EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Segmentation-Guided Deep Learning Framework for Automated Respiratory Disease Diagnosis Using Enhanced Chest X-Ray Images

Muthulakshmi V, K Ramalakshmi

Computer and Decision Making: An International Journal · 03.11.2026 · 3 (6S) · 2948-2966

DOI: 10.1922/comdem.v3i6s.19047 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Computer and Decision Making: An International Journal, online erschienen am 03.11.2026. Verfasst von Muthulakshmi V und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Respiratory diseases are a major worldwide health problem, contributing to over 4.2 million deaths annually and demanding the automatization of screening procedures for early diagnosis and clinical diagnosis. The current research contributions include a Segmentation-Guided Deep Learning (DL) framework for automated respiratory disease diagnosis using chest X-ray (CXR) images to optimize. Detection of pneumonic abnormalities are less sensitive to low contrast and heterogeneous imagination conditions. Unlike conventional enhancement strategies, the proposed method uses Multi-Scale Retinex with Color Restoration (MSRCR)-based radiographic enhancement to improve boundary preservation, pneumonic texture visibility, and light standardization free from magnifying noise. Then, the optimized Attention U-Net segmentation network was used to precisely isolate lung regions and remove irrelevant anatomic organizations, thus improving disease-focused Feature Learning (FL). The metameric lung territories are then processed by the hybrid DenseNet121–EfficientNetB3 DL architecture combined with the channel attention and feature fusion module for discriminative classification of respiratory diseases. For lung segmentation, the Shenzhen chest X-ray dataset includes 662 chest radiograms with manually labeled lung masks used to train the Attention U-Net model. The respiratory disease classification was performed using the NIH ChestX-ray14, RSNA Pneumonia Detection, and COVID-19 Radiography Database. The segmentation module obtained a Dice Similarity Coefficient of 98.5% and a 96.8% IoU, enabling precise extraction of lung regions and improved disease-specific feature learning for subsequent classification. In the training process, progressive augmentation, focal loss optimization, and transfer learning strategies were integrated to deal with class imbalance and heterogeneous datasets. The experimental evaluation shows that, alongside 99.3% accuracy, 98.7% recall, F1-score of 98.2%, and a Dice similarity coefficient of 98.5%. The diagnostic accuracy is comparable to that of the current CNN, Transformer, and ensemble respiratory disease detection techniques.

Zitieren

APA

V, M., & Ramalakshmi, K. (2026). Segmentation-Guided Deep Learning Framework for Automated Respiratory Disease Diagnosis Using Enhanced Chest X-Ray Images. Computer and Decision Making: An International Journal, 3(6S), 2948-2966. https://doi.org/10.1922/comdem.v3i6s.19047

Vancouver

V M, Ramalakshmi K. Segmentation-Guided Deep Learning Framework for Automated Respiratory Disease Diagnosis Using Enhanced Chest X-Ray Images. Computer and Decision Making: An International Journal. 2026;3(6S):2948-2966. doi:10.1922/comdem.v3i6s.19047

BibTeX
@article{V2026,
  author = {V, Muthulakshmi and Ramalakshmi, K},
  title = {Segmentation-Guided Deep Learning Framework for Automated Respiratory Disease Diagnosis Using Enhanced Chest X-Ray Images},
  journal = {Computer and Decision Making: An International Journal},
  year = {2026},
  volume = {3},
  number = {6S},
  pages = {2948--2966},
  doi = {10.1922/comdem.v3i6s.19047}
}
Kostenlos starten →