Publication date: Jun 24, 2026
Lung disease classification using chest X-ray (CXR) images has become essential for early diagnosis and improved clinical decision-making. However, challenges such as low image quality, feature similarity among diseases, and classification instability reduce diagnostic reliability. To address these issues, this study proposes a novel ELSTM-AZOA framework for multiclass lung disease classification using the NIH CXR dataset. Initially, the collected CXR images are preprocessed using the balance contrast enhancement technique to improve image quality. U-Net + + is then employed for accurate lung region segmentation, followed by feature extraction using statistical and gray level co-occurrence matrix features. The extracted features are classified using an enhanced long short-term memory (ELSTM) network, while the American zebra optimization algorithm (AZOA) optimizes the model parameters to maximize classification accuracy. The proposed framework classifies six categories: healthy lung, tuberculosis, pneumonia, lung cancer, COPD, and COVID-19. Experimental results demonstrate that the proposed ELSTM-AZOA model achieves superior performance compared with existing methods, obtaining 6. 36% higher accuracy and 6. 43% higher precision. The findings confirm that the proposed framework provides robust, reliable, and promising computer-aided lung disease classification.
Open Access PDF
| Concepts | Keywords |
|---|---|
| Cancer | GLCM features |
| Matrix | Lung disease classification |
| Pneumonia | NIH CXR dataset |
| Reliable | Statistical features |
| Zebra | U-Net + + |