Biom3d

Biom3d provides a modular, PyTorch-based framework that enables reproducible and interoperable 2D and 3D semantic segmentation pipelines for bioimage analysis while adhering to FAIR principles.

Key Features:

• Modular Architecture: The framework is organized around seven core module types enabling fine-grained control over data handling, model configuration, optimization, and evaluation.
• Auto-configuration: Its default configuration, nnCore, autoconfigures optimal pipelines based on new datasets.
• Model Interoperability: The architecture supports seamless substitution of transformer-based MONAI models.
• OMERO Compatibility: The framework integrates with OMERO (Open Microscopy Environment Remote Objects) for data management and reproducible computation including execution on HPC servers.

Scientific Applications:

• 3D Bioimage Segmentation: Biom3d outperforms classical tools such as NucleusJ/NODeJ and matches the robustness of nnU-Net across diverse imaging modalities.
• Advanced Bioimage Analysis: The framework supports building, sharing, and reusing advanced bioimage analysis solutions through its extensible architecture.

Methodology:

Biom3d implements deep learning-based 2D and 3D semantic segmentation using a PyTorch-based architecture with seven core module types for data handling, model configuration, optimization, and evaluation.