[](https://github.com/Gitlab /morphodeep/raw/main/LICENSE) [](https://github.com/Gitlab /morphodeep/actions) [](https://codecov.io/gh/Gitlab /morphodeep)

MorphoDeep

MorphoDeep is a Python package for performing membrane instance segmentation using a pre-trained U-Net model based on the Semantic format.
It produces a dense labels image where each unique integer corresponds to a segmented cell instance. It also includes a napari plugin

The plugin is designed for:

• fluorescence images of cell membranes
• 2D and 3D volumes (with optional isotropic rescaling)
• interactive use in napari or scripted use in Python

Key features

• Pre-trained U-Net for 2D / pseudo-3D membrane instance segmentation
• Napari widget with:
  • image layer selection
  • network patch size (128 or 256)
  • 2D / 3D mode
  • optional tiling (“patches mode”) for large images
  • voxel size and Z-axis selection
  • optional isotropic rescaling for anisotropic volumes
  • Automatic model download and caching on first use

Installation

Setup your environnement

conda create -y -n morphodeep-env python=3.10
conda activate morphodeep-env

You can install morphodeep via ffpip:

pip install morphodeep

Installation with napari

Setup your environnement

conda create -y -n napari-env -c conda-forge python=3.11
conda activate napari-env

You can install morphodeep plugin with napari and Qt via:

pip install "napari[all]"
pip install "morphodeep[napari]"

MorphoDeep library Usage

You have 3 way to use the morphodeep library:

1. with a command line : python morphodeep.py <args>
2. directly inside the python code using the semantic library: MorphoModel(args)
3. directly in napari with the morphodeep plugin

Arguments

JOBS (run the command line in a job )

• --job : Launch the corresponding job
• --exe : Execute the method ( ground_truth ... )
• --path_jobs : path job output
• --job_number : Launch a sepecif job in job array list
• --job_filename : The file of the job array list

ACTION

• --train : Launch the training
• --predict : Launch the prediction
• --ground_truth : Generate all the ground truth
• --setdata : Generate the data list files used for train/test/predict
• --export : Compute the testing set and plot the results in figures
• --evolution : Evaluation the testing set throught the different epochs
• --export_evol : Predict for each epochs
• --eval : Evaluation of training database
• --plot : Compute export and plot loss
• --plot_loss : Plot the loss and accuracy

METHOD

• -mode : 2D or 3D
• --img_size : image size
• --full : Use a several layers to complete a tiles network to a work with image size
• --weight_files : weight files

TRAINING

• --batch_size : "batch_size
• --steps_per_epoch : steps_per_epoch
• --epochs : number of epochs
• --log_path : log path for tensorboard
• --specie : which specie ? (PM,DR,AT,DM,or PM-AT, PM-DM-AT, CP (for CellPose)
• --microscope : which microscope ? (SPIM (for lightsheet), CONF (for confocal),SIM (for simulated))
• --augmentation : Use Data Augmentation ? Default is False
• --pondere : Keep only the worst element (10%) for training

DATA Management

• --input_path : Input Path (depend on the called method )
• --output_path : Output Path (depend on the called method
• --membrane_filename : Filename membrane")
• --nucleus_filename : Filename nucleus")
• --segmented_filename : Filename segmented")
• --embryo_name : Embryo Name")
• --ratio : Resize the full database")
• --input_method : Input Method (for Ground Truth Calculcation)")

DATASET

• --dataset_file : Dataset File without .train, .valid and .test
• --test_split : Float between 0 and 1. Fraction of the data to be used as test data.
• --validation_split : Float between 0 and 1. Fraction of the data to be used as validate data.

PREDICTION

• --input_file : Input Filename to predict
• --output_file : Output Filename of the predicted image
• --patches : uses for prediction

RESULT

• --export_path : The export path for the results

DATA PREPARATION

1. Extract segmentation and Raw Images data

SPIM PHALLUSIA MAMILATA

example :python morphodeep.py --microscope SPIM --specie PM --phallusia_data

CONF ARABIDOPSSIS

example :python morphodeep.py --microscope CONF --specie AT --arabidopsis_data

CONF LATERAL ROOTS

example :python morphodeep.py --microscope CONF --specie LP --plantseg_data

CONF OVULES

example :python morphodeep.py --microscope CONF --specie OV --plantseg_data

CONF SEA STAR

example :python morphodeep.py --microscope CONF --specie SS --seastar_data

CONF C ELEGANS

example :python morphodeep.py --microscope CONF --specie CE --celegans_data

2. Generate all Ground Truth

args : --ground_truth example :python morphodeep.py --ground_truth --microscope CONF --specie SS

3. Split Train,Test Valid data

args : --setdata example :python morphodeep.py --setdata --microscope CONF --specie SS

3.b To transfert your data in 128 patches

example : python morphodeep.py --setdata --img_size 128

4. Predict Cellpose (on test data)

args : --predict_cellpose example :python morphodeep.py --predict_cellpose --exe --microscope SPIM --specie PM

5. Predict PlantSeg (on test data)

args : --predict_plantseg example :python morphodeep.py --predict_plantseg --exe --microscope CONF --specie OV

6. DataManagement

To compress or uncompress data from $STORE path in Jean Zay : morphodeep/DataManagement/store_data.py example :python store_data.py --compress --microscope CONF --specie PM

7. Train Semantic Networks

args : --train example : python morphodeep.py --train --microscope SPIM --specie PM

8. Compute accuracy and plot loss evolution

args : --plot example : python morphodeep.py --plot --microscope CONF --specie AT

9. Benchmark Test data (from figures path)

example : python Benchmark.py --predict

11. Benchmark on external data

args : python Benchmark.py --predict -f figure_5/EXTERNAL_DATA/external_data_3D.txt

2D

Everything is by default in 3D, add --mode 2D to comptue evertyhing in 2D

1. Convert in 2D DATA (and generate txt files)

We extract all the 3D datbase to 2D slices args : --extract_2D example :python morphodeep.py --mode 3D --microscope ALL --specie all --extract_2D [-wh <test;train;valid>]

2. Generate all Ground Truth for all species (includ in all)

args : --ground_truth example :python morphodeep.py --mode 2D --ground_truth --microscope SPIM --specie PM

3. Predict Cellpose (on test data)

args : --predict_cellpose example :python morphodeep.py --mode 2D --predict_cellpose

4. Predict PlantSseg (on test data)

args : --predict_plantseg example :python morphodeep.py --mode 2D --predict_plantseg

6. DataManagement

To compress or uncompress data from $STORE path in Jean Zay : semantic/DataManagement/store_data.py example :python store_data.py --mode 2D --compress --microscope CONF --specie PM

7. Train Semantic Networks

args : --train example : python morphodeep.py --mode 2D --batch_size 32 --train

8. Compute accuracy and plot loss evolution

args : --plot example : python morphodeep.py --mode 2D --plot

12. Benchmark Test data (from figures path)

example : python Benchmark.py --predict -t 2D -p True

12. Benchmark on external data

example : python Benchmark.py --predict -t 2D -p True -n DUNNET -f /lustre/fsn1/projects/rech/dhp/uhb36wd/DATA_INTEGRATION//EXTERNAL_DATA_2D/external_data_2D.txt

This napari plugin was generated with copier using the napari-plugin-template (None).

Contributing

Contributions are very welcome. Tests can be run with tox, please ensure the coverage at least stays the same before you submit a pull request.

License

Distributed under the terms of the GNU LGPL v3.0 license, "morphodeep" is free and open source software

Issues

If you encounter any problems, please [file an issue] along with a detailed description.