Deep learning Models for Multi-class Pneumonia Detection with emphasis on Covid-19

Abstract

Pneumonia is a dangerous condition that can lead to death. Bacteria, fungi, or viruses are common causes of this condition, which can affect one or both lungs.
The x-ray scans can be used to identify this lung ailment. Pneumonia is a dangerous condition that can lead to death. Bacteria, fungi, or viruses are common causes of this condition, which can affect one or both lungs. The x-ray scans can be used to identify this lung ailment. This illness usually presents itself with a cloudy white structure on the chest x-rays. The Kaggle datasets used for chest X-rays contains a variety of images that are divided into 4 categories – “Normal”, “Viral Pneumonia”, “Bacterial Pneumoia”, “Covid-19”. A deep learning model will be developed that can truly tell us whether or not a person has pneumonia disease. The model will be able to accurately assess the illness severity by detecting the degree of this cloudiness with an automated pneumonia detection system. This project is based purely on ‘Deep Learning’ concepts and will be scalable in types of complications covered, model accuracy and utility convenience analysis/application. Statistical results obtained demonstrate that CNN models can be trained to analyze chest X-ray images, specifically to detect Pneumonia and even to differentiate between Covid-19 and Pneumonia. The attempt to differentiate between viral and bacterial pneumonia has been partially successful, and suggests that it is difficult to distinguish between them due to similar patterns on the X-rays. With the help of MobileNet Architecture is possible to scale this model, the results also suggest that it is also possible to differentiate between pneumonia and COVID-19, which present similarly on the x-rays, which most models fail to undertake. It is thus deduced that the lung damage is most similar in viral and bacterial pneumonia, but the pattern is different for lungs affected with Covid-19.