COVID-19 Detection from Chest X-Rays

Jordan University of Science and Technology

Course: Deep Learning

Team Members:

• Abdel Rahman Alsheyab
• Mohammad Alkhasawneh
• Osamh Al Shra'h
• Nidal Shahin

Description

For the Semester-Final Team Project, my friends and I have built a Deep Learning model to classify chest X-ray images into Normal, COVID-19, and Pneumonia (Viral/Bacterial) cases. Using almost 34K CXR images of size 256x256 downloaded from COVID-QU-Ex Dataset on kaggle (https://www.kaggle.com/datasets/anasmohammedtahir/covidqu). Citations are mentioned in the link and the uploaded notebook.

Why X-rays? They are fast, non-invasive, and widely available, making them an effective screening tool.

Impact:

• Helps hospitals prioritize treatment and isolate cases early.
• Reduces dependency on PCR tests, which can be slow and resource-intensive.
• Supports underdeveloped regions where advanced diagnostic tools are limited.

Notes:

• We downloaded specific partitions of the original kaggle dataset and restructered it in our own way.
  • Divided into Train (70%), Dev (15%), Test (15%).
  • Three labels.csv files were created, one for each.
  • (CNP_DS) Dataset link: (https://drive.google.com/file/d/1UUJfTyv6R1qYcePiGS6xxp66EYAfq1_H/view?usp=drive_link).
• We used multiple libraries in this project mainly pytorch for the CNN model and matplot for visuals.
• You might have to change the paths to match your own files directories.
• You will have to manually create the (.pth) path files we used in our code to save models with the best results.
• The code works on devices with or without GPUs.
• Two files where uploaded:
  • Notebook (.ipynb) file which includes the code and the outputs.
  • Report (PDF) file that covers the topic, workflow, results and challenges.