DoseRAD2026: Real-time dose calculation in radiotherapy

In modern clinical radiotherapy treatment planning systems, dose calculation times with Monte-Carlo simulations providing clinically acceptable dose accuracy can range from 13 to 70 seconds for GPU-based methods [1,2] and up to several minutes for CPU based approaches, which are still widespread in clinics [1-4]. The acceleration of radiotherapy dose calculation brings obvious benefits to treatment planning, online plan adaptation, quality assurance, and may ultimately foster real-time dose-guided radiotherapy, where radiation delivery is continuously tailored to the patient’s moving anatomy via a real-time feedback loop [5]. Recent developments suggest that AI-based dose calculation directly on MRI may offer a pragmatic solution to close this loop by updating the delivered dose distribution in real-time [6,7]. It remains to be seen what the optimal strategy is to achieve (near-)real-time dose calculation on 3D anatomy to combine accuracy and calculation speed. The DoseRAD2026 Grand Challenge aims to benchmark state-of-the-art methods for real-time dose calculations directly on images (both CT and MRI), in photon and proton radiotherapy.

We propose to organize the DoseRAD2026 challenge, building on our previous experiences with the TrackRAD2025 [8] and SynthRAD2025 MICCAI challenges [9]. The challenge will feature four tasks for real-time photon and proton dose calculation using either CT or MRI images as inputs, and we plan to build on the existing SynthRAD2025 dataset which features paired MRI-CT images to generate ground truth Monte-Carlo based dose distributions for training and testing. We seek a broad dataset of photon beams generated from arbitrary multi-leaf collimator (MLC) apertures, and proton beams for pencil beam scanning. Ground truth data will be obtained from lengthy but accurate full physics modelling via Monte Carlo simulation of particle transport. Four tasks will address the latest technical developments in radiation therapy:

1. Photon therapy dose calculation on CT images is crucial for treatment planning of the delivery of intensitymodulated photon therapy using volumetric modulated arc therapy (VMAT), which is used for the vast majority of patients. 
2. Photon therapy dose calculation on MRI images for cutting edge MRI-linacs which allow in-room online adaptation of radiation plans based on high soft tissue contrast MRIs.
3. Proton therapy dose calculation on CT images is required for treatment planning of high precision proton beams, which deposit dose more locally than photon beams and require advanced proton acceleration beamlines and gantries.
4. Proton therapy dose calculation on MRI images supports both MRI-only proton therapy radiation planning and prototype development of in-room MRI guidance for proton therapy, similarly to MRI-linacs.

Using the paired SynthRAD2025 dataset allows us to address these four tasks with a consistent dataset.

The objective of the challenge will be fast and accurate dose calculation of individual radiation beams on either CT or MRI images of patients. Beams will either be defined by a photon linear accelerator’s MLC or a proton therapy system’s pencil beam scanning parameters. The algorithms will be provided with the CT or MRI and beam parameters, and need to output a beam-specific radiation dose distribution in 3D.

The organization of this challenge will be supported by an international group of experts in photon therapy, proton therapy and MRI-linacs (the Netherlands, Switzerland, Germany, Sweden) and will be hosted on the open access grand-challenge.org platform. The top winning teams will earn prize money. The challenge is expected to launch in March 2026, with the testing phase in July 2026, and result announcement at MICCAI in Abu Dhabi in October 2026, depending on acceptance at MICCAI in early 2026.