The accurate classification of lung diseases through medical imaging like X-Ray and CT-scan is vital for early diagnosis and clinical decisions. Deep learning and transfer learning have shown significant ability to automating this classification process, although the effectiveness of various pretrained models can differ based on the dataset's characteristics. This study employs three pretrained deep learning models viz ResNet50, and DenseNet121 and Vision Transformer (ViT-B/16) to perform the multi-class lung disease classification using X-Ray images. The performance of all the models is measured using various metrics such as accuracy, precision, recall, F1-score, and validation accuracy to identify the best model for the lung disease classification. From the results of these models, it is found that DenseNet121 outperforms than the others by achieving accuracy of 92.49%, precision of 92.43%, recall of 92.49%, and F1-score of 92.45%. ResNet50 also shows good results, with accuracy of 92.05% and F1-score of 91.97%, while ViTB/16 reaches an accuracy of 88.18%. These findings suggest that DenseNet121 and ResNet50 are more suitable for lung disease classification tasks than ViT-B/16.
Lung disease classification, deep learning, transfer learning, Vision Transformer, ResNet50, DenseNet121, medical image classification, federated learning.