NucVerse3D: 3D Nuclei Segmentation in Multi-Modal Microscopy Using a Residual Attention U-Net

NucVerse3D is a collection of volumetric microscopy datasets, trained models, and segmentation results for generalized 3D nuclear instance segmentation. The NucVerse3D framework and associated codebase are publicly available at https://github.com/Segovia-lab/3D-Nuclei-segmentation.git This record includes newly released datasets with manual 3D nuclear instance annotations—mouse liver tissue, mouse liver hepatocellular carcinoma, and Drosophila melanogaster brain glia—as well as preprocessed 3D patches used to train two types of models: dataset-specific models, trained using data from a single dataset, and generalized models, trained by aggregating data from multiple datasets spanning different species, tissues, resolutions, and imaging modalities. In addition, trained model weights and segmentation outputs for all evaluated datasets are provided. For external benchmark datasets, only the processed patches, trained models, and segmentation outputs used in this study are included, while the corresponding raw data remain available through their original repositories.