Early detection of breast cancer remains one of the most critical factors in improving patient outcomes. Magnetic Resonance Imaging (MRI) is increasingly used alongside mammography, particularly for women with dense breast tissue or elevated risk. However, interpreting large 3D MRI volumes is time-consuming for radiologists, limiting its widespread use in screening programs. To address this challenge, the ODELIA consortium launched an ambitious public competition to accelerate the development of AI tools for breast cancer detection in MRI.
The Challenge
The ODELIA Breast MRI Challenge 2025 invited global participants to develop AI models capable of predicting cancer at the case level using a curated public dataset of 511 breast MRI examinations from five European sites. An additional 231 examinations, including data from a sixth site, were withheld as a holdout set to rigorously test generalizability. Submissions were evaluated on three clinically relevant metrics: AUC, sensitivity at 90% specificity, and specificity at 90% sensitivity.
Hosted on the Grand Challenge platform, the event attracted 95 registrants from 24 countries. After a pre-evaluation phase, seven teams submitted final models, which were showcased at the Deep-Breath workshop during MICCAI 2025.
Results
The winning team, DivideAndConquer, achieved an impressive overall score of 0.769 on the main dataset, with an AUC of 0.897. Other strong contenders included BCN-AIM and agaldran. Performance dropped across all models when tested on the holdout dataset, underscoring the complexity and variability of breast MRI data across different sites and vendors.
Impact and Next Steps
This challenge marks a significant milestone for ODELIA’s mission to foster decentralized medical AI innovation. The submitted models will now be considered for integration and further training within the ODELIA swarm learning framework, paving the way for robust, generalizable AI solutions in breast cancer screening.
Learn more here:
