ProbioticFarmer/mindboggle-101-freesurfer

Mindboggle-101 — Complete Lab-Processed FreeSurfer recon-all Release (Baseline + Rescans)

Complete lab-computed FreeSurfer recon-all outputs from Mindboggle-101 (Klein & Tourville, 2012; OSF osf.io/wzqmg & Zenodo 10.5281/zenodo.22072032, CC-BY 4.0), covering 146 total FreeSurfer subject-sessions — including 43 dedicated short-interval test-retest rescan sessions (OASIS-TRT-20-rescan and MMRR-21-rescan) and paired 3T/7T sessions (MMRR-3T7T-2).

Dual-Access Cloud Organization for Hugging Face Jobs

1. Ready-to-Load NIfTI-1 Volumes (volumes_nifti/{cohort}/{subject_id}/*.nii.gz): T1.nii.gz, brain.nii.gz, norm.nii.gz, aseg.nii.gz, aparc+aseg.nii.gz, aparc.a2009s+aseg.nii.gz, wmparc.nii.gz, and ribbon.nii.gz losslessly converted from .mgz so Python / nibabel / ANTsPy / Lean FFI workflows load FreeSurfer posteriors & segmentations directly without installing FreeSurfer.
2. Pre-Parsed Morphometry Parquet Tables (metadata/aseg_stats.parquet & metadata/aparc_stats.parquet): Every subject's stats/aseg.stats and stats/{lh,rh}.aparc.stats parsed into queryable Parquet tables.
3. Complete Unmodified recon-all Subject Trees (recon_all_tarballs/{cohort}/{subject_id}.tar.gz): 100% lossless FreeSurfer directories (mri/, surf/lh.white, surf/lh.pial, surf/lh.sphere.reg, label/*.annot, stats/) packaged per subject to avoid 65k+ tiny-file Git overhead.

Fast Usage on Hugging Face Jobs

from huggingface_hub import snapshot_download
import pandas as pd
import nibabel as nib

root = snapshot_download(
    repo_id="ProbioticFarmer/mindboggle-101-freesurfer",
    repo_type="dataset",
    allow_patterns=["metadata/*", "volumes_nifti/*"],
)
sessions = pd.read_parquet(f"{root}/metadata/sessions.parquet")
aseg = pd.read_parquet(f"{root}/metadata/aseg_stats.parquet")

Citation

Klein, A., & Tourville, J. (2012). 101 labeled brain images and a consistent human cortical labeling protocol. Frontiers in Neuroscience, 6, 171. (doi:10.5281/zenodo.22072032)