Deep Learning for Bone Mineral Density and T-Score Prediction from Chest X-rays: A Multicenter Study

Area of Science:

• Artificial Intelligence in Medicine
• Radiology and Medical Imaging
• Osteoporosis Research

Background:

• Osteoporosis diagnosis and treatment remain inadequate globally, despite rising patient numbers.
• Chest X-rays are a common, accessible, and cost-effective imaging modality.

Purpose of the Study:

• To develop a deep learning model for predicting bone mineral density (BMD) and T-scores from chest X-rays.
• To assess the model's efficacy in screening for osteoporosis.

Main Methods:

• A deep learning model was trained using ensemble learning on chest X-rays, age, and sex data from six hospitals (2010-2021).
• The model predicted BMD via regression and T-scores via multiclass classification.
• Performance was evaluated by correlating predicted and true BMDs and assessing T-score classification consistency.

Main Results:

• The model achieved BMD prediction correlation coefficients of 0.75 for the hip and 0.63 for the lumbar spine.
• Areas under the curve for T-score predictions (normal, osteopenia, osteoporosis) were 0.89, 0.70, and 0.84, respectively.
• These metrics indicate strong performance in predicting osteoporosis-related bone density metrics.

Conclusions:

• The developed deep learning model demonstrates significant potential for osteoporosis screening.
• Predicting BMD and T-scores from chest X-rays using AI can improve early detection and management of osteoporosis.