CNN Deep-Learning Technique to Detect Covid-19 Using Chest X-ray

Abstract

Most of the countries around the world are locked down due to the pandemic.Every country has imposed strict travel restrictions and has stopped all types of visas and tourist activities.This created a major impact on the aviation sector and the tourist sector.Even the people not affected by Covid-19 and in real emergence are not able to travel from one place to another.Some countries have laid down quarantine rules, which will be a major hindrance to emergency travellers and tourists.All passengers travelling are tested for COVID-19 using RT-PCR, which can take between 48 to 72 hours to produce the result.But in some cases, people who are tested negative even after 3 or 4 RT-PCR tests show typical pneumonia in the CT Scan or a chest X-ray.If the aviation sector relies only on the RT-PCR test, many patients may be missed.To reduce the risk to some extent and prevent a high-risk patient from travelling, the passenger can be asked to upload his / her chest X-ray before travel.Using an X-ray of the chest, we can predict the possibility of Covid-19 cases before the patients are physically examined.This technique cannot replace the RT-PCR test but can be a stand-by tool to help detect Covid-19 before the RT-PCR test.It would also help to identify patients who are highly prone to the infection.In this paper, we developed a CNN from scratch to identify a patient infected with COVID from a chest X-ray image.The model was trained with the chest X-ray of normal and COVID patients.Later the model was tested on two datasets, one publicly available in GitHub, and the other dataset was compiled from the Italian Society of Medical and Interventional Radiology website using web scrapping.The model produced an accuracy of 96.48 percent with the training dataset.To further improve accuracy, we used the same dataset on a pre-trained network (VGG16) and achieved an accuracy of around 99 percent.

Ähnliche Arbeiten