This notebook provides a (hopefully) comprehensive tutorial on processing volumetric electron microscopy (EM) data for cell segmentation using deep learning.
Specifically, this notebook will guide you through:
-
Loading and Visualizing Volumetric EM Data: Learn how to handle large 3D datasets from various formats.
-
One-Shot Prediction with BioImage.IO Models: Use pre-trained models from the BioImage.IO model zoo for boundary prediction.
-
Fine-Tuning on Ground-Truth Data: Adapt the model to your specific dataset using labeled training cubes.
-
Robust Pipeline Implementation: Best practices for reproducibility, error handling, and bioimage.io compliance.
-
Load and preprocess 3D EM volumes from various file formats
-
Utilize community-shared models from BioImage.IO
-
Implement fine-tuning workflows for domain adaptation & task transfer
-
Evaluate segmentation quality and export results
-
Package models according to BioImage.IO standards
***Note - Make sure you are in the directory you want to download this repository to...
Clone this repository to your machine:
git --clone https://github.com/kreshuklab/zoocell_seg_ws_2026.git
Move into the repository directory
cd ./zoocell_seg_ws_2026
Set up the environment we will use for this pipeline/workflow
conda env create --file environment.yml
- BioImage.IO: https://bioimage.io/ - Model zoo and specifications
- ELF: https://github.com/constantinpape/elf - Segmentation algorithms
- PyTorch: https://pytorch.org/ - Deep learning framework
- BioIO: https://github.com/bioio-devs/bioio - Modern bioimage I/O
- CebraNet: - available in the Bioimage Model Zoo (bioimage.io) // (CebraNET @bioimage.io, CebraNET @zenodo)
Remember: The field of bioimage analysis is rapidly evolving. Stay updated with the latest models and techniques from the BioImage.IO community!
This notebook was created for educational purposes. For production use, consider additional validation, error handling, and performance optimization.