A deep learning-based model to estimate pulmonary function from chest x-rays: multi-institutional model development...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonary Medicine

Background:

• Chest x-rays are standard for anatomical assessment but their utility for dynamic pulmonary function estimation is unclear.
• Pulmonary function tests like spirometry are crucial for diagnosing and managing lung diseases.
• Estimating pulmonary function from readily available chest x-rays could improve accessibility and patient care.

Purpose of the Study:

• To develop and validate a deep learning model to estimate forced vital capacity (FVC) and forced expiratory volume in 1 second (FEV1) from chest x-rays.
• To assess the model's performance against traditional spirometry measurements.

Main Methods:

• A retrospective study involving over 140,000 chest x-ray and spirometry pairs from five Japanese institutions.
• Development and external validation of a deep learning-based AI model using chest x-rays.
• Performance evaluation using Pearson's correlation coefficient, ICC, MSE, RMSE, and MAE.

Main Results:

• The AI model demonstrated high accuracy in estimating FVC and FEV1 from chest x-rays, with strong correlations (r ≈ 0.90-0.91) and low error metrics in external validation.
• External testing showed excellent agreement between AI-estimated FVC/FEV1 and spirometry results.
• The model achieved robust performance across different institutions.

Conclusions:

• A deep learning model can accurately estimate FVC and FEV1 from chest x-rays, serving as a viable alternative to spirometry.
• This AI tool can benefit patients unable to undergo spirometry and potentially optimize CT imaging protocols.
• Further research integrating clinical data can enhance the model's diagnostic and management capabilities for lung diseases.