Yuhan Hou commited on
Commit
a500017
1 Parent(s): 02ffd0b
Files changed (1) hide show
  1. FracAtlas_dataset.py +257 -0
FracAtlas_dataset.py ADDED
@@ -0,0 +1,257 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ # -*- coding: utf-8 -*-
3
+ """
4
+ Created on Sun Feb 18 23:13:51 2024
5
+
6
+ @author: houyuhan
7
+ """
8
+
9
+
10
+ #Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
11
+ #
12
+ # Licensed under the Apache License, Version 2.0 (the "License");
13
+ # you may not use this file except in compliance with the License.
14
+ # You may obtain a copy of the License at
15
+ #
16
+ # http://www.apache.org/licenses/LICENSE-2.0
17
+ #
18
+ # Unless required by applicable law or agreed to in writing, software
19
+ # distributed under the License is distributed on an "AS IS" BASIS,
20
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
21
+ # See the License for the specific language governing permissions and
22
+ # limitations under the License.
23
+ """
24
+ FracAtlas Dataset Loader
25
+
26
+ This script provides a Hugging Face `datasets` loader for the FracAtlas dataset, a comprehensive collection
27
+ of musculoskeletal radiographs aimed at advancing research in fracture classification, localization, and segmentation.
28
+ The dataset includes high-quality X-Ray images accompanied by detailed annotations in COCO JSON format for segmentation
29
+ and bounding box information, as well as PASCAL VOC XML files for additional localization data.
30
+
31
+ The loader handles downloading and preparing the dataset, making it readily available for machine learning models and analysis
32
+ tasks in medical imaging, especially focusing on the detection and understanding of bone fractures.
33
+
34
+ License: CC-BY 4.0
35
+ """
36
+
37
+
38
+ import csv
39
+ import json
40
+ import os
41
+ from typing import List
42
+ import datasets
43
+ import logging
44
+ import pandas as pd
45
+ from sklearn.model_selection import train_test_split
46
+ import shutil
47
+ import xml.etree.ElementTree as ET
48
+ from datasets import load_dataset
49
+
50
+
51
+
52
+ # TODO: Add BibTeX citation
53
+ # Find for instance the citation on arxiv or on the dataset repo/website
54
+ _CITATION = """\
55
+ @InProceedings{huggingface:yh0701/FracAtlas_dataset,
56
+ title = {FracAtlas: A Dataset for Fracture Classification, Localization and Segmentation of Musculoskeletal Radiographs},
57
+ author={Abedeen, Iftekharul; Rahman, Md. Ashiqur; Zohra Prottyasha, Fatema; Ahmed, Tasnim; Mohmud Chowdhury, Tareque; Shatabda, Swakkhar},
58
+ year={2023}
59
+ }
60
+ """
61
+
62
+ # TODO: Add description of the dataset here
63
+ # You can copy an official description
64
+ _DESCRIPTION = """\
65
+ The "FracAtlas" dataset is a collection of musculoskeletal radiographs for fracture classification, localization, and segmentation.
66
+ It includes 4,083 X-Ray images with annotations in multiple formats.The annotations include bbox, segmentations, and etc.
67
+ The dataset is intended for use in deep learning tasks in medical imaging, specifically targeting the understanding of bone fractures.
68
+ It is freely available under a CC-BY 4.0 license.
69
+ """
70
+
71
+ # TODO: Add a link to an official homepage for the dataset here
72
+ _HOMEPAGE = "https://figshare.com/articles/dataset/The_dataset/22363012"
73
+
74
+ # TODO: Add the licence for the dataset here if you can find it
75
+ _LICENSE = "The dataset is licensed under a CC-BY 4.0 license."
76
+
77
+ # TODO: Add link to the official dataset URLs here
78
+ # The HuggingFace Datasets library doesn't host the datasets but only points to the original files.
79
+ # This can be an arbitrary nested dict/list of URLs (see below in `_split_generators` method)
80
+ _URL = "https://figshare.com/ndownloader/files/43283628"
81
+
82
+ # TODO: Name of the dataset usually matches the script name with CamelCase instead of snake_case
83
+ class FracAtlasDataset(datasets.GeneratorBasedBuilder):
84
+ """TODO: Short description of my dataset."""
85
+
86
+ _URL = _URL
87
+ VERSION = datasets.Version("1.1.0")
88
+
89
+ def _info(self):
90
+ return datasets.DatasetInfo(
91
+ description=_DESCRIPTION,
92
+ features=datasets.Features(
93
+ {
94
+ "image_id": datasets.Value("string"),
95
+ "image": datasets.Image(),
96
+ "hand": datasets.ClassLabel(num_classes=2,names=['no_hand','hand']),
97
+ "leg": datasets.ClassLabel(num_classes=2,names=['no_leg','leg']),
98
+ "hip": datasets.ClassLabel(num_classes=2,names=['no_hip','hip']),
99
+ "shoulder": datasets.ClassLabel(num_classes=2,names=['no_shoulder','shoulder']),
100
+ "mixed": datasets.ClassLabel(num_classes=2,names=['not_mixed','mixed']),
101
+ "hardware": datasets.ClassLabel(num_classes=2,names=['no_hardware','hardware']),
102
+ "multiscan": datasets.ClassLabel(num_classes=2,names=['not_multiscan','multiscan']),
103
+ "fractured": datasets.ClassLabel(num_classes=2,names=['not_fractured','fractured']),
104
+ "fracture_count": datasets.Value("int32"),
105
+ "frontal": datasets.ClassLabel(num_classes=2,names=['not_frontal','frontal']),
106
+ "lateral": datasets.ClassLabel(num_classes=2,names=['not_lateral','lateral']),
107
+ "oblique": datasets.ClassLabel(num_classes=2,names=['not_oblique','oblique']),
108
+ "localization_metadata": datasets.Features({
109
+ "width": datasets.Value("int32"),
110
+ "height": datasets.Value("int32"),
111
+ "depth": datasets.Value("int32"),
112
+ }),
113
+ "segmentation_metadata": datasets.Features({
114
+ "segmentation": datasets.Sequence(datasets.Sequence(datasets.Value("float"))),
115
+ "bbox": datasets.Sequence(datasets.Value("float")),
116
+ "area": datasets.Value("float")
117
+ }) or None
118
+ }
119
+ ),
120
+ # No default supervised_keys (as we have to pass both question
121
+ # and context as input).
122
+ supervised_keys=None,
123
+ homepage=_HOMEPAGE,
124
+ citation=_CITATION
125
+ )
126
+
127
+ def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
128
+ url_to_download = self._URL
129
+ downloaded_files = dl_manager.download_and_extract(url_to_download)
130
+
131
+ # Adjusted path to include 'FracAtlas' directory
132
+ base_path = os.path.join(downloaded_files, 'FracAtlas')
133
+
134
+ # Split the dataset to train/test/validation by 0.7,0.15,0.15
135
+ df = pd.read_csv(os.path.join(base_path, 'dataset.csv'))
136
+ train_df, test_df = train_test_split(df, test_size=0.3)
137
+ validation_df, test_df = train_test_split(test_df, test_size=0.5)
138
+
139
+ # store them back as csv
140
+ train_df.to_csv(os.path.join(base_path, 'train_dataset.csv'), index=False)
141
+ validation_df.to_csv(os.path.join(base_path, 'validation_dataset.csv'), index=False)
142
+ test_df.to_csv(os.path.join(base_path, 'test_dataset.csv'), index=False)
143
+
144
+ annotations_path = os.path.join(base_path, 'Annotations/COCO JSON/COCO_fracture_masks.json')
145
+ images_path = os.path.join(base_path, 'images')
146
+ localization_path = os.path.join(base_path, 'Annotations/PASCAL VOC')
147
+
148
+ return [
149
+ datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"dataset_csv_path": os.path.join(base_path, 'train_dataset.csv'),
150
+ "images_path": images_path,
151
+ "annotations_path": annotations_path,
152
+ "localization_path":localization_path
153
+ }),
154
+ datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"dataset_csv_path": os.path.join(base_path, 'validation_dataset.csv'),
155
+ "images_path": images_path,
156
+ "annotations_path": annotations_path,
157
+ "localization_path":localization_path
158
+ }),
159
+ datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"dataset_csv_path": os.path.join(base_path, 'test_dataset.csv'),
160
+ "images_path": images_path,
161
+ "annotations_path": annotations_path,
162
+ "localization_path":localization_path
163
+ })
164
+ ]
165
+
166
+ def _generate_examples(self, annotations_path, images_path, dataset_csv_path,localization_path):
167
+ logging.info("Generating examples from = %s", dataset_csv_path)
168
+ split_df = pd.read_csv(dataset_csv_path) # Load the DataFrame for the current split
169
+
170
+ # Function to convert numeric ID to formatted string
171
+ def format_image_id(numeric_id):
172
+ return f"IMG{numeric_id:07d}.jpg" # Adjust format as needed
173
+
174
+ # Function to extract information from xml files
175
+ def parse_xml(xml_path):
176
+ tree = ET.parse(xml_path)
177
+ root = tree.getroot()
178
+
179
+ # Extract the necessary information
180
+ width = int(root.find("./size/width").text)
181
+ height = int(root.find("./size/height").text)
182
+ depth = int(root.find("./size/depth").text)
183
+ segmented = int(root.find("./segmented").text)
184
+ return width, height, depth, segmented
185
+
186
+ # Load annotations
187
+ with open(annotations_path) as file:
188
+ annotations_json = json.load(file)
189
+
190
+ for item in annotations_json['annotations']:
191
+ item['image_id'] = format_image_id(item['image_id'])
192
+
193
+ annotations = {item['image_id']: item for item in annotations_json['annotations']}
194
+
195
+
196
+ # Iterate through each row in the split DataFrame
197
+ for _, row in split_df.iterrows():
198
+ image_id = row['image_id']
199
+ # Determine the folder based on the 'fractured' column
200
+ folder = 'Fractured' if row['fractured'] == 1 else 'Non_fractured'
201
+
202
+ # Check if the formatted_image_id exists in annotations
203
+ annotation = annotations.get(image_id)
204
+ image_path = os.path.join(images_path, folder, image_id)
205
+
206
+ # Initialize variables
207
+ segmentation, bbox, area = None, None, None
208
+ segmentation_metadata = None
209
+
210
+ if annotation:
211
+ segmentation = annotation.get('segmentation')
212
+ bbox = annotation.get('bbox')
213
+ area = annotation.get('area')
214
+
215
+ segmentation_metadata = {
216
+ 'segmentation': segmentation,
217
+ 'bbox':bbox,
218
+ 'area': area
219
+ }
220
+ else:
221
+ segmentation_metadata = None # Default if not present
222
+
223
+ xml_file_name = f"{image_id.split('.')[0]}.xml"
224
+ xml_path = os.path.join(localization_path, xml_file_name)
225
+
226
+ # Parse the XML file
227
+ width, height, depth, _ = parse_xml(xml_path)
228
+
229
+ localization_metadata = {
230
+ 'width': width,
231
+ "height":height,
232
+ 'depth': depth
233
+ }
234
+
235
+
236
+ # Construct example data
237
+ example_data = {
238
+ "image_id": row['image_id'],
239
+ "image":image_path,
240
+ "hand": row["hand"],
241
+ "leg": row["leg"],
242
+ "hip": row["hip"],
243
+ "shoulder": row["shoulder"],
244
+ "mixed": row["mixed"],
245
+ "hardware": row["hardware"],
246
+ "multiscan": row["multiscan"],
247
+ "fractured": row["fractured"],
248
+ "fracture_count": row["fracture_count"],
249
+ "frontal": row["frontal"],
250
+ "lateral": row["lateral"],
251
+ "oblique": row["oblique"],
252
+ "localization_metadata": localization_metadata,
253
+ 'segmentation_metadata': segmentation_metadata
254
+ }
255
+ yield image_id, example_data
256
+
257
+