Main content

Files | Discussion Wiki | Discussion | Discussion
default Loading...

Home

Menu

Loading wiki pages...

View
Wiki Version:
The original dataset *VerSe 2019* ([https://doi.org/10.1148/ryai.2020190138][1]) - includes 160 CT image series of 141 patients with segmentation masks of 1725 vertebrae. - is split into training (80), validation (40), and test set (40). - was prepared for a vertebral labeling and segmentation challenge hosted at the 2019 International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI). Please read the readme.txt and licence.txt: https://osf.io/pg6s3/ (Creative Commons Attribution-ShareAlike 2.0). Data Structure ------------------- This dataset is available in two different ways: 1. **Image series based (MICCAI):** 160 image series of 141 patients are divided into a training (n=80), validation (n=40), and test (n=40) set as originally published for the MICCAI challenge. 2. **Subject based:** 141 patients holding 160 image series are divided into a training (n=67), validation (n=37), and test (n=37) set. Subject identifiers equal image series identifiers ('verseXXX'). For patient with 2 or 3 image series, new subject identifiers ('verse4XX') were introduced. This is an adaption of the Brain Imaging Data Structure (BIDS; https://bids.neuroimaging.io/). We recommend to use the subject-based format, as this is completely consistent in VerSe 2019 and VerSe 2020. Please note, that the VerSe 2019 cases that were re-used in the VerSe 2020 challenge have to be downloaded from the subject-based VerSe 2019 repository and are not again included in the subject-based VerSe 2020 repository. Additionally, we offer python scripts to easily work with the subject-based data-structure here: https://github.com/anjany/verse Citation --------- Please respect our work, as we spent >2 years for algorithmic development and >2000 working hours for manual corrections of segmentation masks. **By downloading this data you agreed to cite these papers in your work:** 1. Löffler M, Sekuboyina A, Jakob A, Grau AL, Scharr A, Husseini ME, Herbell M, Zimmer C, Baum T, Kirschke JS., A Vertebral Segmentation Dataset with Fracture Grading. Radiology: Artificial Intelligence, 2020 https://doi.org/10.1148/ryai.2020190138. 2. Sekuboyina A. et al., Labelling Vertebrae with 2D Reformations of Multidetector CT Images: An Adversarial Approach for Incorporating Prior Knowledge of Spine Anatomy. Radiology: Artificial Intelligence, 2020, https://doi.org/10.1148/ryai.2020190074. 3. Sekuboyina A, Bayat AH, Husseini ME, Löffler M, Menze BM, ..., Kirschke JS. VerSe: A Vertebrae Labelling and Segmentation Benchmark. https://arxiv.org/abs/2001.09193 An overwiew of the data is provided in reference 1. The methods to generate the initial segmentations that were manually corrected afterwards are detailed in reference 2 and 3. This work has been supported by the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 637164 — iBack — ERC-2014-STG). Data Format ----------- Scans and segementation masks are stored in NIFTI format (https://nifti.nimh.nih.gov/). Coordinates of vertebral body centroids per vertebral level are stored in JSON format. Please refer to the Component Wiki Pages for further details. Licence and Ethics ------------------ The data is published under the licence CC BY-SA 2.0 (see licence.txt). When using the data you must cite the three papers mentioned above. Ethical approval to publish this data has been obtained from the local ethics committee at the Technical University of Munich (Proposal 27/19 S-SR). Subsequent Project ------------------ The follow-up of this project is available here: https://osf.io/t98fz/ [1]: https://doi.org/10.1148/ryai.2020190138