Retina-chatbands

A CNN-based detection of retina ChAT-bands.

Further details can be found in: Muellner & Roska 2023, https://doi.org/10.1101/2023.03.22.533751

Please cite our manuscript if you use this code.

Installation

Frist, clone or download this repository.

Create a python environment (Optional, but recommended):

conda create -n retina-env python=3.6
conda activate retina-env

If you are on a machine with a CUDA-compatible GPU, you should first install the gpu version of tensorflow:

conda install tensorflow-gpu=1.12

Install the package into the current environment

cd faim-retina-chatbands/
pip install .

Usage

The following usage examples assume you are using a terminal (e.g. anaconda prompt) from the base folder of this repo on your storage and that you have activated the corresponding virtual environment (e.g. conda).

conda activate retina-env  # only necessary you have setup a conda environment called retina-env.
cd faim-retina-chatbands/

Run a given model

bash segment_data.sh

Adjust the parameters in segment_data.sh according to your needs.

The most important parameters are:

--output-folder: directory into which the segmentations are written.

--model-dir: directory containing the trained model

--input-folder: input folder which is scanned to identify stacks to process.

--fname-patterns: list of file patterns matching stacks that need to be processed. E.g. *conf488*.stk, *.czi.

--split-fraction: split the work into split-fraction number of chunks. This can make it easier to distribute work over different machines and to restart processing in case of failures.

If you would like to see all parameters of the task, use:

luigi --module retchat.tasks.run_segmentation retina.ParallelChatbandPredictionTask --help

Train a new model

Training a model is done in two steps. First, the training data is being prepared with prepare_dataset.py. Then, the model is trained on the dataset using train_segmentation.py.

Each of them can be called using python <script.py> .... For details on the arguments, use

python prepare_dataset.py --help
python train_segmentation.py --help

You find example calls of those two scripts in scripts/. To run these examples, simply call:

bash scripts/run_preparation.sh
bash scripts/run_single_training.sh

Note that these bash scripts contain path name placeholders - make sure to adjust the paths accordingly.