SegRap 2023 Banner
Join

Segmentation of Organs-at-Risk and Gross Tumor Volume of NPC for Radiotherapy Planning (SegRap2023)

News

  • [2023.12.21] Now, the challenge summary report of SegRap2023 is available. If you used the SegRap2023 dataset for your research, please consider citing this ArXiv, BibTeX.
  • [2023.10.31] The training set with labels and the validation set without labels can be downloaded without any additional requirements, Google-Drive, Pan.Baidu, the unzipped password is *segrap2023@uestc. *
  • [2023.10.12] The final ranking results are available here. Thanks to all the participants.
  • [2023.09.18] The final detailed results of each task have been sent to each participant individually. In total 12/11 teams submitted the solutions for Task 01/02, and the top 5 teams will be invited to introduce their methods during MICCAI2023. We are looking forward to seeing you at the MICCAI2023&SegRap2023! We will release the training set and validation set for academic research later. Thanks for your attention.
  • [2023.08.21] The testing phase will be opened within 2 days, each team can successfully submit a testing phase docker image for each task once from 21 Aug. to 10. Sep. When submitting the docker images, please prepare a four or more-page technical report (please use the latex template of MICCAI) summarising your methods and details (like pre- or post-processing, data augmentation strategies, networks, training, testing strategies, etc.). Please note that each technical report can include three authors, please highlight the contact authors (we will email you the results or update the docker error). Please send the technical report to the email (xiangde.luo@std.uestc.edu.cn) when submitting the docker images, and we will provide feedback on the evaluation status.
  • [2023.07.13] The online evaluation submission is open, the docker submission tutorial and example can be found here. We are so sorry for the three-day delay due to some technical issues.
  • [2023.06.06] We revised the labels and converted these labels (some labels have overlap) into a one-hot way, and also provided a fixed strategy to re-convert the one-hot label to multi organs labels (revised labels, one-hot labels and code).
  • [2023.05.26] The cash awards of SegRap2023 are raised to \$4000 in total from \$2000. Call for challengers!!!
  • [2023.04.22] The dataset is now available for download, please follow the dataset to register and download data.
  • [2023.04.15] SegRap2023 website is now fully open. Please check the timeline.

Motivation

Radiotherapy is one of the most important cancer treatments for killing cancer cells with external beam radiation. Treatment planning is vital for radiotherapy, which sets up the radiation dose distribution for tumors and ordinary organs. The goal of planning is to ensure the cancer cells receive enough radiation and to prevent normal cells in organs-at-risk (OARs) from being damaged too much. For instance, optical nerves and chiasma in the head cannot receive too much radiation. Otherwise, the patient risks losing his/her vision. Gross Target Volume (GTV) is the position and extent of gross tumor imaged by CT scans, i.e., what can be seen. A critical step in radiation treatment planning is to delineate the boundaries of GTV and tens of OARs. However, manual delineation slice-by-slice in CT scans is tedious and time-consuming for radiation oncologists. Automatic delineation of GTV and OARs would substantially reduce the treatment planning time and therefore improve the efficiency of radiotherapy.

In this challenge, we will provide a dataset of CT scans of patients with nasopharyngeal carcinoma (NPC), where the segmentation targets will include OARs, Gross Target Volume of the nasopharynx (GTVnx), and Gross Target Volume of the lymph nodes (GTVnd). The dataset will consist of CT scans from 200 patients (120, 20, and 60 patients for training, validation and testing, respectively), each with two CT scans (a no-contrast CT and a contrast-enhanced CT) with pixel-level annotations of GTVnx, GTVnd, and 45 OARs. Based on this extensive and comprehensive dataset, two sub-tasks will be held in SegRap2023:

Task01: 45 OARs segmentation from no-contrast and contrast-enhanced CT scans: Brain, Brainstem, Chiasm, Cochlea left, Cochlea right, Esophagus, Eustachian tube left, Eustachian tube right, Eye left, Eye right, Hippocampus left, Hippocampus right, Internal auditory canal left, Internal auditory canal right, Larynx, Larynx glottic, Larynx supraglottic, Lens left, Lens right, Mandible left, Mandible right, Mastoid left, Mastoid right, Middle Ear left, Middle ear right, Optic nerve left, Optic nerve right, Oral cavity, Parotid left, Parotid right, Pharyngeal constrictor muscle, Pituitary, Spinal cord, Submandibular left, Submandibular right, Temporal lobe left, Temporal lobe right, Thyroid, Temporomandibular joint left, Temporomandibular joint right, Trachea, Tympanic cavity left, Tympanic cavity right, Vestibular semicircular canal left, Vestibular semicircular canal right.

Task02: GTVnx and GTVnd segmentation from no-contrast and contrast-enhanced CT scans: Gross Target Volume of nasopharynx (GTVnx) and Gross Target Volume of lymph node (GTVnd).

Figure 1. An example illustration of the SegRap dataset. SegRap provides CT scans from 200 patients (each patient has two CT scans) with voxel-level annotations of 45 OARs and 2 GTVs.

Timeline

1. Release of training data: May 10th April 22 (12:00 AM GMT), 2023;

2. Release of validation data: July 10th (12:00 AM GMT), 2023;

3. Docker and short paper submission of the testing set opening: Aug. 20th (12:00 AM GMT), 2023;

4. Submission deadline for results: ~~Sept. 10th (12:00 AM GMT) ~~Sept. 13th (3:00 AM GMT), 2023;

5. Announcement of final results: Oct. 8th, 2023.

Rule

1. All participants should register for this challenge with their real names, affiliation (including department, the full name of university/institute/company, country), and affiliation emails. Incomplete or redundant registrations will be removed without notice.

2. All participants must submit a complete solution to this challenge during the validation and testing phase (if you don't submit the results in the validation phase, we can not evaluate your solution in the testing phase). A complete solution includes a Docker container (tar file) and a 2-8-page qualified short paper.

3. All participants should agree that the submitted short papers can be publicly available to the community on the challenge website, and organizers can use the information provided by the participants, including scores, predicted labels, and short papers.

4. Participants should have docker expertise and the submitted Docker tar file size is preferred less than 8 GB (please don't ensemble more than 5 models). A Docker size of over 12 GB will raise an error. The Docker should execute for at most  3 hours and occupy no more than 10 GB GPU Memory (we are so sorry that we just have 2080Ti for evaluation) to generate segmentation results of the testing set (60 cases). Otherwise, an error will be raised.  

5. Participants are not allowed to register multiple teams and accounts ( only listed names in the signed document will be considered ). Participants from the same research group are also not allowed to register multiple teams. One participant can only join one team.  Organizers keep the right to disqualify such participants. 

6. The dataset will be emailed to the participants after they submitted the signed end-user agreement. Redistribution or transfer of data or data links are not allowed. Participants should use the data only by themselves. The challenge data and results will be free to use after MICCAI2023.

7. For a fair comparison, participants are not allowed to use any additional private data about the 45 OARs, GTVnx and GTVnx, but the pre-trained or foundation models (self-/partially- supervised learning without using the above organs or lesions) are allowed.