Brain age estimation pipeline: trained weights and Apptainer image (Rajabli et al., 2025)

Résumé fourni par la source

This record provides the trained models and a ready-to-run container for the brain age estimation pipeline described in: Rajabli R, Soltaninejad M, Fonov VS, Bzdok D, Collins DL. "Brain Age Prediction:Deep Models Need a Hand to Generalize." Human Brain Mapping, 2025;46(11).doi:10.1002/hbm.70254 CONTENTS brain-age-model.sif: a self-contained Apptainer/Singularity image that runs the full pipeline (preprocessing and inference) on CPU. weights.zip: the nine trained model checkpoints used by the ensemble (needed only if you rebuild the image from source). WHAT IT DOES The pipeline estimates brain age from a single T1-weighted MRI. It performs skull-stripping (SynthStrip), denoising, N4 bias-field correction, and linear registration to MNI152 space, then predicts brain age with a 9-model deep ensemble and applies a per-model bias correction. All neuroimaging tools and Python dependencies are bundled inside the image; no GPU is required. Source code and usage: GitHub Repo IMPORTANT NOTES For research use only; not a medical device. DO NOT RUN the model on UK Biobank (UKBB) data. The models were trained and validated on 35,247 UKBB subjects, so predictions on those subjects are pulled toward their chronological age and artificially shrink the brain age gap. LICENSE The authors' own contributions in this record (the trained weights and the pipeline/container build recipe) are released for non-commercial use under CC BY-NC 4.0. For commercial use, please contact the authors to arrange a license. The container image additionally bundles the third-party software listed below; those components are not relicensed and remain under their own terms, with copyright held by their respective authors. If you use these models or the container, please cite this Zenodo record and the paper above, and follow the citation requirements of the bundled components. THIRD-PARTY COMPONENTS BUNDLED IN THE IMAGE Third-party components bundled in the image (each under its own license): SynthStrip (FreeSurfer): MIT or CC BY 4.0. https://surfer.nmr.mgh.harvard.edu/docs/synthstrip/ MNI ICBM152 2009c atlas (MNI, McGill): MNI/BIC permissive license. https://www.bic.mni.mcgill.ca/ServicesAtlases/ICBM152NLin2009 MINC toolkit and the ANTs tools it uses (antsRegistration, N4BiasFieldCorrection): open-source (BSD / Apache 2.0). https://bic-mni.github.io/ , https://github.com/ANTsX/ANTs PyTorch and other Python dependencies (numpy, scipy, nibabel, torchio, SimpleITK, and more): open-source, mostly BSD-3-Clause, MIT, or Apache-2.0. See requirements.txt in the repository. Please cite each tool as its authors request.