fastMRI Dataset

About Us

Here at the Center for Advanced Imaging Innovation and Research (CAI²R), in the Department of Radiology at NYU School of Medicine and NYU Langone Health, we bring people together to create new ways of seeing. We are committed to the translation of new imaging techniques and technologies into clinical practice, for the improvement of human health. In particular, we are pushing the boundaries of rapid image acquisition and advanced image reconstruction, with the aim of providing uniquely valuable biomedical information to advance the understanding of disease and improve the care of patients.

fastMRI

We are partnering with Facebook AI Research (FAIR) on fastMRI – a collaborative research project to investigate the use of AI to make MRI scans up to 10X faster.

NYU Langone and FAIR are providing open-source AI models, baselines, and evaluation metrics.

The Dataset

The deidentified imaging dataset provided by NYU Langone comprises raw k-space data in several sub-dataset groups. Curation of these data are part of an IRB approved study. Raw and DICOM data have been deidentified via conversion to the vendor-neutral ISMRMD format and the RSNA clinical trial processor, respectively. We also performed manual inspection of each DICOM image for the presence of any unexpected protected health information (PHI), with spot checking of both metadata and image content.

Knee MRI: Data from more than 1,500 fully sampled knee MRIs obtained on 3 and 1.5 Tesla magnets and DICOM images from 10,000 clinical knee MRIs also obtained at 3 or 1.5 Tesla. The raw dataset includes coronal proton density-weighted images with and without fat suppression. The DICOM dataset contains coronal proton density-weighted with and without fat suppression, axial proton density-weighted with fat suppression, sagittal proton density, and sagittal T2-weighted with fat suppression. The exact distribution of contrasts is given in table 1. Please note that this table does not include stats about the data that was originally held back for the fastMRI reconstruction challenge.

Field Strength
PD	697
PD with fat suppression	701
Total	1398

Table 1: Number of scans for the different contrasts of the knee raw dataset.

Brain MRI: Data from 6,970 fully sampled brain MRIs obtained on 3 and 1.5 Tesla magnets. The raw dataset includes axial T1 weighted, T2 weighted and FLAIR images. Some of the T1 weighted acquisitions included admissions of contrast agent. The exact distribution of contrasts and field strengths is given in table 2.

Field Strength	1.5T	3T
T1	382	409
T1 post contrast	849	646
T2	1655	2524
FLAIR	126	411
Total	3012	3990	7002

Table 2: Number of scans for the different contrasts and scanner field strengths of the brain raw dataset.

Prostate MRI: Data from 312 prostate MRI exams obtained on 3 Tesla magnets. The raw dataset includes axial T2-weighted and axial diffusion-weighted images for each of the 312 exams. For more information about the prostate dataset, please refer to our article https://www.nature.com/articles/s41597-024-03252-w and code repository https://github.com/cai2r/fastMRI_prostate

Breast MRI: Data from 300 clinical breast MRI exams obtained on 3 Tesla magnets. The raw dataset includes axial DCE-MR using a 3D GRASP sequence for each of the 300 exams For more information about the breast dataset, please refer to our article https://arxiv.org/abs/2406.05270 and code repository https://github.com/eddysolo/demo_dce_recon

Apply for Access

The application process includes acceptance of the Data Sharing Agreement (found below) and submission of an online application form. The application must include the investigator’s institutional affiliation and the proposed uses of the data. NYU fastMRI data may be used for internal research or educational purposes only as described in the data use agreement and may not be redistributed in any way without prior permission.
Read and agree to the data use agreement below to apply for access.

Note: Each dataset has its own Data Sharing Agreement. Obtaining access to and using NYU fastMRI data requires adherence to the NYU fastMRI Data Sharing Agreement and the publication policies outlined in the documents listed above.

Use of this Dataset

Interested scientists may apply for access to fastMRI data for the purposes of internal research or education only.* Access is contingent on adherence to the fastMRI Dataset Sharing Agreement shown below, which also outlines policies for publication and citation. Note: This agreement is subject to updates

Contact Us

fastmri@med.nyu.edu

For permission to publish images from the dataset,
To report any violations of the Data Sharing Agreement,
Should you discover protected health information (PHI). Note that all metadata previously containing PHI have been stripped, and fields requiring entries have been assigned anonymized dummy characters which no longer represent PHI.

Lead investigators with regard to the dataset:
Florian Knoll, PhD
Patricia M. Johnson, PhD
Daniel K. Sodickson, MD PhD
Michael P. Recht, MD
Yvonne W. Lui, MD

*Use of the fastMRI dataset for creating and submitting benchmark results for publication on an MLCommons™ benchmark, such as an MLPerf™ benchmark, is considered non-commercial use of fastMRI. It is further considered non-commercial use of fastMRI to, in locations other than and in addition to mlcommons.org, (i) republish results published on mlcommons.org, and (ii) create and publish unverified benchmark results consistent with the relevant MLCommons rules. Please note that you must adhere to the fastMRI data usage guidelines and appropriate MLCommons policies such as the Trademark Guidelines, Terms of Use, and Result Guidelines at https://mlcommons.org/en/policies/ and include appropriation citations in the aforementioned publications. Additionally, please note that a non-commercial use of the dataset is not necessarily a non-commercial use of the MLPerf™ benchmarks.

NYU Langone Health fastMRI Dataset Sharing Agreement

By registering for downloads from the fastMRI Dataset, I agree to this Dataset Sharing Agreement, as well as to the terms of use as posted and updated periodically at: https://nyulangone.org/policies-disclaimers/disclaimer.)

The fastMRI Dataset is considered proprietary to and owned by New York University and NYU Langone Health (together “NYU”). Other than the rights granted herein, NYU retains all rights, title, and interest in the fastMRI Dataset.
Subject to the provisions of this Agreement, NYU shall give to me access to and the right to download the fastMRI Dataset, and NYU hereby grants to me a non-exclusive, royalty-free license to use the fastMRI Dataset for internal research or educational purposes only and only as permitted by this Agreement. This Agreement conveys no other rights of any sort with respect to the fastMRI Dataset or the intellectual property rights embodied therein.
I will receive a download link to access the fastMRI Dataset without charge for internal research or educational purposes only. The link will permit me to download a verbatim copy of the fastMRI Dataset solely for such use. I will NOT SHARE THE DOWNLOAD LINK to the fastMRI Dataset with others. If another user within my organization or elsewhere wishes to obtain a copy of and use the fastMRI Dataset, they must register as an individual user and comply with all the terms of this Agreement.
I will use the fastMRI Dataset for internal research or educational purposes only and only as permitted by this Agreement.
I further agree to:
1. Use the fastMRI Dataset in compliance with all laws, governmental regulations and guidelines, including obtaining all approvals for the proposed use of the fastMRI Dataset to comply with federal, state, local and institutional policies and regulations.
2. Not SELL OR OTHERWISE MONETIZE any portion or all of the fastMRI Dataset.
3. Not DISTRIBUTE, PUBLISH, REPRODUCE, RETRANSMIT, COMMERCIALLY EXPLOIT, COPY OR OTHERWISE TRANSFER any portion or all of the fastMRI Dataset or any subsequent variables or data files derived from the fastMRI Dataset to any person or entity other than individuals under my direct supervision and other than for uses in academic publications and presentations with citations as provided in paragraph 6 below. I will ensure that no one will be allowed to take or send fastMRI Dataset to any other location, unless prior written permission is obtained from NYU.
4. Require anyone who receives or has access to the fastMRI Dataset to agree to the same restrictions and conditions on the use and/or disclosure of the fastMRI Dataset that apply to me.
5. Maintain the security and confidentiality of the fastMRI Dataset and use appropriate safeguards to prevent any unauthorized use or disclosure of the fastMRI Dataset.
I agree all publications and presentations resulting from any use of the fastMRI Dataset must cite use of the fastMRI Dataset as follows:
1. In any published abstract, I will cite “NYU fastMRI” as the source of the data in the abstract.
2. In any published manuscripts using data from NYU fastMRI, I will reference the following paper: Knoll et al Radiol Artif Intell
  . 2020 Jan 29;2(1):e190007.
  doi: 10.1148/ryai.2020190007.
  https://pubs.rsna.org/doi/10.1148/ryai.2020190007
  and the arXiv paper, https://arxiv.org/abs/1811.08839.
  
  I will include language similar to the following in the methods section of my manuscripts in order to accurately acknowledge data source: “Data used in the preparation of this article were obtained from the NYU fastMRI Initiative database (fastmri.med.nyu.edu).[citation of Knoll et al Radiol Artif Intell
  . 2020 Jan 29;2(1):e190007.
  doi: 10.1148/ryai.2020190007. (https://pubs.rsna.org/doi/10.1148/ryai.2020190007), and the arXiv paper: https://arxiv.org/abs/1811.08839] As such, NYU fastMRI investigators provided data but did not participate in analysis or writing of this report. A listing of NYU fastMRI investigators, subject to updates, can be found at:fastmri.med.nyu.edu.The primary goal of fastMRI is to test whether machine learning can aid in the reconstruction of medical images.” I acknowledge that, depending upon the length and focus of the manuscript, it may be appropriate to include more or less than the above, and further acknowledge that inclusion of some variation of the language shown above is mandatory.
  
  Other than citations as provided in this paragraph 6, I will not use the name, symbol, or mark of NYU or any of its affiliates or of the fastMRI Dataset in connection with any products, promotion, financing, advertising, or sales literature or in any form of publicity without the prior written approval of NYU.
The fastMRI Dataset has been collected from human subjects. I should not receive any protected health information (PHI). All metadata in the fastMRI Dataset has been de-identified and anonymized using dummy numbers and no longer represents PHI. If I suspect that certain data includes PHI, I will not use the PHI for any purpose, I will immediately destroy such data, and I will immediately contact the NYU fastMRI principal investigator listed at fastmri.med.nyu.edu.
I will make no attempt to use the information in the fastMRI Dataset to identify the individual subjects or to contact such individuals or link the fastMRI Dataset with any other dataset.
Except as contemplated by paragraph 7 above, I will not contact NYU or anyone at NYU concerning any individual subjects or seeking information about them.
The fastMRI Dataset has not been reviewed or approved by the Food and Drug Administration, and is not provided for clinical use. In no event shall any data or images in the fastMRI Dataset or generated through the use of the fastMRI Dataset be used or relied upon in the diagnosis or provision of patient care.
I will report to NYU any use or disclosure of the fastMRI Dataset that is not permitted by this Agreement or required by law within 5 business days of becoming aware of such use or disclosure.
When I have completed my work using the fastMRI Dataset, I will destroy all copies of the fastMRI Dataset and retain no copies thereof.
I understand that failure to abide by the terms of this Agreement will result in immediate termination of use as well as privileges to access the fastMRI DataSet. In the event that NYU determines that I have violated this Agreement or other impermissible use has been made, NYU may direct that I immediately destroy all copies of the fastMRI Dataset and retain no copies thereof even if I did not cause the violation or impermissible use.
The fastMRI dataset is provided “as is”. NYU MAKES NO WARRANTIES, EXPRESS OR IMPLIED, INCLUDING ANY WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE OR OF NON-INFRINGEMENT, AND HEREBY DISCLAIMS THE SAME.
In no event shall NYU, or NYU partners be liable for any use by me of the fastMRI Dataset resulting in any loss, claim, damage or liability, of whatsoever kind or nature, which may arise from or in connection with this AGREEMENT or the use, handling or storage of the fastMRI Dataset.
I agree to indemnify, defend, and hold NYU and its affiliates and affiliated hospitals, and each of their trustees, directors, officers, representatives, partners, faculty, students, volunteers, employees, or agents (each, an “NYU Indemnitee”) harmless from any claims, losses or damages, including legal fees, arising out of or resulting from my access to and/or use of the fastMRI Dataset or any failure to abide by the terms of this Agreement, except to the extent caused by an NYU Indemnitee.
This Agreement shall be governed by and interpreted in accordance with the laws of the State of New York. The exclusive venue for any dispute relating to this Agreement shall be in the state or federal courts within New York, New York.