CATENA

Overview of CATENA

CATENA is an end-to-end, developer-friendly pipeline for large-scale connectomics—engineered to train and evaluate on terabyte-scale EM datasets. It integrates state-of-the-art Funke-lab components for neuron segmentation (Local Shape Descriptors; Sheridan et al., 2022), synapse detection (Synful; Buhmann et al., 2020), microtubule tracking (Micron; Eckstein et al., 2019), and neurotransmitter classification (Synister; Eckstein, Bates et al., 2024), alongside EM-to-EM domain adaptation, mitochondria segmentation, tissue vs. non-tissue masking, and robust pre-/post-processing tools.

CATENA brings together tools and models, including some state-of-the-art models for large-scale connectomics under one hood Designed for technically proficient users, each module remains decoupled and self-contained, yet collectively they lower barriers with elaborate documentation, default examples, and error reporting from our own trials, making advanced connectomics accessible without sacrificing flexibility.

PLEASE NOTE THIS IS UNDER HEAVY DEVELOPMENT. FOLLOW DEV BRANCH LINKS BELOW!

📦 Models, Packages and Tools: - Neuron Segmentation Local Shape Descriptors (Sheridan et al. 2022): Installation and Usage - Synapse Detection Synful(Buhmann et al. 2020): Installation and Usage - Microtubule tracking Micron (Eckstein et al. 2019): Installation and Usage [TO BE INTEGRATED WITH CATENA SOON..]

[!WARNING] Microtubule Tracking uses TENSORFLOW 1.x and Gurobi dependencies for ILP - Neurotransmitter classification Synister (Eckstein, Bates et al. 2024): Installation and Usage - Mitochondria segmentation using MONAI and adapted Residual UNets for isotropic FIBSEM data from Xie et al. - EM Tissue/No-Tissue Mask generation models and conventional CV pipelines. - Generative AI for EM-to-EM translation: TO BE ADDED

• For visualisation: Napari and Neuroglancer

🛠️ Detailed Features: - Pytorch implementations of LSDs and Synful. - Exploration of LSDs and Synful for other task objectives. - Docker-based containerisation and release of development environments. - Style transfer and domain adaptation with Generative AI models. - Mitochondria segmentation pipelines that use both LSDs and MONAI - MONAI Tissue vs Non-Tissue detection pipelines - Large scale data analysis over public and local EM datasets. - Artefact logging with Weights and Biases.

Please check Issues for basic troubleshooting tips. Kindly note these packages are being tested gradually and not all issues have made it to the list yet.

References

The pipeline has been built upon pre-existing work: - Local Shape Descriptors: Github, Paper - Synful: GitHub, Paper - Micron: Github, Paper - Synister: GitHub, Paper - Generative AI: To do

Citations

If you use this codebase, please cite us. However, please do not forget to cite the original authors of the algorithms/models. @software{Mohinta_Catena_Neuron_Segmentation_2022, author = {Mohinta, Samia}, month = aug, title = {{Catena: Neuron Segmentation, Synapse Detection, Microtubule tracking and more...}}, version = {0.1}, year = {2022} }

Funding

This work has been supported by generous funding from:

• Symons MCR Conference Fund
• Hugh Paton - JP Morgan Bursaries
• Dr Teresa Tiffert Research Innovation Award

Usage Collaborations

This work is being used in other institutes:

💬 What People Are Saying About Catena

Winding Lab, The Crick

"I was very positively surprised by the quality of the segmentations, especially given that the model had not been trained on our data and that only minimal enhancement was applied to the EM images. Larger spines, in particular, are segmented with incredible precision and the identities of individual neurons appear to be well maintained across the z-plane. I was especially impressed to see the model perform well even on noisier regions with low contrast or staining residue in the intracellular space. There are occasional minor errors around small dendritic spines, so I’m very excited to see how the model performs on a dataset that has not undergone the full suite of preprocessing steps. -- Anna Seggewisse " View on X →

💥 Research Outputs

🤝 Conferences

• Berlin Connectomics 2024, MPI Berlin, Germany - accepted for Poster Presentation
• UK Neural Computation 2024, Sheffield University, Sheffield UK - accepted for Poster Presentation
• UCL NeuroAI 2024, UCL, London UK - accepted for Poster Presentation
• AI Revolution Meets 4D Cellular Physiology March 2025, HHMI Janelia, USA - accepted for Poster Presentation
• Analysis and Modelling of Connectomes June 2025, HHMI Janelia, USA - accepted for Poster Presentations