Deep learning meets chest X-rays: a promising approach for predicting future compression fracture risk.

Area of Science:

• Radiology
• Artificial Intelligence
• Osteoporosis Research

Background:

• Osteoporotic fractures pose a significant global health challenge, causing disability and reduced quality of life.
• Current diagnostic tools like DXA have limitations in identifying all at-risk individuals.
• There is a need for accessible and noninvasive methods for osteoporotic fracture risk assessment.

Purpose of the Study:

• To develop and validate an AI-enabled chest X-ray (AI-CXR) model for predicting osteoporotic fracture risk.
• To offer a noninvasive and accessible alternative to existing diagnostic methods.
• To assess the AI-CXR model's predictive accuracy compared to DXA-based T-scores.

Main Methods:

• Retrospective analysis of 166,571 chest X-rays from 78,548 patients in Taiwan.
• Internal validation on 31,977 X-rays and external validation on 36,677 X-rays.
• Extraction of radiological features (e.g., costophrenic angle blunting, degenerative joint disease) for the predictive model.

Main Results:

• The AI-CXR model demonstrated superior predictive accuracy over DXA, especially for patients without T-scores (internal validation: 0.896 vs 0.829; external validation: 0.778 vs 0.818).
• High-risk groups identified by AI-CXR showed significantly higher 5-year fracture incidence (internal: 2.6% vs 0.3%; external: 3.5% vs 0.5%).
• Key radiological features were more prevalent in high-risk groups, and the model showed consistent performance across demographic subgroups.

Conclusions:

• The AI-CXR model serves as a cost-effective, noninvasive tool for osteoporotic fracture risk assessment.
• This AI-driven approach enables improved early detection of fractures.
• Personalized interventions can be facilitated across diverse clinical settings through this novel risk assessment tool.