Overview

Schema Version

https://atlas.rsna.org/schemas/2025-11/dataset.json

Name

fastMRI brain subset (NYU, 2018) used in ZSNET study

Link

https://arxiv.org/abs/1811.08839

Indexing

Keywords: fastMRI, brain MRI, 2D spin-echo, T1-weighted, T2-weighted, FLAIR, acceleration, undersampling, reconstruction, deep learning

Content: MR, NR, RS

RadLex: RID10746, RID10312, RID10721, RID12726, RID35806, RID12692, RID10795, RID10794, RID12681

Organization(s)

New York University (NYU) Langone Health

Meta AI (Facebook)

Friedrich-Alexander Universität Erlangen-Nürnberg

Funding

Supported by NIH grants R01EB024532 and P41EB017183 (as acknowledged in the article).

Ethical review

Retrospective study with local IRB approval; written informed consent waived due to minimal risk and use of de-identified data.

Comments

Retrospective study on the brain subset of the fastMRI dataset collected at New York University in 2018; data de-identified by NYU. Models trained on 5847/6405 images; standard fastMRI brain splits used. IRB approved with consent waived.

Date

Published: 2022-11-02

Created: 2018

References

[1] Radmanesh A, Muckley MJ, Murrell T, Lindsey E, Sriram A, Knoll F, Sodickson DK, Lui YW. "Exploring the Acceleration Limits of Deep Learning Variational Network–based Two-dimensional Brain MRI". Radiology: Artificial Intelligence. 2022. doi:10.1148/ryai.210313. PMID: 36523647. PMCID: PMC9745443.

[2] Zbontar J, Knoll F, Sriram A, et al.. "fastMRI: An open dataset and benchmarks for accelerated MRI". arXiv 1811.08839. 2018. Available from: https://arxiv.org/abs/1811.08839

[3] Knoll F, Zbontar J, Sriram A, et al.. "fastMRI: A Publicly Available Raw k-Space and DICOM Dataset of Knee Images for Accelerated MR Image Reconstruction Using Machine Learning". Radiology: Artificial Intelligence. 2020. doi:10.1148/ryai.2020190007. PMID: 32076662. PMCID: PMC6996599.

Dataset

Motivation

To explore limits of deep learning–based reconstruction for accelerated 2D brain MRI and identify useful acceleration ranges.

Sampling

Retrospective undersampling of fully sampled raw multicoil k-space for acceleration experiments.

Partitioning scheme

Standard fastMRI brain splits: training, validation, and test with no patient overlap.

Missing information

Patient counts, demographics, and site counts are not specified; only volume counts are provided.

Relationships between instances

Volumes include normal and abnormal clinical studies; no patient overlap across splits.

Noise

Not specified at dataset level; retrospective undersampling used for experiments.

External data

Out-of-distribution experiments also used fastMRI knee data and pure noise inputs (for model evaluation).

Confidentiality

De-identified clinical MRI data; IRB-approved use with consent waived.

Re-identification

Data were de-identified by NYU.

Sensitive data

Medical imaging data with PHI removed (de-identified).