ADNI T1w to BIDS

Reproducible selection and BIDSification of T1ws of ADNI1, 2/GO/3, and 4. ADNI 1,2/GO,3 QC determined by ADNIs QC standards. ADNI 4 QC has not been released. Two version will be available: all ADNI4 T1ws, and a set QC'd by CoBra Lab member Olivier Parent. Raw DICOMS were obtained from IDA and converted to bids with dcm2bids.

ADNI-derived data: subject to the ADNI Data Use Agreement, not for redistribution.

Selection

Phase	Script	QC source	Rule	n
ADNI1	adni1_build.py	MRIMPRANK	min RANK per (subject, visit, field strength)	3,535
GO/2	adni_go2_build.py	MAYOADIRL_MRI_IMAGEQC	keep accelerated + unaccelerated per session; drop grade 4	8,861
ADNI3	adni3_build.py	MAYOADIRL_MRI_QUALITY_ADNI3	SERIES_SELECTED=TRUE per session	2,329
ADNI4	adni4_build.py	in-lab ADNI4_QC_t1_cobralab.tsv	--bids: all T1w; --qc: grade 0/1	1,470 / 398

ADNI4 Mayo QC is unreleased; qc_olivier = 0 good, 1 artifacts, 2 fail.

The scripts build the list of images that pass QC. The output is then pasted into the IDA Advanced Search by Image-ID. This will give a list of images that can be added to collection. The images and the corresponding _cohort_manifest.csv can then be downloaded.

Requirements

• Python 3.12.3
• dcm2niix/v1.0.20250506
• dcm2bids/3.1.1

Inputs

QC/metadata sheets (IDA -> Study Data, newest dated version):

File	Phase	IDA name
MRIQC_*.csv	all	MAYO ADIR LAB MRI Quality [ADNI1,GO,2,3,4]
MRIMPRANK_*.csv	ADNI1	MRI MPRAGE Ranking [ADNI1]
MAYOADIRL_MRI_QUALITY_ADNI3_*.csv	ADNI3	Mayo ADNI3 MRI QC
MAYOADIRL_MRI_IMAGEQC_05_07_15_*.csv	GO/2	Mayo (Jack Lab) ADNI GO/2 MRI QC (_Archive)
ADNI4_QC_t1_cobralab.tsv	ADNI4	in-lab

Set ROOT (top of each script) to the directory holding the sheets and zips.

Usage

python3 adni1_build.py            # -> <phase>_cohort_manifest.csv + Image-ID download list
module load dcm2bids
python3 adni1_build.py --bids     # -> bids_adni1/

ADNIGO/2 and 3 are run in the same way. ADNI4 has two independent targets:

python3 adni4_build.py --bids     # bids_adni4_all/  (all T1w)
python3 adni4_build.py --qc       # bids_adni4_qc/   (grade 0/1; standalone)

Validate (converted vs missing, per phase):

python3 validate.py adni1_cohort_manifest.csv bids_adni1

Output

One BIDS dataset per phase (bids_adni1/, bids_adni3/, bids_adni_go2/; ADNI4 yields two, bids_adni4_all/ and bids_adni4_qc/), each laid out as:

bids_<phase>/
├── dataset_description.json
└── sub-<ID>/
    └── ses-<VISIT>/
        └── anat/
            └── sub-<ID>_ses-<VISIT>_acq-<ACQ>[_run-N][_rec-N]_T1w.nii.gz   (+ .json)

Entities:

• acq — field strength / acceleration: 15T, 3T, 3Taccel, 3Tunaccel, 3Tultrafast
• run-N — present only when a session has >1 scan of the same acq
• rec-N — Philips reconstruction variant (see Notes)

Notes

• acq (field strength / acceleration) is taken from MRIQC, not the DICOM sidecar (MagneticFieldStrength is inconsistent: 1.5 / 1.494 / 15000 G); injected per-scan via a Modality=MR dcm2bids config.
• dcm2niix may emit >1 NIfTI/scan: Philips multi-ImageType → run-01/02, resolved to canonical + rec-NN (resolve_run_splits); _Eq/sidecar-less outputs are left under tmp_dcm2bids/. The canonical _T1w is always written.
• ADNI1 m48 visits scanned in the GO/2 era are filed under the GO/2 project in IDA (search unscoped to retrieve); they remain ADNI1 by MRIProtocolPhase.

Acknowledgements

Pipeline code was developed with assistance from Claude (Anthropic).