Hi, I am participating in Challenge 1 (Brain Metastases) and have a question about training data policy for BraTS 2026. The BraTS-METS 2023 rules explicitly restricted the use of external annotated datasets for training. However, I could not find an equivalent statement in the BraTS 2026 Challenge 1 rules documentation. Could you please clarify: 1. Are participants allowed to use external publicly available annotated datasets (e.g. NYUMets, UCSF-BMSR, BrainMetShare) as additional training data for BraTS 2026 Challenge 1? 2. If allowed, is disclosure in the methods paper sufficient, or are there additional requirements? 3. Are pre-trained model weights from other segmentation tasks (e.g. glioma, general brain MRI) permitted as initialization? Thank you.

Created by vineet jangir vineet3jangir
Thank you very much for your observation! We will solve the discrepancy. If any other questions arise, please reach out to us. Thank you for your understanding! Cordially, Nikolay
Dear @Niki_Yordanov_99, Thank you a lot for the clarification, Could I flag a discrepancy for you to resolve on the rules page? The Challenge Rules (Terms & Conditions, "Use of Data in addition to BraTS and pre-trained models") currently state that participants "are allowed to use additional publicly available annotated datasets and publicly available pre-trained models," restricting only models generated from previous BraTS data. That reads as the opposite of the policy you've stated here. It would help everyone to have the written rules updated, so participants relying on the T&C page aren't misled. Thanks a lot. Best, Francesco
Dear participant, The use of external annotated datasets for training the model you are about to submit to the challenge is not allowed!!! Pre-trained model weights from other segmentation tasks are not allowed as well !!! Thank you for your understanding!

Training Data Policy for BraTS 2026 Challenge 1 — External Datasets and Pre-trained Weights page is loading…