Development of Voluseg for volumetric segmentation of calcium imaging data
Led by Luiz Tauffer in collaboration with Dr. Mika Rubinov’s laboratory at Vanderbilt University, we integrated modern software engineering practices with Voluseg, a specialized tool for volumetric segmentation of calcium imaging data. This project focused on creating advanced algorithms and tools that enable neuroscientists to efficiently process and analyze three-dimensional calcium imaging datasets, facilitating the study of neural activity patterns in complex brain structures.
Scope of Work
The Rubinov lab, known for its contribution to “Glia Accumulate Evidence that Actions Are Futile and Suppress Unsuccessful Behavior” (Cell, 2019), used the Voluseg pipeline for cell segmentation and component detection in zebrafish studies. Interest from the wider community created demand to turn Voluseg into a mature, user-friendly package, so we brought to it the practices we maintain in tools such as PyNWB and SpikeInterface.
Software Engineering
We refactored the code to follow current practice, with automated formatting, pre-commit hooks, and continuous integration to keep quality and formatting consistent, and comprehensive function docstrings following a standard convention.
Packaging
The project moved from setup.py to pyproject.toml, was tested across the major operating systems and Python versions, and was set up for straightforward distribution through PyPI with continuous deployment.
Testing
We built a test suite covering the major functionality, integrated it into continuous integration through GitHub Actions so it runs across multiple environments, and targeted test coverage of at least 70 percent.
Documentation
We revised the README so the package introduces itself properly, and built full documentation covering installation, self-contained tutorials, and an API reference.
Containerization and Cloud Deployment
We created a Dockerfile and container for running Voluseg, wrote infrastructure code to launch that container in the cloud, and demonstrated the pipeline running there end to end.
Reading HDF5 Files
Voluseg was extended to read 4D HDF5 files, including NWB data from Dandiset 000350, which holds the two-photon data behind the 2019 paper.