Generation of Synthetic Chest X-Ray Images and Detection of COVID-19: a Deep Learning based Approach

This is the official implementation of the paper Generation of Synthetic Chest X-Ray Images and Detection of COVID-19: a Deep Learning based Approach.

Usage

Required Directory Structure:

.
+--Train
|  +--.
|  +--COVID
|  +--NORMAL
+--Test
|  +--.
|  +--COVID
|  +--NORMAL

loader.py contains the loading requirements for the dataset.

main.py contains the discriminator, generator, and the acgan function.

trainer.py contains the training methodology for the ACGAN, trained for 1200 epochs.

utils.py has the label smoothing function, the print logs function, the function to generate noise and labels, and the function to plot the loss graph.

generate.py loads the trained generator weights and generates the CXR image.

We have added 50 synthetic images in COVID-19 (Synthetic). Remaining synthetic images are available on request, mail yashkarbhari17@gmail.com .

Training dataset was used from the following: 1) https://github.com/ieee8023/covid-chestxray-dataset 2) https://github.com/agchung/Figure1-COVID-chestxray-dataset 3) https://github.com/agchung/Actualmed-COVID-chestxray-dataset 4) https://www.kaggle.com/tawsifurrahman/covid19-radiography-database

Model structure

Generated Images

COVID-19 CXR	Normal CXR

Citation

@Article{diagnostics11050895,
author = {Karbhari, Yash and Basu, Arpan and Geem, Zong Woo and Han, Gi-Tae and Sarkar, Ram},
title = {Generation of Synthetic Chest X-ray Images and Detection of COVID-19: A Deep Learning Based Approach},
journal = {Diagnostics},
volume = {11},
year = {2021},
ARTICLE-NUMBER = {895}
}