Jungle15/GDP-HMM_Challenge

Announcements

The paper for this dataset has been submitted for single-blind review. During the review period, access requests are temporarily disabled. If you are not a reviewer, please note that downloading the dataset implies agreement with the following terms:

• I agree to use this dataset for non-commercial use ONLY

• I agree to register the GDP-HMM challenge if I use the data before the end of the challenge

• I understand that my access to the data may be declined/removed if I have not correctly finished above steps

• I agree to read the README file and follow citation terms before downloading data

• I understand plans in this dataset is not clinical-approved

Dataset Card for GDP-HMM

This dataset is connected to the GDP-HMM challenge at AAPM 2025. The task is about generalizable dose prediction for radiotherapy.

The tutorials of using this dataset can be found in GitHub.

By downloading the dataset before the end of GDP-HMM challenge (May 2025), you are agreed to participate Phase I through III of the challenge and need to register the challenge first under "My Submissions" of the platform.

This repo provides the data with Numpy format which can be directly used with our GitHub baseline. For raw DICOM format, please visit Radiotherapy_HaN_Lung_AIRTP.

Dataset

In total, there are over 3700 RT plans included in the challenge covering head-and-neck and lung sites and IMRT & VMAT planning modes. There are three splits for the dataset.

The training split includes both input and label. The input include CT image, PTVs, OARs, helper structures, beam geometries, prescribed dose, etc.

The validation split only has the input shared to public. The participants of the challenge and researchers can submit their prediction to the challenge platform to get evalution results. We plan support this evaluation even during post-challenge and post the ranking in leaderboard.

The test split will be full hidden to public. During the challenge, participants need to submit their solution via docker. After the challenge, we will release the reference plans of the validation set and researchers can contact the lead organizer for collaboration to test on the hidden split.

• Curated by: Riqiang Gao and colleagues at Siemens Healthineers
• Funded by: Siemens Healthineers
• Shared by: Riqiang Gao
• Language(s) (NLP): English
• License: cc-by-nc-sa-4.0

Uses

The dataset is for research only. commercial use is not allowed. Although the plans in this dataset is not clinical-approved. the dataset should be helpful developing and validating AI models.

Dataset Creation

Documented in the Reference [1]. We sincerely acknowledge the support of TCIA (https://www.cancerimagingarchive.net) for data release.

Citation

If you use the dataset for your research, please cite below papers:

[1] Riqiang Gao, Mamadou Diallo, Han Liu, Anthony Magliari, Wilko Verbakel, Sandra Meyers, Masoud Zarepisheh, Rafe Mcbeth, Simon Arberet, Martin Kraus, Florin Ghesu, Ali Kamen. Automating RT Planning at Scale: High Quality Data For AI Training. arXiv preprint arXiv:2501.11803. 2025.

[2] Riqiang Gao, Bin Lou, Zhoubing Xu, Dorin Comaniciu, and Ali Kamen. "Flexible-cm gan: Towards precise 3d dose prediction in radiotherapy." In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023.