Generalizable framework for multi-site bone density prediction using non-dominant wrist optical biomarkers.

Area of Science:

• Biomedical Engineering
• Radiology
• Machine Learning

Background:

• Osteoporosis is a prevalent skeletal disorder marked by low bone mineral density (BMD) and increased fracture risk, especially in aging individuals.
• Current standard screening methods like dual-energy X-ray absorptiometry (DXA) involve radiation exposure and can be costly and time-consuming.

Purpose of the Study:

• To develop and validate a machine learning framework for intelligent optical bone densitometry (iOBD) using near-infrared (NIR) wrist imaging.
• To establish iOBD as a non-invasive, radiation-free alternative for preliminary osteoporosis screening.

Main Methods:

• NIR wrist images were acquired using six configurations (3 wavelengths x bilateral wrists).
• The optimal configuration (940 nm, left wrist) was selected based on tissue penetration and reduced variability.
• Features from NIR images (VGG16 network) and physiological data were used to predict BMD at five DXA sites.
• Model performance was evaluated using mean absolute percentage error (MAPE) and correlation coefficient (r).

Main Results:

• The VGG16 + KNN model achieved an 8.5% MAPE and r=0.718 for BMD prediction at the one-third distal radius.
• Prediction accuracy decreased at more distant skeletal sites (spine: r=0.460, femur: r<0.15), with reduced performance in elderly subjects.
• The iOBD framework demonstrated feasibility for radiation-free BMD assessment.

Conclusions:

• Intelligent optical bone densitometry (iOBD) using NIR wrist imaging shows potential as a radiation-free triage tool in primary care.
• This method can help identify patients who require confirmatory DXA scans, potentially reducing costs and scan times.
• Further research is needed to optimize performance at various skeletal sites and in diverse populations.