AI-powered diagnosis of respiratory diseases: Evaluating vision transformers and ResNet architectures for covid-19 and lung pathologies

This study systematically evaluates the efficacy of advanced deep learning architectures, namely Vision Transformers (ViT) and various ResNet models (ResNet50, ResNet101, ResNet152), in the classification of chest radiographs into four clinically significant diagnostic categories: Normal, Lung Opacity, Viral Pneumonia, and COVID-19. A meticulously curated dataset comprising 21,165 chest X-ray images was utilized to benchmark the models' performance across key evaluation metrics, including precision, recall, F1-score and accuracy. The experimental evaluation reveals that ViT model achieved 90.25% accuracy, 91.56% precision, 89.22% recall, and a 90.25% F1-score. These findings highlight the potential of AI-driven approaches in augmenting medical diagnostics, improving diagnostic accuracy, and enhancing healthcare delivery, particularly in resource-limited settings. The study underscores the applicability of Vision Transformers in complex medical imaging tasks and contributes to the growing body of research supporting AI-based solutions for respiratory diseases and other healthcare challenges.