Synthetic MRI Artifact Dataset

The Synthetic MRI Artifact Dataset provides paired brain magnetic resonance imaging (MRI) data with and without synthetic linear artifacts, including ringing, herringbone, and zipper patterns. The dataset is structured following the Brain Imaging Data Structure (BIDS) specification and is intended to support the development and evaluation of artifact correction methods in MRI, particularly in supervised learning settings where paired clean and corrupted images are required. Synthetic artifacts were generated using controlled transformations in both the Fourier domain and spatial domain. Ringing artifacts were simulated through radial perturbations in frequency space, herringbone artifacts were produced by copying and displacing frequency components with smoothing, and zipper artifacts were generated by introducing localized Gaussian noise with controlled variability, amplitude, and frequency. Each artifact-corrupted image is paired with its corresponding clean version, enabling direct supervised learning. The dataset is organized according to the BIDS standard, where clean anatomical images are stored under sub-XXX/anat/sub-XXX_T1w.nii.gz, while artifact-corrupted images and their associated parameters are provided in the derivatives directory under derivatives/synthetic_artifacts. For each subject, a JSON file accompanies the artifact image, containing the parameters used for artifact generation, including axis, position, depth, intensity, variability, number of artifacts, and amplitude. This dataset is derived from the Dallas Lifespan Brain Study, and users are required to cite the original work: Park, D. C. et al., "The Dallas Lifespan Brain Study: A Comprehensive Adult Lifespan Data Set of Brain and Cognitive Aging," Scientific Data (2025), available at https://www.nature.com/articles/s41597-025-04847-7. Additionally, users should cite the methodology used to generate the synthetic artifacts when applicable. It is important to note that the artifacts in this dataset are synthetically generated and may not fully capture the variability observed in real clinical settings. Therefore, while the dataset is suitable for benchmarking and methodological development, direct comparisons with other studies may be limited due to differences in data sources and artifact characteristics. The dataset follows the licensing terms of the original Dallas Lifespan Brain Study.