Scholarlink

In this study, we evaluate and compare the performance of two deep learning architectures—TinyViT and ResNet50—for the classification of COVID-19 from the chest CT scans of the publicly available HUST COVID-19 CT dataset. The dataset comprises 5705 non-informative CT (NiCT) images with no visible lung parenchyma features, 4001 positive CT (pCT) images with evident findings of COVID-19 pneumonia, and 9979 negative CT (nCT) images with no findings of COVID-19. Both the models were learned to make a prediction of CT images into any one of the three classes: NiCT, pCT, or nCT. Experiment results indicate that TinyViT outperforms ResNet50 with accuracy and F1-score of 98%, whereas ResNet50 achieves merely 96% on both the scores. The experiment results indicate the potential of light-weight vision transformers like TinyViT in optimizing diagnostic accuracy while maintaining computational efficiency, making them prime contenders for being deployed to resource-constrained clinical settings.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Learning for COVID-19 Diagnosis: A Comparative Study of Vision Transformers and CNNs

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Elmehdi Benmalek,
  • Najat Lechhab,
  • Wajih Rhalem,
  • Najib El Idrissi,
  • Atman Jbabi,
  • Abdelilah Jilbab,
  • Jamal Elmhamdi

摘要

In this study, we evaluate and compare the performance of two deep learning architectures—TinyViT and ResNet50—for the classification of COVID-19 from the chest CT scans of the publicly available HUST COVID-19 CT dataset. The dataset comprises 5705 non-informative CT (NiCT) images with no visible lung parenchyma features, 4001 positive CT (pCT) images with evident findings of COVID-19 pneumonia, and 9979 negative CT (nCT) images with no findings of COVID-19. Both the models were learned to make a prediction of CT images into any one of the three classes: NiCT, pCT, or nCT. Experiment results indicate that TinyViT outperforms ResNet50 with accuracy and F1-score of 98%, whereas ResNet50 achieves merely 96% on both the scores. The experiment results indicate the potential of light-weight vision transformers like TinyViT in optimizing diagnostic accuracy while maintaining computational efficiency, making them prime contenders for being deployed to resource-constrained clinical settings.