Reproduce “BrainNext: A General-Purpose Self-Supervised Foundation Model for Brain MRI Analysis” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Moona Mazher, Abdul Qayyum, Steven A. Niederer, Daniel C. Alexander
BrainNext introduces a general-purpose self-supervised foundation model for volumetric brain MRI analysis, utilizing a native 3D Bi-Directional xLSTM-UNet architecture and masked autoencoder pretraining. The model learns transferable anatomical representations from large-scale unlabeled data, achieving second place overall and first place in meningioma segmentation in the FOMO 2025 Method Track.
Key insights
- Self-supervised pretraining with a 3D MAE objective can effectively learn robust, transferable volumetric anatomical representations from large unlabeled brain MRI datasets.
- A native three-dimensional Bi-Directional xLSTM-UNet architecture can successfully model both fine-grained local features and long-range spatial dependencies in 3D medical images.
- A single pretrained encoder can be adapted to diverse downstream neuroimaging tasks (classification, segmentation, regression) with strong generalizability, reducing dependency on extensive task-specific annotations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.