ODELIA BREAST MRI Challenge 2025 Banner
Join

ODELIA Breast MRI Challenge 2025

Early breast cancer detection is paramount and MRI increasingly supplements mammography for high-risk or dense-breasted women. Reading large 3D volumes demands significant radiologist time and is a barrier to the use of MRI in screening. To accelerate the development of AI tools for earlier, more accurate breast cancer detection in MRI, the ODELIA consortium launched this challenge.

Top five contestants/teams will be invited to present their work as part of the Deep-Breath workshop at MICCAI 2025.

Participation

Participants need to verify their Grand Challenge account before they can submit. Verification is done manually and can take some time, so please do this as soon as possible. Other information is given on the General Rules and Info page.

The Data

500 breast MRI studies labelled "no lesion" / "benign lesion" / "malignant lesion" for each breast. This heterogenous dataset has been curated from six European centres, covering multiple vendors at both 1.5 and 3 T.

Each MRI study contains a T2-weighted acquisition alongside a DCE acquisition with one pre-contrast phase and between 2 and 7 post-contrast phases.

Ground truth definitions

  • Malignant lesion: lesion with enhancement observed and proven to be malignant with either histopathology or follow up information. Non mass enhancement proven to be malignant.
  • Benign lesion: well-defined lesion with enhancement observed and proven to not to be malignant with either histopathology or follow up information.
  • No lesion: none of the lesions described above are present.

Note: if both a malignant and a benign lesion are present, the classification given is "malignant lesion"

Full description of our dataset can be found here.
Our data is hosted on HuggingFace and can be found here along with code provided by the Grand Challenge team to set up an example algorithm.

The full data is now available in both default and unilateral format. The dataset used in the evaluation phases of the challenge will be .mha files of the T2 and DCE acquisitions (1 pre and a number of post-contrast phases) and your code will need to handle making a prediction per breast. We have provided the pre-processing script used to produce our unilateral data here (note: you will need to edit path variables etc. in the script to get it working on your system). You are free to use this code, adapt it, or scrap it in favour of your own code. We hope contestants will choose the latter so we can see what pre-processing methods you come up with!

The Challenge

Classification of breast MRI scans on a highly heterogeneous dataset.

Pre-evaluation phase

Teams will have three attempts to test their submissions on a subset of the 50 cases of the test dataset, that will be chosen to contain outliers or extreme examples of the test dataset (e.g. largest case, smallest case, diverse image quality, etc.) for participants to check their algorithms work in those cases. Prior to submission of final results, teams will also have to submit an abstract (approximately 3 pages) as part of pre-evaluation describing their methods, if any other publicly available data was used, and their training results.

A successful submission to this phase will be required to make a final submission.

Final submission

Limited to one submission per team and limited to teams that have successfully completed the pre-evaluation phase.

Evaluation will be run over the full test dataset and teams will be scored on AUC, specificity (at 90% sensitivity), and sensitivity (at 90% specificity). The average scoring will be computed by averaging the raw ranks of the three metrics. If a tie occurs, the highest AUC score will be used to rank tied teams.

Awards

  • 1,000€ prize for challenge winners
  • Top 5 contestants/teams will be invited to present at the Deep-Breath workshop at MICCAI 2025

Key dates

June

  • 03 June 2025 - Training data released
  • 05 June 2025 - Further data added
  • 16 June 2025 - Pre-evaluation available

July

  • 17 July 2025 - Pre-evaluation deadline (required for final submission)
  • 31 July 2025 - Final submission deadline

August

  • 15 August 2025 - Final evaluation and results announced

September

  • 23 September 2025 - Deep-Breath workshop at MICCAI 2025