Abstract

Worldwide, infectious illnesses have always posed a serious concern. Millions of people die from illnesses that affect the lungs, such as tuberculosis, pneumonia, and COVID-19. Early screening and diagnosis are necessary to administer appropriate care and stop the disease's spread. A simple, low-cost imaging option is chest X-ray (CXR) imaging. Therefore, the main diagnostic method for all of these chest infections and anomalies is a CXR examination. Paramedics and researchers are working nonstop to create a reliable and accurate method for diagnosing diseases in an effort to save lives. This study aims to classify cases of pneumonia, COVID-19, tuberculosis, and normal using a CXR. The model's goal is to classify the diseases using convolution neural networks (CNN). One kind of deep learning technique that works especially well for computer vision and image recognition applications is the CNN. We will compare and assess the two CNN models in this project: EfficientNetB0 and MobileNetV2 and create an ensemble of these models for an enhanced performance. The dataset that was used in this instance was obtained from Kaggle. The collection has 14,000 images overall, with 3500 images per class and 4 classes (Normal, Tuberculosis, Pneumonia, and COVID-19). This project evaluates how accurate its classification is.