This study aimed to construct a paired low-field and high-field healthy-brain MRI dataset to provide a valuable resource for low-field MRI super-resolution(SR) research.

Postgraduate students enrolled at Beijing University of Posts and Telecommunications from January to March 2026 were recruited and underwent brain MRI scans on both a 0.2 T low-field scanner and a 3.0 T high-field scanner. The low-field protocol comprised three axial sequences—fluid-attenuated inversion recovery(FLAIR), fast spin echo(FSE), and multi-spin-echo(MSSE). The high-field protocol included a magnetization-prepared rapid gradient echo(MPRAGE) sequence with isotropic resolution, as well as diffusion-weighted imaging (DWI). Post-acquisition processing included format conversion, orientation unification, bias-field correction, intensity normalization, automated brain parcellation, and SynthSR-hyperfine SR reconstruction using paired 0.2T MSSE/FSE inputs.

After quality control and selection, a total of 76 imaging volumes were obtained, including complete 0.2 T/3.0 T paired data for 10 subjects and 0.2 T-only data for 2 additional subjects. Thirty-two label brain parcellations were successfully generated for all 10 paired subjects. SR volumes generated from MSSE/FSE pairs showed a certain level of improvement in image quality. Across the 10 paired subjects, mean metrics were as follows: SNR 168.470±60.329, CNR 1.729±0.087.

This dataset provides a multi-sequence paired brain MRI resource at 0.2 T and 3.0 T. It can serve as a benchmark for training and evaluating low-field MRI super-resolution models and for cross-field-strength image quality studies.

Ultra high field diffusion magnetic resonance imaging (dMRI) provides diffusion‐weighted (DW) images with a high signal‐to‐noise ratio, but increases inhomogeneity, which affects the accuracy of dMRI metric reconstruction. Current methods for correcting inhomogeneity rarely consider the accuracy of the reconstructed dMRI metrics. Deep learning models for reconstructing metrics from dMRI signals typically assume that DW images have a homogeneous intensity. To address these challenges, we propose a deep learning model capable of directly reconstructing high‐accuracy dMRI metric maps from inhomogeneous DW images.

An attention‐based q‐space inhomogeneity‐resistant reconstruction network (qIRR‐Net) is proposed for the voxel‐wise reconstruction of diffusion tensor imaging and diffusion kurtosis imaging metrics. A training procedure based on data augmentation and consistency loss is introduced to ensure that the reconstruction results of qIRR‐Net are not affected by signal inhomogeneity. The 3T and 7T dMRI data from the Human Connectome Project are used for model training, testing, and evaluation.

On the 3T dMRI data with simulated inhomogeneity, qIRR‐Net improves the peak signal‐to‐noise ratio by 5.39 and the structural similarity index measure by 0.18 compared with weighted linear least‐squares fitting. On the 7T dMRI data, the metric maps reconstructed by qIRR‐Net not only exhibit clearer tissue structures but also demonstrate greater stability compared with the weighted linear least‐squares results.

The proposed qIRR‐Net enables the accurate reconstruction of dMRI metrics from inhomogeneous DW images. This approach could potentially be expanded to obtain multiple artifact‐free metric maps from ultrahigh field dMRI for neuroscience research and neurology applications.