Deep Learning For Classification Of Chest X-Ray Images (Covid 19)

TL;DR

This paper develops a deep learning approach using CNNs and pre-trained models to classify chest X-ray images for Covid-19 diagnosis, achieving high accuracy and faster convergence, aiding medical practitioners during pandemics.

Contribution

It introduces an effective CNN-based method with pre-trained models for classifying Covid-19 and other lung conditions from chest X-rays, emphasizing improved speed and accuracy.

Findings

ResNet 18 achieved 94.1% accuracy

Pre-trained models converge faster than CNNs

GPU implementation reduces diagnosis time

Abstract

In medical practice, the contribution of information technology can be considerable. Most of these practices include the images that medical assistance uses to identify different pathologies of the human body. One of them is X-ray images which cover much of our work in this paper. Chest x-rays have played an important role in Covid 19 identification and diagnosis. The Covid 19 virus has been declared a global pandemic since 2020 after the first case found in Wuhan China in December 2019. Our goal in this project is to be able to classify different chest X-ray images containing Covid 19, viral pneumonia, lung opacity and normal images. We used CNN architecture and different pre-trained models. The best result is obtained by the use of the ResNet 18 architecture with 94.1% accuracy. We also note that The GPU execution time is optimal in the case of AlexNet but what requires our attention is that the pretrained models converge much faster than the CNN. The time saving is very considerable. With these results not only will solve the diagnosis time for patients, but will provide an interesting tool for practitioners, thus helping them in times of strong pandemic in particular.