sphero-vem

Python library for quantitative analysis of volumetric electron microscopy (vEM) data.

sphero-vem was developed for the end-to-end analysis pipeline described in:

Bottone et al., 3D Reconstruction of Nanoparticle Distribution in Tumor Spheroids with Volume Electron Microscopy, Preprint: https://doi.org/10.64898/2026.04.17.719153

While the library was originally developed for SBF-SEM data of nanoparticle-loaded tumor spheroids, the individual components are designed to be reusable for other vEM datasets and workflows.

Capabilities

• Denoising: self-supervised Noise2Void via CAREamics on large zarr volumes
• Registration: intensity-based pairwise slice alignment with multi-resolution PyTorch optimization
• Segmentation: fine-tuning and inference with Cellpose-SAM for cells and nuclei; empirical Bayes approach for nanoparticles
• Shape analysis: 3D morphological descriptors via signed distance functions and mesh-based curvature (mean curvature, curvedness, shape index, fractal dimension)
• Spatial analysis: nanoparticle-to-nucleus distance quantification per cell
• Data management: zarr-native I/O with OME-NGFF multiscale support and processing metadata tracking

Installation

The library is available on PyPI. We recommend installing it in a dedicated virtual environment.

pip install sphero-vem

For full CUDA 12.x GPU acceleration (Linux only; Windows untested):

pip install "sphero-vem[cuda]"

PyTorch-based stages (denoising, registration, Cellpose-based segmentation) support GPU execution via PyTorch's native device management, including CUDA and MPS (Apple Silicon), with no additional dependencies.

Array operation stages (nanoparticle segmentation, shape analysis, spatial analysis) additionally support CUDA acceleration via CuPy and CuCIM, available with the cuda extra above.

Development install

Clone the repository and install with Poetry:

git clone https://github.com/dv-bt/sphero-vem.git
cd sphero-vem
poetry install           # base install
poetry install -E cuda   # with CUDA 12.x GPU acceleration (Linux only; Windows untested)

Documentation

Full documentation is available on the official website.

Requirements

• Python 3.12–3.13
• PyTorch ≥ 2.0
• NumPy ≥ 2.0
• zarr ≥ 3.0
• Cellpose ≥ 4.0
• CAREamics ≥ 0.0.9

Dataset and model weights

The annotated SBF-SEM dataset used to develop this pipeline is available at BioImage Archive (https://doi.org/10.6019/S-BIAD3263). Fine-tuned Cellpose-SAM model weights are available on Zenodo (https://doi.org/10.5281/zenodo.19616546).

Citation

If you use this library, please cite the accompanying paper:

@article {Bottone2026,
	author = {Bottone, Davide and Gerken, Lukas RH and Habermann, Sebastian and Mateos, Jose Maria and Lucas, Miriam S and Riemann, Johannes and Fachet, Melanie and Resch-Genger, Ute and Kissling, Vera M and Roesslein, Matthias and Gogos, Alexander and Herrmann, Inge K},
	title = {3D Reconstruction of Nanoparticle Distribution in Tumor Spheroids with Volume Electron Microscopy},
	year = {2026},
	doi = {10.64898/2026.04.17.719153},
	eprint = {https://www.biorxiv.org/content/early/2026/04/21/2026.04.17.719153},
}

License

See LICENSE for details.