Hello, On the BraTS 2023 page it is stated that the 2023 dataset has been updated, with changes in preprocessing: > While you can download the BraTS 2023 Adult Glioma dataset from this challenge website, please be aware that it has been updated and incorporated into the BraTS-Lighthouse 2025: Pre- and Post-Treatment Adult Glioma Challenge. The updated dataset for the new challenge underwent different preprocessing, so the BraTS 2023 and BraTS 2025 Adult Glioma datasets are no longer identical. Given this, I would like to clarify the following: 1. The dataset currently available at this link ([https://www.synapse.org/Synapse:syn51514105](https://www.synapse.org/Synapse:syn51514105)), although listed on the BraTS 2023 page, has been updated and is therefore not identical to the dataset originally used in the 2023 challenge. Is that correct? 2. If so, does this mean that results obtained using the currently available training data are not directly comparable to results reported in papers that used the original 2023 dataset? 3. Are results obtained using this dataset ([https://www.synapse.org/Synapse:syn51514105](https://www.synapse.org/Synapse:syn51514105)) comparable to submissions evaluated on the validation/test sets of the still-open 2023 challenge? Thank you in advance for the clarification.

Created by Lucas Santana lssantana
``` In other words, are these the exact same files that were used in the original 2023 challenge? ``` Yes ```.... in the MNI Atlas space?``` Yes SRI and MNI space have two different dimensions, coordinates, spacing etc...
Thank you very much for the clarification! I would greatly appreciate it if you could clarify one more point. 1. Are the BraTS GLI 2023 files (`ASNR-MICCAI-BraTS2023-GLI-Challenge-TrainingData.zip`) available at this link: [Synapse BraTS 2023 GLI Dataset](https://www.synapse.org/Synapse:syn51514105) still in the SRI24 Atlas space? In other words, are these the exact same files that were used in the original 2023 challenge? 2. And are the BraTS GLI 2025 files (`BraTS2025-GLI-PRE-Challenge-TrainingData.zip`) available at this link: [Synapse BraTS 2025 GLI Dataset](https://www.synapse.org/Synapse:syn65773245) in the MNI Atlas space? I am asking because I downloaded both datasets and noticed that the files have different sizes, so I would like to confirm whether this difference is indeed due to the change in atlas/preprocessing.
@lssantana Thanks for reaching out. The BraTS GLI 2023 dataset includes pre-operative MRIs in the SRI24 Atlas space. In the BraTS 2025 GLI challenge, the data were transformed into MNI Atlas space. Therefore, the subjects remained identical; however, volumes (images and masks) were transformed. So the short answer is NO. The MNI152 template is larger than the SRI24 atlas. Since HD95 is measured in millimeters (mm), a 5mm error in a larger MNI brain represents a "smaller" relative error than 5mm in a smaller SRI24 brain. While Dice is a ratio and theoretically scale-invariant, the resampling process to fit a brain into a larger space (MNI) vs. a smaller space (SRI24) involves different interpolation math. This can subtly "smooth" or "alias" the tumor boundaries, shifting your Dice score by a few percentage points regardless of your model's actual performance. If you want to ensure that results are directly comparable, you need to register your predicted masks from one doman into another. Best regards, /Mehdi

BraTS 2023 Dataset Update: Are Results Still Comparable to Original Challenge Submissions? page is loading…