Related Experiment Video
Updated: Dec 28, 2025
Four-Dimensional CT Analysis Using Sequential 3D-3D Registration
Published on: November 23, 2019
fastMRI: A Publicly Available Raw k-Space and DICOM Dataset of Knee Images for Accelerated MR Image Reconstruction
Florian Knoll1, Jure Zbontar1, Anuroop Sriram1
1Department of Radiology, NYU School of Medicine, 650 First Ave, New York, NY 10016 (F.K., M.J.M., M.B., M.P., K.J.G., J.K., H.C., D.W., N.Y., M.P.R., D.K.S., Y.W.L.); Department of Artificial Intelligence Research, Facebook, Menlo Park, Calif (J.Z., A.S., J.P., E.O., C.L.Z.); Department of Artificial Intelligence Research, Facebook, New York, NY (A.D.); Center for Data Science, New York University, New York, NY (K.J.G.); Department of Artificial Intelligence Research, Facebook, Montreal, Canada (M.D., A.R., M.R., P.V.); and Department of Computer Science, University of Florida, Gainesville, Fla (Z.Z.).
A new dataset of knee MRI scans is now available for researchers. This data supports the development of faster magnetic resonance imaging (MRI) reconstruction techniques using machine learning.
Area of Science:
- Medical Imaging
- Machine Learning
- Data Science
Background:
- Accelerated Magnetic Resonance Imaging (MRI) is crucial for reducing scan times and improving patient comfort.
- Machine learning (ML) offers promising avenues for enhancing MR image reconstruction speed and quality.
- Access to comprehensive, high-quality datasets is essential for developing and validating ML-based reconstruction algorithms.
Purpose of the Study:
- To present a publicly available dataset for accelerated MR image reconstruction.
- To provide both raw k-space data and processed Digital Imaging and Communications in Medicine (DICOM) images of knee joints.
- To facilitate research and development in ML-driven accelerated MRI.
Main Methods:
- Dataset compilation included acquiring k-space data and corresponding DICOM images of knee MRIs.
- Data was curated for use in machine learning model training and validation.
- Standardized formats were employed for broad accessibility.
Main Results:
- A comprehensive dataset containing k-space and DICOM images of knee MRIs has been successfully compiled.
- The dataset is structured to support the development of accelerated MR image reconstruction algorithms.
- Public availability enables reproducible research and fosters innovation in the field.
Conclusions:
- The release of this dataset represents a significant resource for the medical imaging and machine learning communities.
- It is expected to accelerate the development and clinical translation of advanced accelerated MRI techniques.
- Further research utilizing this dataset will drive progress in efficient and high-quality MR imaging.
