A decentralised deep learning application to detect Covid-19, Pneumonia or Healthy lungs based on Lung Ultrasound Images. This repo contains the code to run the application. This is a research studies conducted during the final year of my bachelors.
The reason for this work is so that medical organisations can collaboratively share a deep learning model without sharing the datasets. This improves data privacy and also allows the model to be continuously trained to be more accurate with contributions of more datasets.
The federated learning application is realised through a client-server architecture built using REST (FastAPI), and a user-interface built using NextJS to allow clients to upload dataset, train, perform federated learning algorithms and lastly perform inferencing to achieve a score.
Lung Ultrasound images are acquired from publicly available resources.
NOTE: Some of the dataset network might not be publicly available.
conda create -n covidlus python=3.8 -y
conda activate covidlus
pip install -r requirements.txt
cd fl-module
python app.py
Uses material design from ant Design
cd front-end/fl-app
npm install antd --save
npm run build
npm start
To understand the backbone architecture of the deep learning network. I first built a Convolutional Neural Network (CNN) that compares between pretrained models from VGG16, ResNet50 and a custom CNN architecture. You can explore the algorithms in by only changing the DATA_DIR path in ML_Approach(CNN).ipynb.
Alternatively, you can run run both the Jupyter notebook or script, covidLUS.ipynb or covidlus.py. The notebook and script was both built to run in Google Colab and Mac-M1.
Feel free to contact for collaboration or raise a issue, please reach out: [email protected]