COVID-19-X-Ray-Classification

Utilizing Deep Learning to detect COVID-19 and Viral Pneumonia from x-ray images

Research Publication: https://dl.acm.org/doi/10.1145/3431804

Datasets used:

• COVID-19 Radiography Database
• COVID-19 10000 Images

Related Research Papers:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7187882/
• https://arxiv.org/pdf/2004.05758.pdf
• https://arxiv.org/pdf/2003.09871.pdf

Project Structure

Training Set contains:

• 200 COVID-19 X-Rays
• 250 Viral Pneumonia X-Rays
• 250 Normal X-Rays
• Total: 700 images

Testing Set contains:

• 89 COVID-19 X-Rays
• 300 Viral Pneumonia X-Rays
• 300 Normal X-Rays
• Total: 689 images

Model Architecture

Results

• Achieved 93% Accuracy on the Testing Set, with F-1 Score of 93%, after 25 Epochs

The model performance was also evaluated after performing 5-fold cross validation on the entire dataset of 1389 images, in which it produced an average accuracy of 90.64% and average F-1 Score of 89.8%

Findings

It is inherently difficult to differentiate between the occurence of the two diseases from a normal x-ray. In fact, ~20 million radiology reports contain clinically significant errors, where 10% play a role in patient deaths. Deep learning offers a solution to this problem.

Saliency maps can help us better understand the features in the x-rays and visualize what areas of the image are of high importance. The areas of yellow gradient have the greatest influence on the model's prediction.