2026-10-14T07:20:20Z https://www.scidb.cn/oai

10.57760/sciencedb.nb.00057 2026-07-08T20:38:48Z

2026-07-08 MMB-MRI: A High-resolution Multimodal Mouse Brain MRI Dataset with Refined Atlas-derived Anatomical Labels (Version 1.0). doi:10.57760/sciencedb.nb.00057 en MMB-MRI is a high-resolution multimodal mouse brain magnetic resonance imaging dataset designed for public reuse in mouse neuroimaging, atlas-based analysis, image registration, automated segmentation, and multimodal representation learning. The dataset contains MRI data from 75 C57BL/6J mouse brains acquired between September 2019 and November 2019, and the public release was prepared in 2026. All volumes provide whole-brain coverage and were reconstructed at approximately 0.07 mm isotropic spatial resolution, with image dimensions of 165 x 230 x 135 voxels. MRI acquisition was performed on an 11.7T MRI scanner using a three-dimensional multi-gradient-echo sequence. The acquisition protocol generated complementary MRI contrasts that capture structural, relaxation-related, susceptibility-related, and magnitude information. The 47-case annotated cohort includes six modalities for each case: quantitative susceptibility mapping (`QSMs.nii`), R2* mapping (`R2s.nii`), T1-weighted MRI (`T1.nii`), T2-weighted MRI (`T2.nii`), T2*-weighted MRI (`T2s.nii`), and magnitude imaging (`iMag.nii`). The 28-case supplementary cohort includes five modalities for each case: `QSMs.nii`, `R2s.nii`, `T2.nii`, `T2s.nii`, and `iMag.nii`; T1-weighted images are not included for this cohort. All MRI volumes were processed using a standardized workflow that included brain extraction, image registration, N4 bias-field correction, intensity normalization, and quality control. Registration used affine and deformable alignment to improve anatomical correspondence across subjects and modalities.Anatomical labels were derived from the Dorr mouse brain atlas. The original atlas contains 114 anatomical labels; these labels were reviewed and consolidated into a benchmark 50-region label scheme to improve anatomical consistency and robustness for segmentation and quantitative analysis. For the 47-case cohort, the 50-region labels were manually corrected and quality controlled by researchers familiar with mouse brain anatomy. For the 28-case cohort, the released labels were generated by atlas mapping and label remapping. Each case-level label file is named `label_50regions.nii.gz`; label value 0 denotes background, and values 1-50 denote the released anatomical regions. The dataset root also includes `atlas_dorr_114labels.nii`, a shared atlas label volume retaining the original Dorr atlas label scheme for reference.The case-wise directory layout is `dataset_by_case/cohort_47mouse/<case_id>/` and `dataset_by_case/cohort_28mouse/<case_id>/`. Within each case folder, MRI modalities are stored under `images/`, and the corresponding anatomical label map is stored under `labels/label_50regions.nii.gz`. Dataset-level metadata are stored in `metadata/`. The file `metadata/case_manifest.csv` contains one row per case and records the cohort name, case identifier, available modalities, number of modalities, label availability, and label provenance. The files `metadata/label_mapping.csv` provide the mapping from the original 114-label Dorr atlas to the released 50-region benchmark scheme, including the original atlas label identifiers and names, the corresponding benchmark label identifiers and names, mapping status, label composition, and the number of source atlas labels included in each benchmark region.The dataset is distributed in Neuroimaging Informatics Technology Initiative (NIfTI) format. MRI images are uncompressed `.nii` files, while label maps are compressed `.nii.gz` files. These formats can be opened with commonly used neuroimaging and medical image analysis software, including ITK-SNAP, 3D Slicer, ANTs, FSL, MONAI, nibabel, and PyTorch-based workflows. The complete release contains 502 files with a logical size of approximately 4.1 GiB: 422 MRI image volumes, 75 case-level 50-region label maps, one shared 114-label atlas volume, and four documentation or metadata files. Individual MRI images are approximately 9.77 MiB each. The 47-case cohort has a logical size of approximately 2.7 GiB, and the 28-case cohort has a logical size of approximately 1.3 GiB. mouse brain; multimodal MRI; anatomical labels; neuroimaging dataset Yiping Zhang Chuyan Wang Yuting Zhai Shengjie Zhang Qifan Pang Ziyi Wang Ziqi Yu Xiaoyong Zhang PUBLIC https://creativecommons.org/licenses/by/4.0/ dataset Science Data Bank