Estimating Total Lung Volume from Pixel-Level Thickness Maps of Chest Radiographs Using Deep Learning

Area of Science:

• Radiology
• Medical Imaging
• Deep Learning

Background:

• Accurate estimation of total lung volume (TLV) is crucial for diagnosing and managing respiratory diseases.
• Traditional methods for TLV estimation often rely on computed tomography (CT), which is resource-intensive.
• Developing non-invasive methods using standard frontal chest radiographs is highly desirable.

Purpose of the Study:

• To estimate total lung volume (TLV) from frontal chest radiographs using pixel-level lung thickness maps.
• To utilize a U-Net deep learning model for generating these lung thickness maps.
• To validate the model's performance on both synthetic and real radiographic data.

Main Methods:

• A retrospective study utilized chest CT scans from public datasets (Luna16, PE Detection Challenge) and a clinical dataset (Klinikum Rechts der Isar).
• Synthetic frontal chest radiographs and lung thickness maps were generated via forward projection of CT scans.
• A U-Net model was trained on synthetic data to predict lung thickness maps and subsequently estimate TLV, with performance assessed by MSE and Pearson correlation.

Main Results:

• The U-Net model demonstrated low error rates and strong correlations between predicted and CT-derived TLV across synthetic and real datasets (e.g., r=0.91, P < .001 for real data).
• The model achieved the highest performance on the Luna16 test dataset, showing the lowest MSE (0.09 L²) and strongest correlation (r=0.99, P < .001).
• Pixel-level lung thickness maps generated by the U-Net model successfully estimated TLV.

Conclusions:

• U-Net-generated pixel-level lung thickness maps provide a viable method for estimating total lung volume from frontal chest radiographs.
• The approach is effective for both synthetic and real radiographic images, offering a promising tool for clinical applications.
• This deep learning-based method enhances the utility of standard chest radiographs for lung volume assessment.