In this repo you will find the source code, along with EDA notebooks and some (but not all) models used to create the Molo application for detecting ocular myopathies from OCT scans.

The full app is split into 3 services:

• The frontend
• The backend (called molo).
• Tensorflow Serving (Available only when running with Docker-compose) and is predominately used in development.

Currently, the production version of this application is hosted on GCP Cloud Run and can be found at here There are two models that can be used in production, either the full model and the TF Lite model, which is the quantized version of the full model and occupies 10x less space in memory.

A quick demo of what the app looks like at the moment. To run that locally for you, just follow the instructions further down.

In order to be able to properly run the full application locally, a couple of files need to be created.

1. A .env file with environment variables that are used throughout the application. Below you will find the full list of variables that need to be set by the user. This file should be placed inside the molo directory. You will also need to create the secrets directory inside of ./molo

    MODEL_URI="samsung-oct-model:8500"  # this can be kept as is
    GCS_PROJECT_BUCKET=<name-of-your-gcp-bucket>
    GOOGLE_APPLICATION_CREDENTIALS="secrets/<PATH_TO_GCP_KEY>  
    UPLOADED_IMAGES_GCS_PATH=<RELATIVE_PATH_TO_IMAGE_DIR_IN_GCP_BUCKET>
    PDF_REPORTS_GCS_PATH=<RELATIVE_PATH_TO_PDF_REPORT_DIR_IN_GCP_BUCKET>
    IMAGE_SIZE=80  # this should be left at 80 if using models shipped with the repo
   

2. Once you have created this file be sure to upload a GCP service account key for your project into ./molo/secrets. Make sure this service account has permissions to write and read files from your GCS repo. You can follow the instructions here if you are unsure, how to create a new service account or generate json keys for it. These credentials are needed to upload and download image files. In future versions of the app we will try to simplify this process to make running our app more straightforward, but for the time being please bare with us.

3. You will also need to create the file ./frontend/.env.development, which will be used by VUE to set environment varialbes for the frontend app. There is only one variable in this one

   VUE_APP_SERVER_URL="http://127.0.0.1:8000"
   

You should be done and ready to run our app locally with docker-compose!

The good news is that this repo comes with a docker-compose file, which you can use to run the entire app on any machine with Docker installed. (To install Docker click here

Once you have Docker installed, clone this repo

mkdir samsung_oct && git clone https://github.com/djvaroli/samsung_oct.git samsung_oct/

Navigate to the root directory of the repo if you haven't done so already.

cd samsung_oct/

Boot up the two services with docker-compose

docker-compose up --build

You should now be able to access the molo service at localhost:8000 and the frontend service at localhost:80. To make sure things are working, navigate to localhost:8000 and you should see a welcome message.

Clone this repository to your local machine

mkdir samsung_oct && git clone https://github.com/djvaroli/samsung_oct.git samsung_oct/

Create a python virtual environment and activate it

python3 -m venv venv && source venv/bin/activate

Install all the necessary dependencies

pip3 install -r requirements.txt
pip3 install -e utilities/ # this will install the  custom utility functions

You should now be set to run/modify code in this repository.

• Refactor notebooks for model training into executable python scripts
• Add results to repo README
• Re-write existing SimCLR model code
  • Make it more robust and compatible with recent TF versions
  • Better model structure during build to enable proper model summary visualisation
  • Refactor code to make it simpler to use with different SimCLR encoder bases
  • Implement additional logic needed for our experiments

• Repo Cleanup
• Docker-compose set up with TF Serving