medarc/adni-mini-v1-1

ADNI Mini v1.1

B-equal ADNI sMRI evaluation set for frozen-feature / linear-probe evaluation.

• 1,200 subjects / 1,200 scans (one processed T1w per subject)
• Diagnosis: 400 CN / 400 MCI / 400 AD
• Sex: 600 Female / 600 Male
• Age at scan: 60–90
• Single eval split as Hugging Face Arrow shards with embedded NIfTI bytes
• Brain masks are not included

Primary label coverage (approx.)

Target	n
Amyloid centiloid (all / CN / MCI / AD)	~1082 / 400 / 400 / 282
Tau meta-temporal SUVR (all / CN)	~833 / 400
CSF Aβ42 / p-tau / t-tau	~860 / 855 / 858
MCI conversion time (months)	~398

Load

from datasets import load_dataset

ds = load_dataset("medarc/adni-mini-v1-1", split="eval")
print(ds)
# image column: nifti

Local path equivalent: set ADNI_EVAL_PATH to a load_from_disk folder with the same layout.

Design notes

• Continuous biomarker labels preferred over binary amyloid/tau for evaluation.
• Within-CN centiloid/tau are the main preclinical targets (less severity confounding).
• Conversion is continuous time-to-AD among MCI, not a fixed 3y binary as the primary task.
• Subject-grouped CV: use participant_id (one scan per subject in this release).