Cone Segmentation ML (Machine Learning edition)

Cell segmentation on Adaptive Optics Retinal Images using pre-trained A-GAN machine learning model and manual editing.

Jianfei Liu (NEI/NIH), Andrei Volkov (NEI/NIH Contractor), and Johnny Tam (NEI/NIH), with research support from the Intramural Research Program of the National Institutes of Health.

BibTeX

@ARTICLE{9339889,
	author={Liu, Jianfei and Shen, Christine and Aguilera, Nancy and Cukras, Catherine and Hufnagel, Robert B. and Zein, Wadih M. and Liu, Tao and Tam, Johnny},
	journal={IEEE Transactions on Medical Imaging},
	title={Active Cell Appearance Model Induced Generative Adversarial Networks for Annotation-Efficient Cell Segmentation and Identification on Adaptive Optics Retinal Images},
	year={2021},
	volume={40},
	number={10},
	pages={2820-2831},
	doi={10.1109/TMI.2021.3055483}
}

---

Setting up development environment

1. Download and install Miniconda or Anaconda.

2. Check out ConeSegmentationML to a local directory <prefix>/ConeSegementationML. (Replace <prefix> with any suitable local directory).

3. Run Anaconda Prompt (or Terminal), cd to <prefix>/ConeSegmentationML.

4. Create Conda Virtual Environment (do this once, next time skip to the next step):

   conda env create --file conda-environment-win.yml (Windows)

   conda env create --file conda-environment-mac.yml (MAC OS)

5. Activate the Virtual Environment:

   conda activate ConeSegmentation

6. Start the application:

   python __main__.py

7. Build "frozen Python" application:

   pyinstaller --clean --noconfirm build-dir.spec

If successful, the result is the directory ConeSegmentationML inside <prefix>/ConeSegmentationML/dist/. You can copy this directory with all its contents to a different machine, and run the executable __main__ (in MacOS and Linux) or __main__.exe (in Windows). It does not need Conda VEs or other development tools.

In MacOS systems, you can build a Mac application instead:

pyinstaller --clean --noconfirm build-app-dir.spec

The result is <prefix>/ConeSegmentationML/dist/ConeSegmentationML.app.

Creating Windows installer using NSIS

1. Download and install NSIS if you don't have it already.

2. Follow steps 1 through 7 of Setting up development environment to build the directory containing "frozen Python" application.

3. Open Command Prompt (or Conda Prompt), cd to <prefix>/ConeSegmentationML.

4. Run NSIS:

"C:\Program Files (x86)\NSIS\makensis.exe" /V4 build-win64-installer.nsi

(Replace C:\Program Files (x86)\NSIS with the actual installation directory, if different from default). If successful, the result is <prefix>/ConeSegmentationML/dist/ConeSegmentationML-{version}-win64.exe. This is a regular Windows installer, which can be distributed to other Windows systems. It requires admin access.

Creating MAC OS installer (.dmg)

1. Make sure Xcode is installed (normally, via Apple App Store).

2. Install Node.js, npm and dmg-license (require admin/sudo access), if they are not already installed:

   curl -L https://raw.githubusercontent.com/tj/n/master/bin/n -o n
   sudo bash n lts
   sudo npm install --global minimist
   sudo npm install --global dmg-license
   rm n
   

3. Follow steps 1 through 5 of Setting up development environment to setup the development environment.

4. At the Conda prompt with ConeSegmentation activated, cd to <prefix>/ConeSegmentationML and type the command:

   bash make_dmg.sh

If prompted to allow Terminal to run Finder scripts, answer "Allow". The result is <prefix>/ConeSegmentationML/dist/ConeSegmentationML-{version}-Darwin.dmg. It is a Mac OS disk image file; when opened, it asks for accepting the license agreement, then mounts itself as an external drive and opens a Finder window, that looks like this:

You can run the app by double-clicking on the icon, or copy it to your Applications folder by dragging the icon over "Applications". Once ConeSegmentationML is in your Applications folder, you can eject the ConeSegmentationML disk, and delete ConeSegmentationML-Darwin.dmg.

---

Deleting Conda Virtual Environment

To delete the Virtual environment at the Conda prompt, deactivate it first if it is active:

conda deactivate

then type:

conda remove --name ConeSegmentation