Dear Participants,

We would like to share an important update on the evaluation of the multi-modal task in LUMIR25. Concerns were raised about potential data leakage, as different modalities were provided for the same subjects in the validation set.

In practice, registration methods often have access to modality information. Therefore, participants are allowed to adapt their models accordingly, e.g., by switching weights between mono- and multi-modal tasks or using image synthesis when appropriate.

That said, please note that the validation set does not fully represent the test scenarios. The final test set may include a broader range of brain imaging modalities, and the exact combinations will not be disclosed in advance to preserve the zero-shot setting.

Importantly, during testing, each subject will be provided with only one modality, so assumptions like all subjects having T1w images will not hold.

We acknowledge that the original validation set may have encouraged such assumptions. In response, we have updated the dataset and will re-evaluate the leaderboard accordingly. Please see the main challenge page for updated download links.

Thank you for your engagement.

Best regards,

The Learn2Reg Organizing Team