|
|
|
|
10
|
|
|
class RSNAMICCAI(SubjectsDataset): |
|
11
|
|
|
"""RSNA-MICCAI Brain Tumor Radiogenomic Classification challenge dataset. |
|
12
|
|
|
|
|
13
|
|
|
This is a helper class for the dataset used in the |
|
14
|
|
|
`RSNA-MICCAI Brain Tumor Radiogenomic Classification challenge`_ hosted on |
|
15
|
|
|
`kaggle <https://www.kaggle.com/>`_. The dataset must be downloaded before |
|
16
|
|
|
instantiating this class (as oposed to, e.g., :class:`torchio.datasets.IXI`). |
|
17
|
|
|
|
|
18
|
|
|
This `kaggle kernel <https://www.kaggle.com/fepegar/preprocessing-mri-with-torchio/>`_ |
|
19
|
|
|
includes a usage example including preprocessing of all the scans. |
|
20
|
|
|
|
|
21
|
|
|
If you reference or use the dataset in any form, include the following |
|
22
|
|
|
citation: |
|
23
|
|
|
|
|
24
|
|
|
U.Baid, et al., "The RSNA-ASNR-MICCAI BraTS 2021 Benchmark on Brain Tumor |
|
25
|
|
|
Segmentation and Radiogenomic Classification", arXiv:2107.02314, 2021. |
|
26
|
|
|
|
|
27
|
|
|
Args: |
|
28
|
|
|
root_dir: Directory containing the dataset (``train`` directory, |
|
29
|
|
|
``test`` directory, etc.). |
|
30
|
|
|
train: If ``True``, the ``train`` set will be used. Otherwise the |
|
31
|
|
|
``test`` set will be used. |
|
32
|
|
|
ignore_empty: If ``True``, the three subjects flagged as "presenting |
|
33
|
|
|
issues" (empty images) by the challenge organizers will be ignored. |
|
34
|
|
|
The subject IDs are ``00109``, ``00123`` and ``00709``. |
|
35
|
|
|
|
|
36
|
|
|
Example: |
|
37
|
|
|
>>> import torchio as tio |
|
38
|
|
|
>>> from subprocess import call |
|
39
|
|
|
>>> call('kaggle competitions download -c rsna-miccai-brain-tumor-radiogenomic-classification'.split()) |
|
40
|
|
|
>>> root_dir = 'rsna-miccai-brain-tumor-radiogenomic-classification' |
|
41
|
|
|
>>> train_set = tio.datasets.RSNAMICCAI(root_dir, train=True) |
|
42
|
|
|
>>> test_set = tio.datasets.RSNAMICCAI(root_dir, train=False) |
|
43
|
|
|
>>> len(train_set), len(test_set) |
|
44
|
|
|
(582, 87) |
|
45
|
|
|
|
|
46
|
|
|
|
|
47
|
|
|
.. _RSNA-MICCAI Brain Tumor Radiogenomic Classification challenge: https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification |
|
48
|
|
|
""" # noqa: E501 |
|
49
|
|
|
id_key = 'BraTS21ID' |
|
50
|
|
|
label_key = 'MGMT_value' |
|
51
|
|
|
bad_subjects = '00109', '00123', '00709' |
|
52
|
|
|
|
|
53
|
|
|
def __init__( |
|
54
|
|
|
self, |
|
55
|
|
|
root_dir: TypePath, |
|
56
|
|
|
train: bool = True, |
|
57
|
|
|
ignore_empty: bool = True, |
|
58
|
|
|
modalities: Sequence[str] = ('T1w', 'T1wCE', 'T2w', 'FLAIR'), |
|
59
|
|
|
**kwargs, |
|
60
|
|
|
): |
|
61
|
|
|
self.root_dir = Path(root_dir).expanduser().resolve() |
|
62
|
|
|
if isinstance(modalities, str): |
|
63
|
|
|
modalities = [modalities] |
|
64
|
|
|
self.modalities = modalities |
|
65
|
|
|
subjects = self._get_subjects(self.root_dir, train, ignore_empty) |
|
66
|
|
|
super().__init__(subjects, **kwargs) |
|
67
|
|
|
self.train = train |
|
68
|
|
|
|
|
69
|
|
|
def _get_subjects( |
|
70
|
|
|
self, |
|
71
|
|
|
root_dir: Path, |
|
72
|
|
|
train: bool, |
|
73
|
|
|
ignore_empty: bool, |
|
74
|
|
|
) -> List[Subject]: |
|
75
|
|
|
subjects = [] |
|
76
|
|
|
if train: |
|
77
|
|
|
csv_path = root_dir / 'train_labels.csv' |
|
78
|
|
|
try: |
|
79
|
|
|
with open(csv_path) as csvfile: |
|
80
|
|
|
reader = csv.DictReader(csvfile) |
|
81
|
|
|
labels_dict = { |
|
82
|
|
|
row[self.id_key]: int(row[self.label_key]) |
|
83
|
|
|
for row in reader |
|
84
|
|
|
} |
|
85
|
|
|
except FileNotFoundError: |
|
86
|
|
|
warnings.warn('Labels CSV not found. Ignoring MGMT labels') |
|
87
|
|
|
labels_dict = {} |
|
88
|
|
|
subjects_dir = root_dir / 'train' |
|
89
|
|
|
else: |
|
90
|
|
|
subjects_dir = root_dir / 'test' |
|
91
|
|
|
|
|
92
|
|
|
for subject_dir in sorted(subjects_dir.iterdir()): |
|
93
|
|
|
subject_id = subject_dir.name |
|
94
|
|
|
if ignore_empty and subject_id in self.bad_subjects: |
|
95
|
|
|
continue |
|
96
|
|
|
try: |
|
97
|
|
|
int(subject_id) |
|
98
|
|
|
except ValueError: |
|
99
|
|
|
continue |
|
100
|
|
|
images_dict = {self.id_key: subject_dir.name} |
|
101
|
|
|
if train and labels_dict: |
|
|
|
|
|
|
102
|
|
|
images_dict[self.label_key] = labels_dict[subject_id] |
|
103
|
|
|
for modality in self.modalities: |
|
104
|
|
|
image_dir = subject_dir / modality |
|
105
|
|
|
filepaths = list(image_dir.iterdir()) |
|
106
|
|
|
num_files = len(filepaths) |
|
107
|
|
|
path = filepaths[0] if num_files == 1 else image_dir |
|
108
|
|
|
images_dict[modality] = ScalarImage(path) |
|
109
|
|
|
subject = Subject(images_dict) |
|
110
|
|
|
subjects.append(subject) |
|
111
|
|
|
return subjects |
|
112
|
|
|
|
fepegar /
torchio
Loading history...