MFNet whole-brain neuronal cell-type atlas (mouse, CCFv3, 10 µm isotropic)

This dataset provides a deep learning–derived whole-brain neuronal cell-type atlas of the mouse brain generated using the Multi-Diffusion-MRI Fusion Network (MFNet). The atlas integrates high-resolution ex vivo diffusion MRI with light-sheet microscopy (LSM)–based neuronal segmentation and cell-type annotations from the Allen Brain Cell (ABC) Atlas, all registered to the Allen Mouse Brain Common Coordinate Framework (CCFv3). The atlas is provided at 10 µm isotropic resolution and includes voxel-wise annotations for three hierarchical levels of cellular organization: neuronal type (i.e., excitatory (E) and inhibitory (I)), cell neighborhood, and fine-grained cell class. Diffusion MRI features were used to predict cell types, while NeuN-stained LSM data provided spatial localization of neuronal elements. Model training and validation were performed using ABC Atlas neuronal cells as ground truth, and the trained model was applied to generate the full 3D atlas. All data are provided in NIfTI format and can be visualized or analyzed using standard neuroimaging tools such as ITK-SNAP, FSL, MRtrix3, or Python libraries (e.g., NiBabel, Nilearn).