numorph-3dunet 0.1.3

Numorph segmentation of cell nuclei using a 3dunet.

NuMorph 3DUnet

The package performs cell nuclei segmentation on large light-sheet imaging dataset. The models that the package uses can be found here for download.

This package containes the original 3DUnet used in the NuMorph pipeline. A detailed describtion of the architecture and training procedure can be found in the publication.

The PyPi package is intended to be installed in a Nvidia optimized container with Tensorflow and used as a nextflow module in the pipeline nf-core/lsmquant. The container is hosted by the nf-core community repository on quay.io .

Installation

The package can also be used within a conda environment (not recommended).

Clone the repository to your workstation. The numorphunet.yml defines the necessary dependencies for running the prediction. You need to have conda installed to create the environment with the following command:

conda env create -f numorphunet.yml

Activate the environment with:

conda activate 3dunet

Install the numorph 3DUnet in the 3dunetconda env by using the following command in the directory of the pyproject.toml file :

pip install .

Usage

Once installed, you can run the cell segmentation tool using the command:

numorph_3dunet.predict -i /path/to/input/directory -o /path/to/output/directory --n_channels 1 --sample_name TEST1 --model /path/to/model_file.h5

Required arguments:

• -i: Input image directory
• -o: Output directory (will be created if it doesn't exist)
• --n_channels: Number of channels
• --sample_name: Sample name for output files
• --model: Model file (.h5)

Optional arguments:

• -g: GPU tag (default: 0)
• --pred_threshold: Prediction threshold (default: 0.5)
• --int_threshold: Minimum intensity threshold (default: 200)
• --overlap: Overlap between chunks [x y z] (default: 16 16 8)

See full help with numorph_3dunet.predict --help

Credits

The pip package was originally developed by Carolin Schwitalla and contains the original work of Oleh Krupa who is the main developer of the 3DUnet and corresponding models used by the NuMorph toolbox.

NuMorph: Tools for cortical cellular phenotyping in tissue-cleared whole-brain images

Krupa O, Fragola G, Hadden-Ford E, Mory JT, Liu T, Humphrey Z, Rees BW, Krishnamurthy A, Snider WD, Zylka MJ, Wu G, Xing L, Stein JL.

Cell Rep. 2021 Oct 12, doi: 10.1016/j.celrep.2021.109802