Skip to content

Latest commit

 

History

History
59 lines (47 loc) · 3.09 KB

README.md

File metadata and controls

59 lines (47 loc) · 3.09 KB
Raw

CSIC 5011 / MATH 5473 Final Project
Imaging’s Potential to Assist in COVID-19 Crisis

Coronavirus disease 2019 (COVID-19) is a contagious disease caused by severeacute respiratory syndrome coronavirus 2 (SARS-CoV-2). An efficient and accu-rate abnormalities detection in chest X-rays (CXR) of infected patients can assisthealth care staff in the battle against COVID-19.

We investigate the potential of synthetic images additional to existing dataset to improveCOVID-19 CXR classification accuracy.

Members

  • YS. CHEN
  • WH. SUM
  • HP. IP
  • ZP. WU

Instruction

  • csic_5011_project2_report_ChenSumIpWu.pdf is the final report.
  • csic_5011_project2_slides_ChenSumIpWu.pdf is the presentation slides.
  • presentation video link: https://youtu.be/i8ON-oZdBFk
  • "Code" contains main source codes in the report.
  • "Data" contains data collection information.
  • "Results" contains some results.

Result Gallery

Real vs Synthetic Chest X-rays

Real COVID-19 CXRs Real Lung Opacity CXRs Real Normal CXRs Real Viral PNA CXRs Synthetic COVID-19 CXRs Synthetic Lung Opacity CXRs Synthetic Normal CXRs Synthetic Viral PNA CXRs

Data Visualization

Visualization of Blended Data

Classification Performance

Real Data - Overall Accuracy: 84.1% Real + Synthetic Data - Overall Accuracy: 91%
Real data classification Blended data classification

Acknowledgements

Original data were collected from COVID-19 Radiography Database.

The code in "Classification.ipynb" borrows heavily from Kaggle code.

The code in "stylegan2-ada" borrows heavily from stylegan2-ada-pytorch.