Robust X-ray Image Classification using Vision Transformers: A Comparative Study with Convolutional Neural Networks

Accurate classification of X-ray images is crucial for detecting COVID-19 and pneumonia cases. This study compares the performance of Vision Transformers (ViTs) and Convolutional Neural Networks in X-ray image classification tasks, with a focus on robustness and overfitting. ViTs and CNNs, specifically VGG16 and ResNet-50, are trained on a diverse dataset comprising COVID-19 cases, pneumonia cases, and normal cases. The results demonstrate that ViTs achieve high accuracy without overfitting, even after exceeding 100 epochs. In contrast, CNNs exhibit a higher susceptibility to overfitting, limiting their performance and generalization ability. The robustness of ViTs in X-ray image classification tasks highlights their potential for accurate COVID-19 and pneumonia detection. This study emphasizes the need to explore alternative architectures like ViTs to overcome overfitting challenges and enhance diagnostic tools for X-ray image classification in the context of respiratory diseases.