A Review on The Comparison Between Convolutional Neural Networks and Vision Transformers for Disease Classification in Chest X-Ray

This review of the literature contrasts convolutional neural networks (CNNs) and vision transformers (ViTs) in the context of artificial intelligence-based medical image analysis. The focus of the analysis of twenty-four (24) studies is accuracy, scalability, computational efficiency, and robustness. Results indicate that ViTs perform better than CNNs on a variety of tasks and that pre-training is essential to their ability to handle complicated medical datasets.

1. INTRODUCTION

The rapid advancement of medical imaging technologies and artificial intelligence has significantly revolutionized medical diagnostics, especially in radiology [1]. Chest x-ray (CXR) is needed for diagnosing various respiratory and heart conditions owing to its availability as well as its affordability [2] . As respiratory diseases rise, accurate Chest X-ray classification is important. Traditional analysis usually suffers from subjectivity, necessitating automated systems for improved performance [3]. This review paper explores the comparative analysis between Convolutional Neural Networks performance metrics with that of Vision Transformer using Chest X-ray for disease classification with the view of aiding automated diagnosis [4]. A thorough literature search identified peer-reviewed studies from 2022 to 2024, facing limitations in varying designs, preprocessing techniques and transforming methodologies [5]. This review addresses the gap in comparative studies of CNNs and ViTs in chest radiography to enhance automated diagnostic strategies [6]. Advancements in medical imaging modality such as is inspired and motivated with the deployment of machine learning Algorithm.