Explainable pulmonary fibrosis detection using edge-strengthened dilated holistic edge detection-based lung...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonary Medicine

Background:

• Pulmonary fibrosis (PF) is a progressive lung disease requiring early detection for better patient survival.
• Accurate diagnosis of PF from chest X-rays remains challenging.

Purpose of the Study:

• To develop and validate an explainable deep learning framework for pulmonary fibrosis detection using chest X-ray images.
• To integrate segmentation and classification models with explainability features for clinical decision support.

Main Methods:

• An edge-strengthened dilated holistic edge detection (ES-D-HED) segmentation network was combined with a ResNet152V2 classification model.
• The framework incorporated enhanced multi-scale edge fusion and Grad-CAM for explainability.
• The model was trained and validated on the NIH Chest X-ray dataset with expert-annotated PF regions.

Main Results:

• The framework achieved high performance with 98.6% classification accuracy, 98.0% sensitivity, 99.2% specificity, and 98.5% F1-score.
• Segmentation models showed strong alignment with expert annotations (Dice scores of 0.904 for lungs, 0.843 for PF regions).
• Grad-CAM visualizations confirmed the model's ability to identify fibrosis-related lung abnormalities.

Conclusions:

• The proposed explainable deep learning framework demonstrates significant potential for pulmonary fibrosis screening using chest X-rays.
• This approach offers interpretable insights for clinical decision-making in PF diagnosis.
• Future work includes multi-class ILD classification, fibrosis grading, and multi-center validation.