The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
Error code: DatasetGenerationCastError
Exception: DatasetGenerationCastError
Message: An error occurred while generating the dataset
All the data files must have the same columns, but at some point there are 14 new columns ({'assd_mm_macro_q25', 'assd_mm_macro_median', 'dice_macro_n', 'dice_macro_q25', 'nmi_q25', 'assd_mm_macro_n', 'nmi_q75', 'dataset', 'assd_mm_macro_q75', 'dice_macro_median', 'nmi_median', 'n_pairs', 'dice_macro_q75', 'nmi_n'}) and 6 missing columns ({'mean', 'weighting', 'sd', 'metric', 'n', 'total_pairs'}).
This happened while the csv dataset builder was generating data using
hf://datasets/lipeirong/MRI-PairedField/validation/anatomical_correspondence_by_source.csv (at revision 8a023108f5124fd91f18e8bd0fa759c061969cf6), ['hf://datasets/lipeirong/MRI-PairedField@8a023108f5124fd91f18e8bd0fa759c061969cf6/validation/anatomical_correspondence_by_contrast.csv', 'hf://datasets/lipeirong/MRI-PairedField@8a023108f5124fd91f18e8bd0fa759c061969cf6/validation/anatomical_correspondence_by_source.csv', 'hf://datasets/lipeirong/MRI-PairedField@8a023108f5124fd91f18e8bd0fa759c061969cf6/validation/downstream_translation_summary.tsv', 'hf://datasets/lipeirong/MRI-PairedField@8a023108f5124fd91f18e8bd0fa759c061969cf6/validation/image_quality_summary.tsv']
Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1848, in _prepare_split_single
writer.write_table(table)
~~~~~~~~~~~~~~~~~~^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 765, in write_table
self._write_table(pa_table, writer_batch_size=writer_batch_size)
~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 773, in _write_table
pa_table = table_cast(pa_table, self._schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
dataset: string
modality: string
n_pairs: int64
dice_macro_n: int64
dice_macro_q25: double
dice_macro_median: double
dice_macro_q75: double
assd_mm_macro_n: int64
assd_mm_macro_q25: double
assd_mm_macro_median: double
assd_mm_macro_q75: double
nmi_n: int64
nmi_q25: double
nmi_median: double
nmi_q75: double
-- schema metadata --
pandas: '{"index_columns": [{"kind": "range", "name": null, "start": 0, "' + 2129
to
{'modality': Value('string'), 'metric': Value('string'), 'total_pairs': Value('int64'), 'n': Value('int64'), 'mean': Value('float64'), 'sd': Value('float64'), 'weighting': Value('string')}
because column names don't match
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1850, in _prepare_split_single
raise DatasetGenerationCastError.from_cast_error(
...<4 lines>...
)
datasets.exceptions.DatasetGenerationCastError: An error occurred while generating the dataset
All the data files must have the same columns, but at some point there are 14 new columns ({'assd_mm_macro_q25', 'assd_mm_macro_median', 'dice_macro_n', 'dice_macro_q25', 'nmi_q25', 'assd_mm_macro_n', 'nmi_q75', 'dataset', 'assd_mm_macro_q75', 'dice_macro_median', 'nmi_median', 'n_pairs', 'dice_macro_q75', 'nmi_n'}) and 6 missing columns ({'mean', 'weighting', 'sd', 'metric', 'n', 'total_pairs'}).
This happened while the csv dataset builder was generating data using
hf://datasets/lipeirong/MRI-PairedField/validation/anatomical_correspondence_by_source.csv (at revision 8a023108f5124fd91f18e8bd0fa759c061969cf6), ['hf://datasets/lipeirong/MRI-PairedField@8a023108f5124fd91f18e8bd0fa759c061969cf6/validation/anatomical_correspondence_by_contrast.csv', 'hf://datasets/lipeirong/MRI-PairedField@8a023108f5124fd91f18e8bd0fa759c061969cf6/validation/anatomical_correspondence_by_source.csv', 'hf://datasets/lipeirong/MRI-PairedField@8a023108f5124fd91f18e8bd0fa759c061969cf6/validation/downstream_translation_summary.tsv', 'hf://datasets/lipeirong/MRI-PairedField@8a023108f5124fd91f18e8bd0fa759c061969cf6/validation/image_quality_summary.tsv']
Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
modality string | metric string | total_pairs int64 | n int64 | mean float64 | sd float64 | weighting string |
|---|---|---|---|---|---|---|
T1W | dice_macro | 161 | 147 | 0.845479 | 0.103245 | equal pair weight; no imputation |
T1W | assd_mm_macro | 161 | 147 | 0.57288 | 0.387901 | equal pair weight; no imputation |
T1W | nmi | 161 | 161 | 1.087679 | 0.039242 | equal pair weight; no imputation |
T2W | dice_macro | 87 | 82 | 0.638754 | 0.250188 | equal pair weight; no imputation |
T2W | assd_mm_macro | 87 | 82 | 1.626473 | 1.490286 | equal pair weight; no imputation |
T2W | nmi | 87 | 87 | 1.036928 | 0.02264 | equal pair weight; no imputation |
T2-FLAIR | dice_macro | 11 | 1 | 0.71341 | null | equal pair weight; no imputation |
T2-FLAIR | assd_mm_macro | 11 | 1 | 1.111576 | null | equal pair weight; no imputation |
T2-FLAIR | nmi | 11 | 11 | 1.025596 | 0.007117 | equal pair weight; no imputation |
Overall | dice_macro | 259 | 230 | 0.771203 | 0.196911 | equal pair weight; no imputation |
Overall | assd_mm_macro | 259 | 230 | 0.950851 | 1.066198 | equal pair weight; no imputation |
Overall | nmi | 259 | 259 | 1.067995 | 0.042092 | equal pair weight; no imputation |
T1W | null | null | null | null | null | null |
T1W | null | null | null | null | null | null |
T2W | null | null | null | null | null | null |
T1W | null | null | null | null | null | null |
T2FLAIR | null | null | null | null | null | null |
T2W | null | null | null | null | null | null |
T1W | null | null | null | null | null | null |
T2W | null | null | null | null | null | null |
T1W | null | null | null | null | null | null |
T1W | null | null | null | null | null | null |
T2W | null | null | null | null | null | null |
T1W | null | null | null | null | null | null |
T2W | null | null | null | null | null | null |
T1W | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
null | null | null | null | null | null | null |
MRI-PairedField
MRI-PairedField harmonizes participant-matched structural brain MRI acquired at 64 mT, 1.5 T, 3 T and 7 T into a common full-volume format for cross-field image translation and related neuroimaging research. It combines eight source datasets and preserves explicit pairing, provenance and access metadata.
- Archival data record and DOI: Zenodo v1.0.0
- Processing and data-loading code: PeirongLiL/MRI-PairedField
- Repository metadata tag:
v1.0.1(naming and metadata fields only; image files are unchanged from the archival release)
The complete logical dataset contains 518 volume records. This repository provides 424 full three-dimensional NIfTI volumes from seven openly distributable sources. The remaining 94 ADNI volumes are not redistributed; paired image-selection lists and reconstruction instructions are provided so that authorised users can request them from the LONI Image & Data Archive. Source-specific terms remain applicable, and users should cite each original source they use.
Contents
The complete logical dataset contains 161 paired cases, 259 same-contrast image pairs and 518 volume records. Of these, 424 processed volumes from seven sources are publicly downloadable here. The remaining 94 ADNI volumes must be requested separately through the LONI Image & Data Archive. A paired case represents a participant-specific acquisition or visit at two field strengths. An image pair contains two volumes of the same contrast, one at each field strength.
| Source | Paired cases | Image pairs | Field strengths | Contrasts | Volumes |
|---|---|---|---|---|---|
| ULF | 23 | 46 | 64 mT, 3 T | T1W, T2W | 92 |
| LUMC | 11 | 33 | 64 mT, 3 T | T1W, T2W, T2-FLAIR | 66 |
| UNC | 10 | 20 | 3 T, 7 T | T1W, T2W | 40 |
| Subcortical | 10 | 10 | 3 T, 7 T | T1W | 20 |
| ADNI | 47 | 47 | 1.5 T, 3 T | T1W | 94 |
| Penn | 30 | 53 | 3 T, 7 T | T1W (30 pairs), T2W (23 pairs) | 106 |
| Hippocampal | 20 | 40 | 3 T, 7 T | T1W, T2W | 80 |
| Vim | 10 | 10 | 3 T, 7 T | T1W | 20 |
| Total | 161 | 259 | 518 |
Image format
Images are full-volume float32 NIfTI files (.nii.gz) on the MRIxFields2026 reference grid: shape (364, 436, 364), spacing 0.5 mm isotropic. This is the resampled spacing, not the native acquisition resolution. The affine is:
[[0.5, 0, 0, -91.5],
[0, 0.5, 0, -126.0],
[0, 0, 0.5, -72.0],
[0, 0, 0, 1.0]]
Source-specific alignment is followed by masked N4 bias correction and in-mask min-max normalization to [0, 1]. Background voxels are set to zero. N4 uses a shrink factor of 2, iterations [50, 50, 30, 20] and convergence threshold 1e-6.
Code and MRIxFields interface
Preprocessing and adapter code: PeirongLiL/MRI-PairedField.
Download
Use Hugging Face for direct or source-selective downloads. Use the Zenodo record for the immutable v1.0.0 archival packages, package manifest and checksums.
pip install huggingface_hub
hf download lipeirong/MRI-PairedField --type dataset --local-dir ./MRI-PairedField-data
To download one source only:
hf download lipeirong/MRI-PairedField --type dataset --include 'data/unc/**' --include '*.tsv' --local-dir ./MRI-PairedField-data
pairs.tsv has one row per logical volume record, including ADNI records. availability.tsv states whether each record is public or controlled and whether its file is present. The fields in pairs.tsv are volume_id (unique volume identifier and filename stem), source_dataset (machine-readable source code), pair_id (paired-case identifier), modality, field_strength, paired_field_strength, relative_path, and case_group (paired-case partitioning key). Two rows with the same source_dataset, pair_id and modality form an image pair and must have reciprocal field-strength values. Keep all rows with the same case_group together when partitioning the data.
The 424 public image files occupy approximately 20.36 GB in compressed NIfTI form. The .nii.gz files are meant to be downloaded and read with NiBabel or SimpleITK; the Hub's tabular viewer does not display their three-dimensional contents.
The adapter supports bidirectional, same-modality records and axial slices z=72:292 (zero-based, end-exclusive). Optional 15-slice windows are a training interface; the dataset retains complete volumes. Physical 64 mT acquisitions are documented as 64 mT, with an adapter mapping for software that uses the label 0.1T.
Source datasets and use conditions
The release contains processed derivatives of the following sources. The source label in the first column is the human-readable name used in the manuscript; source_dataset in pairs.tsv is the stable machine code. Access, redistribution and citation remain subject to the original source terms. ADNI is controlled-access and must be obtained through the ADNI application and data-use agreement; this release does not redistribute ADNI raw files. For every other source, users should check the linked record for its current licence or access requirements and cite the linked paper or dataset record.
| Manuscript label | source_dataset |
Official source and released subset | Use condition |
|---|---|---|---|
| ULF | openneuro_hfc |
Ultra-low-field brain MRI: test–retest reliability and correspondence to high-field MRI; OpenNeuro ds006557 v1.0.2. Same-day HFC 64 mT/3 T, T1W and T2W. | Follow OpenNeuro and source-publication terms; cite the study. |
| LUMC | zenodo_20281403 |
Paired 64mT and 3T Brain MRI Scans of Healthy Subjects for Neuroimaging Research; 11 matched cases, T1W/T2W/T2-FLAIR. | Follow the Zenodo record and source-publication terms; cite the record. |
| UNC | unc |
A paired dataset of T1- and T2-weighted MRI at 3 Tesla and 7 Tesla; 10 paired cases, T1W/T2W. | Follow the linked dataset and publication terms; cite both. |
| Subcortical | osf |
3 versus 7 Tesla MRI for parcellations of subcortical brain structures; 10 T1W pairs. | Follow OSF and publication terms; cite the study. |
| ADNI | adni |
Alzheimer’s Disease Neuroimaging Initiative; 47 selected T1W 1.5 T/3 T pairs. | ADNI controlled access and data-use agreement apply; do not redistribute raw ADNI files. |
| Penn | penn |
Penn 3T-7T Paired Epilepsy Imaging; 30 T1W and 23 T2W pairs. | Follow Pennsieve and source-publication terms; cite the record. |
| Hippocampal | figshare_26075713 |
A paired dataset of multi-modal MRI at 3 Tesla and 7 Tesla with manual hippocampal subfield segmentations; 20 T1W/T2W pairs. | Follow Figshare and publication terms; cite the study. |
| Vim | lausanne_zenodo1438358 |
Dataset comparison of MRI-based automated segmentation methods and functional neurosurgery targeting with direct visualization of the ventro-intermediate thalamic nucleus at 7 T; 10 T1W pairs. | Follow Zenodo and publication terms; cite the record and study. |
The processed public files in this repository are the released derivatives. They are not a replacement for obtaining permission to access the original raw sources. See LICENSES.md for the source-specific terms; no blanket licence is asserted over all image files.
- Ultra-low-field brain MRI: test–retest reliability and correspondence to high-field MRI (OpenNeuro ds006557 v1.0.2)
- Paired 64mT and 3T Brain MRI Scans of Healthy Subjects for Neuroimaging Research (Zenodo 20281403)
- A paired dataset of T1- and T2-weighted MRI at 3 Tesla and 7 Tesla (UNC)
- 3 versus 7 Tesla MRI for parcellations of subcortical brain structures (OSF 4nrku)
- Alzheimer’s Disease Neuroimaging Initiative (ADNI): MRI methods
- Penn 3T-7T Paired Epilepsy Imaging
- A paired dataset of multi-modal MRI at 3 Tesla and 7 Tesla with manual hippocampal subfield segmentations
- Dataset comparison of MRI-based automated segmentation methods and functional neurosurgery targeting with direct visualization of the ventro-intermediate thalamic nucleus at 7 T (Zenodo 1438358)
Versioning and DOI
The citable archival data release is MRI-PairedField v1.0.0 at https://doi.org/10.5281/zenodo.22816860. The Hugging Face v1.0.1 tag standardizes the repository name and metadata field names; it does not change the released image files.
Citation
Please cite the Zenodo dataset record, the accompanying manuscript when available, and the original publication or data record for each source used:
Li, P., Zhu, H., Wu, X. & Yuan, Z. (2026). MRI-PairedField: a standardized multi-contrast participant-matched brain MRI dataset spanning four field strengths (Version 1.0.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.22816860
@dataset{li_2026_mri_pairedfield,
author = {Li, Peirong and Zhu, Haibo and Wu, Xiangbo and Yuan, Zengbei},
title = {MRI-PairedField: a standardized multi-contrast participant-matched brain MRI dataset spanning four field strengths},
year = {2026},
publisher = {Zenodo},
version = {1.0.0},
doi = {10.5281/zenodo.22816860},
url = {https://doi.org/10.5281/zenodo.22816860}
}
- Downloads last month
- 113