Answers are grounded in this paper and cite their sources.
Analysis of lung disease allows for early detection and treatment of disease, reduced complication rates, and overall best outcomes. Lung Ultrasound (LUS) uses real‑time safe imaging to facilitate disease detection. In this study, a hybrid approach model integrating deep learning and the Transformer model was proposed for automatic classification of lung disease from LUS image. This dataset is used consisting of 112 LUS videos with categories 40 in COVID-19, 40 in Other Lung Disease and 32 as healthy lung videos. Data Preprocessing applied include resizing, batching, shuffling, caching, and pre‑fetching while the Data augmentation technique including random horizontal flipping, rotating, zooming, and translation for models Generalization. A pre-trained VGG16 network with its initial layers frozen was employed for transfer learning and local feature extraction. Mutual Information was subsequently used to select the most informative features, which were processed by a Swin Transformer to capture contextual relationships. The resulting representation was passed through global average pooling, a fully connected layer, and a softmax layer for final classification. The proposed hybrid architecture achieved an accuracy of 96.40%, demonstrating improved classification performance compared with the evaluated existing approaches.
No takes yet. Share an insight, caveat, or question.
K et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: