Coverage for src/mednet/config/data/nih_cxr14/datamodule.py: 79%
39 statements
« prev ^ index » next coverage.py v7.4.2, created at 2024-04-30 11:44 +0200
1# SPDX-FileCopyrightText: Copyright © 2023 Idiap Research Institute <contact@idiap.ch>
2#
3# SPDX-License-Identifier: GPL-3.0-or-later
4"""NIH CXR14 (relabeled) DataModule for computer-aided diagnosis.
6Database reference: [NIH-CXR14-2017]_
7"""
9import importlib.resources
10import os
11import pathlib
13import PIL.Image
14from torchvision.transforms.functional import to_tensor
16from ....data.datamodule import CachingDataModule
17from ....data.split import JSONDatabaseSplit
18from ....data.typing import DatabaseSplit, Sample
19from ....data.typing import RawDataLoader as _BaseRawDataLoader
20from ....utils.rc import load_rc
22CONFIGURATION_KEY_DATADIR = "datadir." + (__name__.rsplit(".", 2)[-2])
23"""Key to search for in the configuration file for the root directory of this
24database.
25"""
27CONFIGURATION_KEY_IDIAP_FILESTRUCTURE = (
28 (__name__.rsplit(".", 2)[-2]) + ".idiap_folder_structure"
29)
30"""Key to search for in the configuration file indicating if the loader should
31use standard or idiap-based file organisation structure.
33It causes the internal loader to search for files in a slightly
34different folder structure, that was adapted to Idiap's requirements
35(number of files per folder to be less than 10k).
36"""
39class RawDataLoader(_BaseRawDataLoader):
40 """A specialized raw-data-loader for the NIH CXR-14 dataset."""
42 datadir: pathlib.Path
43 """This variable contains the base directory where the database raw data is
44 stored."""
46 idiap_file_organisation: bool
47 """If should use the Idiap's filesystem organisation when looking up data.
49 This variable will be ``True``, if the user has set the configuration
50 parameter ``nih_cxr14.idiap_file_organisation`` in the global configuration
51 file. It will cause internal loader to search for files in a slightly
52 different folder structure, that was adapted to Idiap's requirements
53 (number of files per folder to be less than 10k).
54 """
56 def __init__(self):
57 rc = load_rc()
58 self.datadir = pathlib.Path(
59 rc.get(CONFIGURATION_KEY_DATADIR, os.path.realpath(os.curdir)),
60 )
61 self.idiap_file_organisation = rc.get(
62 CONFIGURATION_KEY_IDIAP_FILESTRUCTURE,
63 False,
64 )
66 def sample(self, sample: tuple[str, list[int]]) -> Sample:
67 """Load a single image sample from the disk.
69 Parameters
70 ----------
71 sample
72 A tuple containing the path suffix, within the dataset root folder,
73 where to find the image to be loaded, and an integer, representing
74 the sample label.
76 Returns
77 -------
78 The sample representation.
79 """
81 file_path = pathlib.Path(sample[0]) # default
82 if self.idiap_file_organisation:
83 # for folder lookup efficiency, data is split into subfolders
84 # each original file is on the subfolder `f[:5]/f`, where f
85 # is the original file basename
86 file_path = file_path.parent / file_path.name[:5] / file_path.name
88 # N.B.: some NIH CXR-14 images are encoded as color PNGs with an alpha
89 # channel. Most, are grayscale PNGs
90 image = PIL.Image.open(self.datadir / file_path)
91 image = image.convert("L") # required for some images
92 tensor = to_tensor(image)
94 # use the code below to view generated images
95 # from torchvision.transforms.functional import to_pil_image
96 # to_pil_image(tensor).show()
97 # __import__("pdb").set_trace()
99 return tensor, dict(label=sample[1], name=sample[0]) # type: ignore[arg-type]
101 def label(self, sample: tuple[str, list[int]]) -> list[int]:
102 """Load a single image sample label from the disk.
104 Parameters
105 ----------
106 sample
107 A tuple containing the path suffix, within the dataset root folder,
108 where to find the image to be loaded, and an integer, representing the
109 sample label.
111 Returns
112 -------
113 list[int]
114 The integer labels associated with the sample.
115 """
117 return sample[1]
120def make_split(basename: str) -> DatabaseSplit:
121 """Return a database split for the NIH CXR-14 database.
123 Parameters
124 ----------
125 basename
126 Name of the .json file containing the split to load.
128 Returns
129 -------
130 An instance of DatabaseSplit.
131 """
133 return JSONDatabaseSplit(
134 importlib.resources.files(__name__.rsplit(".", 1)[0]).joinpath(
135 basename,
136 ),
137 )
140class DataModule(CachingDataModule):
141 """NIH CXR14 (relabeled) DataModule for computer-aided diagnosis.
143 This dataset was extracted from the clinical PACS database at the National
144 Institutes of Health Clinical Center (USA) and represents 60% of all their
145 radiographs. It contains labels for 14 common radiological signs in this
146 order: cardiomegaly, emphysema, effusion, hernia, infiltration, mass,
147 nodule, atelectasis, pneumothorax, pleural thickening, pneumonia, fibrosis,
148 edema and consolidation. This is the relabeled version created in the
149 CheXNeXt study.
151 * Reference: [NIH-CXR14-2017]_
152 * Raw data input (on disk):
154 * PNG RGB 8-bit depth images
155 * Resolution: 1024 x 1024 pixels
157 * Labels: [CHEXNEXT-2018]_
158 * Split reference: [CHEXNEXT-2018]_
159 * Output image:
161 * Transforms:
163 * Load raw PNG with :py:mod:`PIL`
164 * Convert to torch tensor
166 * Final specifications:
168 * RGB, encoded as a 3-plane tensor, 32-bit floats, square
169 (1024x1024 px)
170 * Labels in order:
171 * cardiomegaly
172 * emphysema
173 * effusion
174 * hernia
175 * infiltration
176 * mass
177 * nodule
178 * atelectasis
179 * pneumothorax
180 * pleural thickening
181 * pneumonia
182 * fibrosis
183 * edema
184 * consolidation
186 Parameters
187 ----------
188 split_filename
189 Name of the .json file containing the split to load.
190 """
192 def __init__(self, split_filename: str):
193 super().__init__(
194 database_split=make_split(split_filename),
195 raw_data_loader=RawDataLoader(),
196 database_name=__package__.split(".")[-1],
197 split_name=pathlib.Path(split_filename).stem,
198 )