Please reply to this thread to report any data issues encountered with the BraTS-PED datasets.
Lead PED organizers: @anahita.kazerooni @mglinguraru
Created by Verena Chung vchung @alfakih
Of course, you can include any pre-processing and/or post-processing in your segmentation pipeline.
Regarding the skull-tripping, you may need to pay attention to the distribution of tumoral tissues and ensure they don't overlap with skull-stipped regions.
Best regards,
/Mehdi @astaraki
Thank you for the clarification.
I understand now that the provided data was defaced, not skull-stripped. However, I would like to clarify that I performed skull stripping myself on the training data as part of my preprocessing pipeline. Therefore, to maintain consistency, I plan to apply the same skull stripping to the test data before making predictions.
I just wanted to confirm whether this approach is acceptable from the challenge perspective.
@alfakih
Thanks for getting in touch.
All the MRI data (training, validation, and testing) for the PEDs task were preprocessed by applying a [Defacing algorithm](https://www.synapse.org/Synapse:syn64153130/wiki/631455), not skull stripping, due to the clinical relevance.
But if you observe training examples that were skull-stripped, can you please give us the names of such cases?
Best regards,
/Mehdi I have a question regarding the pediatric task.
During training, I am using skull-stripped images. However, I noticed that the test images include the skull. Should we include skull stripping as part of the preprocessing pipeline during inference, even though it is time-consuming? Or will the organizers provide skull-stripped test images before evaluation? @dr.amin.tavallaii
Once again, we would like to thank you for your feedback on the Pediatric dataset.
@timmulvany
The problematic cases have been corrected, and the dataset has been updated.
The current dataset is the finalized version so you can use it for model development.
Best regards,
/Mehdi
Hi @joeyx
Thank you for reaching out.
I can see that two out of three submitted jobs have been successfully evaluated. Did you receive the quantified metrics via email?
We are working to reduce the evaluation time of the submitted jobs, and it may take a few days. I appreciate your patience.
Best regards,
/Mehdi Dear BraTS-PED 2025 Organizers
I have uploaded my prediction results for the PED challenge on the website, and the current status shows "EVALUATION_IN_PROGRESS". I would like to know how long the backend validation process typically takes?
Thank you.
Best regards
Joey @the_heavy_shadow
Thanks for getting in touch.
The preprocessing steps of the PEDs dataset include registration to the SRI24 atlas and defacing, unlike e.g. GLI datasets, which were registered to the MNI152 atlas, followed by skull stripping.
Let me know if you have any questions.
/Mehdi I have a question about the pre-processing: Is the dataset (both training and val) already registered to an atlas? If yes, was it SRI24, similar to previous BraTS datasets? Dear @astaraki,
Thank you so much for the clarification and update, it's greatly appreciated!
Kind Regards,
Tim Mulvany Dear @timmulvany
Thanks for getting in touch!
The training and validation datasets for the PED task are the same as those used in BraTS24, with only a few exceptions.
These exceptions are cases @dr.amin.tavallaii identified as needing further review and correction.
That review is now complete, and the updated files for those specific subjects will be uploaded to the PED dataset very soon.
I'll let you know on this thread as soon as they're available.
You can go ahead and develop your model using the available dataset, regardless of the QC process.
Let me know if you have any questions!
/Mehdi Dear BraTS-PED 2025 Organizers,
I am hoping to be a participant in the 2025 BraTS-PED challenge, but am seeking some clarification regarding the final dataset to use for training my model.
On the challenge wiki, it states _**"The final QC-approved datasets are expected to be released by 5 May 2025"**_.
Under the files tab it states _**"BraTS-PED 2025 is using the same training and validation data as BraTS-PED 2024."**_.
Can I confirm the final training dataset is indeed the unchanged dataset the 2024 BraTS-PED challenge?
Additionally, have the below issues identified in this discussion thread by @dr.amin.tavallaii been incorporated into the dataset currently available for download?
Many thanks,
Tim Mulvany Dear @astaraki,
Thank you for your response. I have also noticed that in BraTS-PED-00259-000, we have a substraction map instead of t1c and a contrast enhanced T1 instead of t1n.
Best,
Amin Tavallaii Dear @dr.amin.tavallaii,
We appreciate your comments, and thank you for the very thorough examination of the cases.
We will review the mentioned cases and write back in this thread as soon as we can.
Best regards,
Subject: Request for Clarification and Replacement Files for BraTS-PEDs 2025 Task 6 Training Dataset Issues
Dear BraTS 2025 Organizers,
I am a participant in Task 6 (Pre-Treatment Pediatric Tumor Segmentation) and have visually inspected all 261 training volumes after skull-stripping with HD-BET. I identified the following issues in the dataset:
1. **Modality Sequence Errors** (4 cases):
- BraTS-PED-00128-000: T2W is a duplicate of T1CE (T2W missing).
- BraTS-PED-00177-000: T1CE and T1N are swapped.
- BraTS-PED-00200-000: T1N is contrast-enhanced (T1N missing).
- BraTS-PED-00203-000: T2W is T1N (T2W missing).
Could you provide replacement files for the missing T2W (00128, 00203) and T1N (00200)? For 00177, I plan to relabel files unless advised otherwise.
2. **Coordinate System Disorientation** (1 case):
BraTS-PED-00042-000 has incorrect orientation. Can you provide corrected files, or should I reorient locally?
These issues are critical for preprocessing and training. I’d appreciate your guidance on replacements, corrections, or approval for local fixes. Thank you for your support!
Best regards,
Amin Tavallaii, MD, MSc
Assistant Professor of Neurosurgery
Visiting Faculty at the Department of Health Informatics and AI