XTC-Net: an explainable hybrid model for automated atelectasis detection from chest radiographs

Area of Science:

• Artificial Intelligence in Medicine
• Medical Imaging Analysis
• Deep Learning for Diagnostics

Background:

• Atelectasis (lung collapse) poses diagnostic and treatment challenges.
• Timely identification is crucial for preventing complications and enabling early intervention.
• Automated AI detection can enhance diagnostic efficiency and reduce clinical workload.

Purpose of the Study:

• To develop an interpretable deep learning model for accurate atelectasis detection from chest radiographs.
• To improve diagnostic efficiency and patient care through automated analysis.

Main Methods:

• Developed a deep learning model integrating Xception, Transformer, and Capsule Network components.
• Utilized Xception for spatial feature extraction and Transformer for long-range dependency modeling.
• Employed Capsule Network to enhance sensitivity to subtle structural variations in atelectasis.

Main Results:

• Achieved 99.73% accuracy, 99.74% sensitivity, and 99.73% F1 score on a public chest X-ray dataset.
• Demonstrated consistent performance on external validation using the NIH ChestX-ray dataset.
• Highlighted the model's generalizability and applicability beyond the primary dataset.

Conclusions:

• The proposed interpretable deep learning model reliably detects atelectasis from chest radiographs.
• The system shows potential for integration into clinical workflows for automated diagnosis.
• Results underscore the capability of AI in enhancing medical imaging analysis and patient outcomes.