Pulmonary diseases identification: Deep learning models and ensemble learning
Artykuł w czasopiśmie
MNiSW
70
Lista 2024
| Status: | |
| Autorzy: | Kwaśniewska Patrycja, Zieliński Grzegorz, Powroźnik Paweł, Skublewska-Paszkowska Maria |
| Dyscypliny: | |
| Aby zobaczyć szczegóły należy się zalogować. | |
| Rok wydania: | 2025 |
| Wersja dokumentu: | Drukowana | Elektroniczna |
| Język: | angielski |
| Numer czasopisma: | 3 |
| Wolumen/Tom: | 21 |
| Strony: | 38 - 58 |
| Scopus® Cytowania: | 0 |
| Bazy: | Scopus |
| Efekt badań statutowych | NIE |
| Materiał konferencyjny: | NIE |
| Publikacja OA: | TAK |
| Licencja: | |
| Sposób udostępnienia: | Witryna wydawcy |
| Wersja tekstu: | Ostateczna wersja opublikowana |
| Czas opublikowania: | W momencie opublikowania |
| Data opublikowania w OA: | 18 sierpnia 2025 |
| Abstrakty: | angielski |
| Deep learning models provide tremendous support for medical imaging by understanding lung conditions and indicating multiple lung diseases. Due to the global burden of respiratory diseases, their prevention and control is of great importance. Therefore, this study focuses on the effectiveness of different deep learning architectures in diagnosing lung diseases from chest X-ray images. Five deep convolutional neural networks are involved: VGG16, DenseNet-121, ResNet-50, MobileNet, and Vision Transformers. They are pre-trained using the ImageNet dataset. Both transfer learning and development of custom models based on the above architectures will be applied. The study deals with the determination of the most effective single model for the identification of lung diseases. The gradient-weighted class activation map is used to highlight the key regions that influence model decisions. In addition, soft voting ensemble learning methods are used to improve the performance of lung disease detection. Commonly used metrics are applied to evaluate all models. The results for COVID-19, pneumonia and normal case identification exceeded 95% accuracy, 95% precision, 96% recall and 95% Fβ for individual models. The ViT model outperformed DenseNet-121, achieving 96.66% accuracy. The results for bacterial pneumonia, viral pneumonia, tuberculosis, COVID-19 and healthy case identification exceeded 85% accuracy, 86% precision, 85% recall and 94% Fβ for single models. Ensemble learning further improved performance. These results demonstrate the high potential of deep learning and ensemble approaches to support accurate and efficient diagnosis of lung diseases using chest X-rays. The deep learning models provide promising decision support tools for this type of healthcare diagnosis |
