BrainSegFounder: Towards 3D Foundation Models for Neuroimage Segmentation
Paper • 2406.10395 • Published
You need to agree to share your contact information to access this model
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
BrainSegFounder is distributed under the GNU General Public
License v3.0 (GPL-3.0), a copyleft license. You may use and
redistribute the weights, but any derivative work that
incorporates them must itself be licensed under GPL-3.0 and make
its source available. The UK-Biobank stage-1 pretrained weights
are not included (UK Biobank data-sharing terms); only the
BraTS-finetuned model is provided.
By requesting access you affirm that you understand the GPL-3.0
copyleft obligations governing the weights and any derivative
works, and that you have read the license linked from this model
card.
Log in or Sign Up to review the conditions and access this model content.
BrainSegFounder (BraTS-finetuned SwinUNETR) -- BrainSegFounder BraTS-finetuned SwinUNETR (fold 0)
Description
BrainSegFounder (Cox et al., lab-smile), ported to JAX / Equinox. BrainSegFounder is a 3D medical foundation model built by staged self-supervised pretraining (UK Biobank then BraTS) of a MONAI SwinUNETR, then fine-tuned for BraTS tumour segmentation. This port is the BraTS-finetuned full SwinUNETR (encoder + decoder), reusing the shared nimox SwinUNETR primitive. The UK-Biobank stage-1 foundation weights are not part of the public release (UKB data-sharing terms); the available checkpoints are the BraTS-stage SSL backbone and this BraTS-finetuned segmentation model.
Intended use
Brain-tumour sub-region segmentation from 4-channel BraTS MRI (T1, T1ce, T2, FLAIR), per-channel nonzero z-score normalised, each spatial dim a multiple of 32. Returns 3 raw-logit channels for the overlapping sub-regions tumour core (TC), whole tumour (WT), enhancing tumour (ET); apply a per-channel sigmoid and threshold at 0.5 (multi-label). The v0 bundle is the network forward only (no sliding-window inference).
Usage
from ilex.models.brainsegfounder import BrainSegFounder
model = BrainSegFounder.from_pretrained('ilex-hub/brainsegfounder.brats.1')
Authors
Cox J., et al. (lab-smile)
Citation
Cox J., Liu Z., Zhao Y., et al. (2024). BrainSegFounder: Towards 3D Foundation Models for Neuroimage Segmentation. Medical Image Analysis. arXiv:2406.10395. Backbone: Hatamizadeh A., et al. (2022). Swin UNETR. arXiv:2201.01266.
References
- Cox J., Liu Z., Zhao Y., et al. (2024). BrainSegFounder: Towards 3D Foundation Models for Neuroimage Segmentation. Medical Image Analysis. arXiv:2406.10395.
- Code / weights: https://github.com/lab-smile/BrainSegFounder
- Backbone: MONAI SwinUNETR (Hatamizadeh A., et al. 2022; arXiv:2201.01266).
License
HF Hub license tag: gpl-3.0
Effective terms: GNU General Public License v3.0 (the BrainSegFounder authors / lab-smile), on both the code (https://github.com/lab-smile/ BrainSegFounder) and the released weights. GPL-3.0 is a copyleft license: redistribution is permitted, but derivative works that incorporate the weights must themselves be GPL-3.0-licensed and provide source. The ilex JAX / Equinox port re-expresses the weights and is therefore distributed under GPL-3.0 for this bundle (the ilex port code is Apache-2.0 / GPL-3.0 dual; the GPL-3.0 arm governs this re-expression). The MONAI SwinUNETR architecture is itself Apache-2.0; the GPL-3.0 obligation comes from the BrainSegFounder weights. The UK-Biobank stage-1 pretrained weights are not redistributed here (UKB terms).
Upstream license reference: https://github.com/lab-smile/BrainSegFounder/blob/main/LICENSE
Copyright
Network architecture (MONAI SwinUNETR) and pretrained weights: copyright (c) the BrainSegFounder authors, released under the GNU General Public License v3.0 (GPL-3.0). JAX / Equinox port: copyright (c) the ilex authors, released under the Apache-2.0 / GPL-3.0 dual license used by ilex itself. Redistribution of the weights is subject to GPL-3.0.
Upstream source
Original weights / reference implementation: https://github.com/lab-smile/BrainSegFounder
Provenance
This artefact was produced by ilex's
save/load pipeline. The architecture is implemented in
ilex.models.brainsegfounder.BrainSegFounder and the weights have been converted
from their upstream format. See the upstream source above
for the canonical reference.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support