1
|
|
|
import csv |
2
|
|
|
from typing import List |
3
|
|
|
from pathlib import Path |
4
|
|
|
|
5
|
|
|
from ..typing import TypePath |
6
|
|
|
from .. import SubjectsDataset, Subject, ScalarImage |
7
|
|
|
|
8
|
|
|
|
9
|
|
|
class RSNAMICCAI(SubjectsDataset): |
10
|
|
|
"""RSNA-MICCAI Brain Tumor Radiogenomic Classification challenge dataset. |
11
|
|
|
|
12
|
|
|
This is a helper class for the dataset used in the |
13
|
|
|
`RSNA-MICCAI Brain Tumor Radiogenomic Classification challenge`_ hosted on |
14
|
|
|
`kaggle <https://www.kaggle.com/>`_. The dataset must be downloaded before |
15
|
|
|
instantiating this class (as oposed to, e.g., :class:`torchio.datasets.IXI`). |
16
|
|
|
|
17
|
|
|
If you reference or use the dataset in any form, include the following |
18
|
|
|
citation: |
19
|
|
|
|
20
|
|
|
U.Baid, et al., "The RSNA-ASNR-MICCAI BraTS 2021 Benchmark on Brain Tumor |
21
|
|
|
Segmentation and Radiogenomic Classification", arXiv:2107.02314, 2021. |
22
|
|
|
|
23
|
|
|
Args: |
24
|
|
|
root_dir: Directory containing the dataset (``train`` directory, |
25
|
|
|
``test`` directory, etc.). |
26
|
|
|
train: If ``True``, the training set will be used. Otherwise the |
27
|
|
|
validation set will be used. |
28
|
|
|
ignore_empty: If ``True``, the three subjects flagged as "presenting |
29
|
|
|
issues" (empty images) by the challenge organizers will be ignored. |
30
|
|
|
|
31
|
|
|
.. _RSNA-MICCAI Brain Tumor Radiogenomic Classification challenge: https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification |
32
|
|
|
""" |
33
|
|
|
id_key = 'BraTS21ID' |
34
|
|
|
label_key = 'MGMT_value' |
35
|
|
|
modalities = 'T1w', 'T1wCE', 'T2w', 'FLAIR' |
36
|
|
|
bad_subjects = '00109', '00123', '00709' |
37
|
|
|
|
38
|
|
|
def __init__( |
39
|
|
|
self, |
40
|
|
|
root_dir: TypePath, |
41
|
|
|
train: bool = True, |
42
|
|
|
ignore_empty: bool = True, |
43
|
|
|
**kwargs, |
44
|
|
|
): |
45
|
|
|
self.root_dir = Path(root_dir).expanduser().resolve() |
46
|
|
|
subjects = self._get_subjects(self.root_dir, train, ignore_empty) |
47
|
|
|
super().__init__(subjects, **kwargs) |
48
|
|
|
self.train = train |
49
|
|
|
|
50
|
|
|
def _get_subjects( |
51
|
|
|
self, |
52
|
|
|
root_dir: Path, |
53
|
|
|
train: bool, |
54
|
|
|
ignore_empty: bool, |
55
|
|
|
) -> List[Subject]: |
56
|
|
|
subjects = [] |
57
|
|
|
if train: |
58
|
|
|
csv_path = root_dir / 'train_labels.csv' |
59
|
|
|
with open(csv_path) as csvfile: |
60
|
|
|
reader = csv.DictReader(csvfile) |
61
|
|
|
labels_dict = { |
62
|
|
|
row[self.id_key]: int(row[self.label_key]) |
63
|
|
|
for row in reader |
64
|
|
|
} |
65
|
|
|
subjects_dir = root_dir / 'train' |
66
|
|
|
else: |
67
|
|
|
subjects_dir = root_dir / 'test' |
68
|
|
|
|
69
|
|
|
for subject_dir in sorted(subjects_dir.iterdir()): |
70
|
|
|
subject_id = subject_dir.name |
71
|
|
|
if ignore_empty and subject_id in self.bad_subjects: |
72
|
|
|
continue |
73
|
|
|
try: |
74
|
|
|
int(subject_id) |
75
|
|
|
except ValueError: |
76
|
|
|
continue |
77
|
|
|
images_dict = {self.id_key: subject_dir.name} |
78
|
|
|
if train: |
79
|
|
|
images_dict[self.label_key] = labels_dict[subject_id] |
|
|
|
|
80
|
|
|
for modality in self.modalities: |
81
|
|
|
image_dir = subject_dir / modality |
82
|
|
|
filepaths = list(image_dir.iterdir()) |
83
|
|
|
num_files = len(filepaths) |
84
|
|
|
path = filepaths[0] if num_files == 1 else image_dir |
85
|
|
|
images_dict[modality] = ScalarImage(path) |
86
|
|
|
subject = Subject(images_dict) |
87
|
|
|
subjects.append(subject) |
88
|
|
|
return subjects |
89
|
|
|
|