Revolutionizing COVID-19 Diagnosis with Swin Transformer: A Comparative Study on CT Image Attention Analysisand CNN Models performance

Abstract

In this paper, a novel Swin Transformer-based methodology is proposed for the diagnosis of COVID-19 utilizing computed tomography (CT) images, with the objective of enhancing performance and interpretability compared to prevailing deep learning models. Empirical results demonstrate that the Swin Transformer surpasses VGG16 and Res50 in terms of evaluation metrics, attaining exceptional test accuracy, AUC, precision, and recall values. Notwithstanding the comparable performance of Inception V3,the Swin Transformer exhibits a more efficient training cycle. Moreover, attention visualization substantiates the superior focus of the Swin Transformer during the analysis process. This study underscores the promising potential of the Swin Transformer in augmenting CT image-based diagnostics for COVID-19 and furthering the development of medical image analysis techniques.

Ähnliche Arbeiten