This process gives files vsi come from Olympus cellSens Software. Each file contains a sequence of color images.
In order to analyze video we have used the Stardis method (https://github.com/stardist/stardist/).
> Uwe Schmidt, Martin Weigert, Coleman Broaddus, and Gene Myers. [Cell Detection with Star-convex Polygons](https://arxiv.org/abs/1806.03535). _International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI)_, Granada, Spain, September 2018.
StarDist is a deep-learning-based method of 2D and 3D nucleus detection from Martin Weigert and Uwe Schmidt. StarDist is powered by a deep learning model that is trained to detect specific kinds of nuclei.
For our study, we annotated a database of images and built a model specific to our problem. We searched for chromatophores and classified them in 2 categories: dark and light.
3. Run _AnalyzeSequence.ipnyb_ program on conda environment
#### To build a new model
1. Annotation of the images with Qupath : https://qupath.github.io/
> Bankhead, P. et al. **QuPath: Open source software for digital pathology image analysis**. _Scientific Reports_ (2017). \
> https://doi.org/10.1038/s41598-017-17204-5
INSERT UN EXEMPLE IMAGE QUAPTH AVEC CONTOURS et CLASSES
On your images, annotate the objects that interest you. If you have categories, remember to declare the different classes. For details on how to use Qupath, see the documentation on their site.
2. Once your annotations are done, use the script "_Export Annotations gson.groovy_" to export the annotations as a json file.
3. Then use the python program "_JSON_Qupath_to_ImageLabel.ipynb_" which will allow to format the images and annotations so that they can be used by the program "_Model_Build.ipynb_." This program is directly from an example provided on the github Stardist repository.
> **_Be careful, for the program to work, you need a compatible GPU graphics card._**
4. The program "_AnalyzeSequence.ipnyb_" can then be used by designating your own model in the program code.
