Introduction - SynthRAD2025

SynthRAD2025's organizers welcome you on the challenge website. We are refining the dataset details, hoping to finalize by end of 2024.

SynthRAD2025¶

🎯 Clinical problem¶

Medical imaging has become increasingly important in the diagnosis and treatment of oncological patients, particularly in radiotherapy (RT), which is used in more than half of all cancer patients. In RT, pre-treatment imaging is used to develop a radiation treatment plan, which is subsequently delivered in fractions over up to 30 days.

Traditionally, X-ray-based imaging is widely adopted in RT for patient positioning and monitoring before, during, or after the treatment, however, also magnetic resonance imaging (MRI) is playing an increasingly indispensable role in radiotherapy workflows

More specifically, accurately calculating radiation therapy doses for cancer patients requires detailed information about tissue photon attenuation or proton stopping power, traditionally obtained from computed tomography (CT) images. Ideally, this would be done at each treatment fraction. However, CT exposes patients to ionizing radiation, has a limited soft-tissue contrast and is usually not available in treatment rooms.

Opportunities¶

MRI: Magnetic resonance imaging (MRI) offers superb soft-tissue contrast without radiation but lacks the attenuation/stopping power data needed for radiation therapy dose calculation and treatment plan optimization [Schmidt et al., 2015].
CBCT: Cone-beam computed tomography (CBCT) is widely used in radiation therapy for in-room patient positioning and monitoring. However, its image quality is limited due to artifacts, hindering widespread adoption of in-room CBCT-based treatment planning [Boda-Heggeman J et al., 2011].

Solution Gap¶

Converting both MRI and CBCT to synthetic CT (sCT) images enables [Spadea & Maspero et al. 2021]:

MRI-only radiation therapy: Minimizing radiation exposure, increasing treatment accuracy and simplifying workflows [Edmund and Nyholm, 2017].
Adaptive MRI- or CBCT-based radiation therapy: sCT from either MRI or CBCT data for accurate dose calculations, allowing for accurate dose calculations and daily treatment plan adaptation (at each fraction) to anatomical changes during therapy.

Justification¶

Publicly available datasets and challenges are needed to compare and advance sCT generation methods for both MRI and CBCT. This will ultimately lead to safer, more efficient, and potentially personalized radiation therapy for cancer patients.

Objective & Tasks¶

This challenge aims to provide the first platform offering public data and evaluation metrics to compare the latest developments in sCT generation methods. The accepted challenge design approved by MICCAI can be found at https://doi.org/10.5281/zenodo.14051075. A type 2 challenge will be run, where the participant needs to submit their algorithm packaged in a docker container for testing. Two tasks are defined:

Task 1: MRI-to-sCT generation to facilitate MRI-only and MRI-based adaptive radiotherapy.
Task 2: CBCT-to-sCT generation to facilitate CBCT-based adaptive radiotherapy.

Challenge phases definition¶

Training: input and ground truth dataset are released to allow the teams to develop their algorithms.

Preliminary test: a phase to allow the teams to familiarize themselves with the submission system and how to package their algorithm for testing. The algorithm will run on a couple of cases, providing feedback for the teams based on the image similarity metrics in the related leaderboard.

Validation: the teams will submit the output images of their algorithms and afterwards image similarity and geometric fidelity metrics will be calculated and made openly available in a public leaderboard. Dose metrics will not be computed at this phase due to the high computational costs.

Test: the teams will submit their final algorithm in a docker container (max 2 times, only the last submission counts) and image similarity, geometric fidelity and dose metrics will be calculated and made openly available in a public leaderboard. The final ranking will be made available after the test phase is concluded, and after the organizers will revise, abiding to the award policies. While submitting a description of the method, along with a submission form, needs to be compiled.

Dataset¶

A multi-center dataset is provided, balancing training, validation and test cases to evaluate the methods on different MRI and CBCT devices and sequences/acquisition settings for head-and-neck, thorax and abdomen cases. Data will be provided by five university medical centers (UMC Utrecht, UMC Groningen, Radboud Nijmegen, LMU University Hospital Munich and University Hospital Cologne), which will contain around 65/10/25 (train/validation/test) MRI or CBCT and CT pairs per tumor entity of patients undergoing radiotherapy at the respective institute.

In total, 900-1200 paired MRI-CT and 1500 CBCT-CT sets are provided, along with a body mask that will be considered for evaluation and may be used during inference. The target CT of the validation and test set will not be shared with the participants to avoid optimistic biases. Input data of the validation and test set will be accessible by the algorithm submitted by the participants, but not directly available to the teams, until the end of the challenge.

Participation¶

Challenge participants may choose to participate either in Task 1 or 2 or both. For each task, algorithms should be provided for all the anatomical sites. The participants may decide whether to provide one or more models per task. The center of origin of the data will not be disclosed at a case level.

The evaluation code used to rank the challenge is shared at https://github.com/SynthRAD2025. After completion, the leaderboard will remain open for submission, ensuring that future methods may still be evaluated up to 2030.

We envision that this challenge will enable a fair and open evaluation of different approaches to synthetic CT generation from MRI and CBCT.

We hope you may have fun taking part in this challenge!

The organizers