Assessing the Efficacy of Transfer Learning in Chest X-ray Image Classification for Respiratory Disease Diagnosis: Focus on COVID-19, Lung Opacity, and Viral Pneumonia

Abstract

Compared to alternatives like Polymerase Chain Reaction (PCR), the Chest X-Ray (CXR)-based method, which falls under Computer-Aided Diagnostic (CAD) approaches, provides a cost-effective solution for early-stage diagnosis of respiratory diseases, including Covid-19, Lung Opacity, and Viral Pneumonia. However, the utilization of CXR-based techniques for respiratory disease diagnosis has been relatively limited, with only a few studies exploring this approach. This research paper delves into the utilization of three distinct architectures—VGG16, VGG19, and MobileNet to classify three respiratory ailments mainly COVID-19, lung opacity, Viral Pneumonia and Normal healthy Lungs. The study incorporates a combination of transfer learning and a custom model, training models from the ground up. The methodology employed in this study centres around Computer-Aided Diagnosis (CAD) using Chest X-Rays (CXRs). Notably, this approach stands out as a cost-effective alternative in comparison to other diagnostic techniques such as Polymerase Chain Reaction (PCR), CT scans, and various medical procedures. Despite its cost-effectiveness, the adoption of CXR-based techniques for diagnosing respiratory diseases remains somewhat constrained. The research discussion section addresses this limitation by presenting and analysing the obtained results. We obtained an accuracy of 98% using the famous mobile Net architecture using Transfer learning, with VGG16 and VGG19 we obtained an accuracy of 97%.