Dear BraTS-GoAT Organizers, I would like to request clarification regarding my training protocol. I initially developed my model for the BraTS Brain Metastases Challenge using this year's standalone BraTS-METS training dataset. Later, I decided to participate in BraTS-GoAT and fine-tuned the same model on the official BraTS-GoAT training data. I am an undergraduate researcher working independently, and retraining the model from scratch is unfortunately not feasible due to limited computational and financial resources. Could you please advise whether this training protocol is eligible for the BraTS-GoAT challenge, or whether the model would be considered non-compliant under the challenge rules? Thank you for your time.

Created by Oserebameh Beckley Babzo
Dear @Babzo and @aec8, Thanks for reaching out. Please note that the provided datasets for the GoAT task are all pre-operative MRIs, while, for example, METs2026 or PEDs2026 contain post-surgery/post-treatment scans. You are of course welcome to use other datasets for your experimental studies and comparisons, but your final solution should meet the [requirements of the GoAT task](https://www.synapse.org/Synapse:syn74274097/wiki/639591). Best regards, /Mehdi
We were curious about a similar architectural question - specifically, whether pre-training on pooled sub-challenge data improves cross-task segmentation performance. However, based on our reading of the strict data boundaries in the rules, it appears this approach would make a model non-compliant for the final leaderboard submission. Out of curiosity, could the organizers clarify the rationale for restricting data sharing across the sub-challenges? I understand not wanting external data, but we all have access to the sub-challenge data.

Request for clarification on Brats-GoATs training protocol page is loading…