Processing microscopy image sequences using Matlab, a graphical user-interface and deep learning classifiers

About The Project

DetecDiv provides a comprehensive set of tools to analyze time microscopy images using deep learning methods. The software structure is such that data can be processed either at the command line or using a graphical user-interface. Detecdiv classification models include : image classification and regression, semantic segmentation, LSTM networks to analyze data and image timeseries. Please refer to our pre-print for further details about the software and its applications for yeast cell division counting and replicative lifespan analysis:

DetecDiv, a deep-learning platform for automated cell division tracking and replicative lifespan analysis

Théo Aspert, Didier Hentsch, Gilles Charvin

https://doi.org/10.7554/eLife.79519

(back to top)

Table of contents

1. About the project
2. Installation procedure
3. User guide
4. Available data and models
5. Demo project
6. Acknowledgments

Installation procedure

We recommend using Matlab >= R2021a to ensure the compatibility of the software. DetecDiv requires the following toolboxes:

-MATLAB Version 9.10 (R2021a)

-Computer Vision Toolbox Version 10.0 (R2021a)

-Deep Learning Toolbox Version 14.2 (R2021a)

-Image Processing Toolbox Version 11.3 (R2021a)

-Statistics and Machine Learning Toolbox Version 12.1 (R2021a)

-Parallel Computing Toolbox Version 7.4 (R2021a) (optional)

In addition, you will need to install specific Matlab addons by clicking the "add-ons" button in the matlab main window. Then search for "Googlenet", "Resnet50", and "Resnet18" to install these packages that correspond to pre-trained neural nets.

Make sure to include all DetecDiv folders and subfolders in the Matlab path using the "Set Path" in the main Matlab workspace.

(back to top)

User guide

A guide on how to use the graphical user-interface can be found here:

GUI User guide

A list of command-line instructions to use DetecDiv in scripts or in the Matlab workspace can be found here:

Command-line instructions

(back to top)

Available datasets and models

Trained models and annotated image trainingset bundle

All the classification models used in the paper can be downloaded from Zenodo:

This repository contains classifiers used to score cell division and lifespan in different contexts (geometries, microscopes). Each .zip files contains the trained classification model, along with the training and validation annonated image datasets that were used to train and test the classfiers. The generalist model is located in a different This repository .

Sample images and labels	Description	Link to the zip file
	Image sequence classification for division and lifespan analysis. Trap geometry as used in the Acar lab	Download
	Image sequence classification for division and lifespan analysis. Trap geometry as used in the Charvin lab with a 20x objective and the RAMM microscope	Download
	Image sequence classification for division and lifespan analysis. Trap geometry as used in the Charvin lab with a 60x objective and a Nikon microscope	Download
	Image sequence classification for division and lifespan analysis. Trap geometry similar to that used in the Qin lab	Download
	Image sequence classification for division and lifespan analysis. Trap geometry as used in the Swain lab	Download
	Image sequence classification for cell contour segmentation.	Download
	Image sequence classification for division and lifespan analysis. Works for any trap geometry listed above	Download

Individual images and classification models

The repositories below contain datasets (training sets and validation sets) used to train classification models, as well as the trained model itself (.mat file). Unlike the "bundle" above, this respositories contain either images or the model, so they are not intended to be used seamlessly within Detecdiv. If you want to use a particular model within Detecdiv, please dowanload a "bundle" as indicated above.

Description	Training set repository	Testset repository	Trained model
Image sequence classification for division and lifespan analysis. Trap geometry as used in the Charvin lab with a 20x objective and the RAMM microscope. Relates to Figure 1 in the paper (id01)	Link	Link	Link
Cell segmentation. Trap geometry as used in the Charvin lab with a 20x objective and the RAMM microscope. Relates to Figure 5 in the paper (id02)	Link	Link	Link
SEP detection within lifespan data. Relates to Figure 4 in the paper (id03)	Link	Link	Link

DetecDiv demo project

A demo project that contains all the necessary files (i.e. raw image files, ROIs, groudtruth data, classifier models, trained classifiers) to learn how to use DetecDiv can be found here:

(back to top)

Acknowledgments

Many thanks to those who provided the necessary resources to make this project possible, including the Charvin lab group members, the IGBMC staff and facilities.

(back to top)