Opportunistic screening of low bone mass using knowledge distillation-based deep learning in chest X-rays with...

Area of Science:

• Artificial Intelligence in Medical Imaging
• Radiology and Bone Health

Background:

• Low bone mass (LBM) is an undiagnosed precursor to osteoporosis, increasing fracture risk.
• Early detection of LBM is crucial for preventing osteoporosis progression and associated complications.

Purpose of the Study:

• To develop and validate OsPenScreen, a deep learning model for opportunistic LBM detection.
• To utilize routine chest X-rays (CXRs) for early LBM identification, reducing healthcare burden.

Main Methods:

• A convolutional neural network (CNN) model, OsPenScreen, was trained on 77,812 CXR and DXA datasets.
• Knowledge distillation and validation across four independent datasets (5,935 images) were employed.
• Model performance was evaluated using AUC, accuracy, sensitivity, and specificity; Grad-CAM visualized decision-making.

Main Results:

• OsPenScreen achieved an AUC of 0.95 (95% CI: 0.94-0.97) on external test datasets.
• Consistent performance was observed across sex and age subgroups.
• The model accurately identified reduced bone mass, with attention focused on relevant bone areas.

Conclusions:

• OsPenScreen offers a scalable, effective tool for opportunistic LBM screening via routine CXRs.
• The model's robust performance across diverse datasets supports its potential for early detection.
• Early LBM detection can prevent osteoporosis progression and lower healthcare costs.