https://github.com/lcwong0928/covid-diagnosis

A machine learning tool based on chest radiographs can aid radiologists or healthcare professionals in diagnosing COVID-19.
https://github.com/lcwong0928/covid-diagnosis

chest-xray-images covid-19 machine-learning

Last synced: 9 days ago
JSON representation

A machine learning tool based on chest radiographs can aid radiologists or healthcare professionals in diagnosing COVID-19.

• Host: GitHub
• URL: https://github.com/lcwong0928/covid-diagnosis
• Owner: lcwong0928
• License: mit
• Created: 2021-01-28T04:51:26.000Z (about 4 years ago)
• Default Branch: main
• Last Pushed: 2022-04-23T17:12:15.000Z (almost 3 years ago)
• Last Synced: 2024-12-05T22:42:48.992Z (2 months ago)
• Topics: chest-xray-images, covid-19, machine-learning
• Language: Jupyter Notebook
• Homepage:
• Size: 21.8 MB
• Stars: 1
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# COVID-19 Diagnoses using Chest Imaging

## Introduction

Coronavirus disease 2019 (COVID-19) first emerged in the city of Wuhan in Hubei province and has become a global

pandemic affecting millions around the globe. Due to the limited availability in traditional antibody testing, rapid and

accurate diagnosis of res- piratory diseases became more urgent in the midst of the pandemic. In this paper, we propose

a machine learning tool based on chest radiographs (X-rays) that can aid radiologists or healthcare professionals in the

diagnosis of COVID-19. X-rays is an imaging technique used to aid the diagnosis of many respiratory diseases, including

tuberculosis (TB) and pneumonia. This image classification task is best accomplished by leveraging effective

convolutional neural network (CNN) architec- tures. We aim to validate several methods, including lo- gistic regression,

K-nearest neighbors, and various CNN architectures, in the classification of posterior-anterior X-ray images of patients

with different respiratory dis- eases. The experimental results show promising results in differentiating chest X-rays

of COVID-19 from normal cases. Specifically, the ResNet architecture achieved a weighted accuracy of 99.0% (with a

sensitivity of 97.0%, a specificity of 100.0%, and a precision of 100.0%) on the binary dataset. However, the

performance dropped significantly as other respiratory images are mixed in to create the multiclass dataset. Saliency

maps, filters, and activation visualization are used to interpret these techniques. While X-rays should not be used as

first-line tests to diagnose COVID-19, we believe our findings can aid in decisions made by medical professionals.

## Links

[Report](https://github.com/lcwong0928/covid-diagnosis/blob/main/results/report.pdf) \

[Presentation](https://github.com/lcwong0928/covid-diagnosis/blob/main/results/presentation.pdf)