AutoCellLabeler

This package contains code for producing training data for the AutoCellLabeler network, and post-processing code for interpreting its outputs. The actual AutoCellLabeler network architecture is implemented in the pytorch-3dunet package.

Setup

To install, download the source code and install with pip:

git clone git@github.com:flavell-lab/AutoCellLabeler
cd AutoCellLabeler
pip install .

The name of the installed package is autolabel. You will also want to install the pytorch-3dunet package using its setup instructions.

Create training data

To create training, validation, and testing data for AutoCellLabeler from a set of multi-spectral images and human labels in CSV format, see the make_AutoCellLabeler_input.ipynb notebook in this package.

To use the training data in our preprint, please refer to our Dropbox, under the AutoCellLabeler/train_val_test_data directory. (You will have to run the make_AutoCellLabeler_input.ipynb notebook to format this data.)

Train and run AutoCellLabeler

For training and running AutoCellLabeler, please refer to the pytorch-3dunet package.

The model weights used in our preprint are available in our Dropbox, under the AutoCellLabeler/model_weights directory.

Post-process AutoCellLabeler output and evaluate performance

To post-process AutoCellLabeler output, including exporting it to CSV files, and comparing it to human labels for accuracy evaluation, please see the parse_AutoCellLabeler_output.ipynb notebook in this package.

The additional testing datasets used in our preprint are available in our Dropbox, under the AutoCellLabeler/noisy_immobilized_eval directory for the low-SNR TagRFP-only immobilized images, and the AutoCellLabeler/freely_moving_eval for the low-SNR TagRFP-only freely-moving images. (You will have to run the make_AutoCellLabeler_input.ipynb notebook to format this data.)

Citation

To cite this work, please refer to this article:

Deep Neural Networks to Register and Annotate the Cells of the C. elegans Nervous System

Adam A. Atanas, Alicia Kun-Yang Lu, Jungsoo Kim, Saba Baskoylu, Di Kang, Talya S. Kramer, Eric Bueno, Flossie K. Wan, Steven W. Flavell

bioRxiv 2024; doi: https://doi.org/10.1101/2024.07.18.601886