ProbioticFarmer/corr-multiscanner-test-retest

CoRR — Multi-Scanner Structural T1w Test-Retest BIDS Cohort (N = 400 Pairs, 12 Sites)

Complete paired baseline (ses-1) and repeat (ses-2) T1-weighted structural brain MRI cohort curated from the Consortium for Reliability and Reproducibility (CoRR) (Zuo et al., Scientific Data 2014, CC-BY 4.0), organized in drop-in BIDS format with Parquet manifests for high-throughput Conformal Risk Control and empirical null calibration on Hugging Face Jobs.

Cohort Highlights

• 400 independent human scan-rescan (ses-1 ↔ ses-2) T1w pairs (800 volumes) across 12 international MRI scanner sites, enabling exact finite-sample Conformal Risk Control down to $\alpha \le 0.005$ ($1 / (M_{\text{cal}} + 1) \ll 0.01$).
• Drop-in BIDS Compatibility: Preserves sub-{SITE}{ID}/ses-{1,2}/anat/sub-*_ses-*_T1w.nii.gz, scanner acquisition .json sidecars, dataset_description.json, and participants.tsv (compatible out-of-the-box with NeuroCert bids_runner.py and acp_calibrator.py).
• Phenotypic & Scanner Metadata: Joins per-site participants.tsv demographics (age, sex, handedness) and scanner protocol parameters (Manufacturer, ManufacturersModelName, MagneticFieldStrength, RepetitionTime, EchoTime) into metadata/scan_rescan_pairs.parquet.

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/corr-multiscanner-test-retest", repo_type="dataset")
pairs = pd.read_parquet(f"{root}/metadata/scan_rescan_pairs.parquet")
ses1_img = nib.load(f"{root}/{pairs.loc[0, 'ses1_t1w_path']}")
ses2_img = nib.load(f"{root}/{pairs.loc[0, 'ses2_t1w_path']}")

Citation

Zuo, X.-N., Anderson, J. S., Bellec, P., Birn, R. M., Biswal, B. B., Blautzik, J., ... & Milham, M. P. (2014). An open science resource for establishing reliability and reproducibility in functional connectomics. Scientific Data, 1, 140049. https://doi.org/10.1038/sdata.2014.49