DoseRAD2026 Grand Challenge dataset

DoseRAD2026 Dataset The DoseRAD2026 dataset is a large-scale, multimodal radiotherapy dataset designed to support the development and benchmarking of fast and accurate radiation dose calculation and prediction methods and is accompanied by the DoseRAD2026 deep learning challenge. It provides paired computed tomography (CT) and magnetic resonance imaging (MRI) data alongside beam-level Monte Carlo (MC)–simulated dose distributions for both photon and proton therapy. The dataset is specifically tailored to enable research in MR-guided radiotherapy, MRI-only workflows, and real-time adaptive treatment planning.A detailed description of the dataset is provided in the attached PDF, which has also been uploaded to arXiv and will be submitted to Medical Physics. 💾 Download 💾 Due to the dataset size, the dataset is hosted on Hugging Face and can be downloaded via the following link: Download Data Dataset Composition The public dataset comprises 115 patients with thoracic and abdominal malignancies: - Training set: 75 patients (release April 2026)- Test set: 40 patients (release March 2030) The dataset is divided into a photon and proton dose subset and each subset can be downloaded separately.Each patient includes: - A planning MRI volume acquired on a 0.35T MR-Linac system (bSSFP sequence) - A corresponding CT volume, deformably registered to MRI - Beam configuration files (JSON format) - Beam-level Monte Carlo dose distributions All image and dose data are provided in MetaImage (.mha) format with consistent spatial metadata. Pre-processing and Data Curation To ensure high-quality multimodal consistency, the dataset underwent: - Deformable CT-to-MRI registration with manual quality control - Air cavity correction for abdominal CTs- Body masking - Task-specific resampling: - Photon tasks: 2 × 2 × 2 mm³ - Proton tasks: 1 × 1 × 3 mm³ Only cases with high-quality spatial alignment between CT and MRI were included, ensuring reliable voxel-wise correspondence for supervised learning tasks. Dose Simulation Ground truth dose distributions were generated using Monte Carlo simulations (Geant4) Photon Dose (VMAT segments) Simplified model of an Elekta Versa HD Linac with an Agility 160-leaf MLC (scriptable VMAT plans based on matRad) Defined by: Multi-leaf collimator (MLC) apertures Gantry angles (full arc sampling) Isocenter positions Dataset scale: 40,500 photon beam segments (training) Proton Dose (pencil beam scanning) Simplified proton beam model employing an energy-dependent, single-Gaussian approximation for both the beamlet energy spread and the spot size Defined by: Source position and gantry angle Energy levels (31.7–200.8 MeV) Pencil beam spots arranged in beam’s-eye-view grids Dataset scale: 81,000 proton beamlets (training) All dose distributions are: - Computed as dose-to-medium (Gy) - Spatially aligned with CT and MRI volumes - Masked to the patient CT body contour Data Access and Licensing The dataset is released under the CC BY-NC 4.0 license, permitting non-commercial use with appropriate attribution. The training set is publicly available from April 2026, while the test set remains withheld until March 2030 to support objective benchmarking in the DoseRAD2026 challenge.