FracAtlas: A Dataset for Fracture Classification, Localization and Segmentation of Musculoskeletal Radiographs

FracAtlas is a dataset for semantic segmentation, object detection, and classification tasks.

The majority of the existing X-Ray datasets are either very small in size to be used in training machine learning models or lack proper annotation to be used in localization and segmentation. This creates a limitation while comparing different state-of-the-art (SOTA) methods as each study uses its own set of data that is not made public. It also hinders the development of machine-learning algorithms for classification, localization and segmentation. In this work, we introduce a new dataset of more than 14,000 X-Ray scans which was collected from 3 major hospitals in Bangladesh. From this raw data, we isolated 4,083 images which have been manually annotated for classification, localization and segmentation of bone fractures with the help of 2 expert radiologists and later validated by a medical officer. Each scan in the dataset was independently labeled by both radiologists and later confirmed by the medical officer. The annotation was done with the help of makesense.ai, an open-source labeling platform. All the scans are available in JPG format along with proper annotations.\

Citation -
@article{Abedeen_2023,
title = {{FracAtlas}: A Dataset for Fracture Classification, Localization and Segmentation of Musculoskeletal Radiographs},
author = {Iftekharul Abedeen and Md. Ashiqur Rahman and Fatema Zohra Prottyasha and Tasnim Ahmed and Tareque Mohmud Chowdhury and Swakkhar Shatabda},
year = 2023,
month = {aug},
journal = {Scientific Data},
publisher = {Springer Science and Business Media {LLC}},
volume = 10,
number = 1,
doi = {10.1038/s41597-023-02432-4},
url = {https://doi.org/10.1038%2Fs41597-023-02432-4}
}