End-to-End Semi-Supervised Opportunistic Osteoporosis Screening Using Computed Tomography

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Osteoporosis is a prevalent metabolic bone disease leading to fragility fractures.
• Low screening rates for osteoporosis persist in at-risk populations.
• Automated bone mineral density (BMD) estimation via computed tomography (CT) offers an alternative to dual-energy X-ray absorptiometry (DXA).

Purpose of the Study:

• To investigate the feasibility of an opportunistic and population-agnostic osteoporosis screening method using abdominal CT scans.
• To evaluate automated BMD estimation without bone densitometry phantom-based calibration.

Main Methods:

• Retrospective study of 268 abdominal CT-DXA pairs and 99 abdominal CT scans.
• Deep learning models trained to localize lumbar vertebrae (L1-L4) and estimate BMD.
• Analysis of CT scans from an oncology specialty clinic.

Main Results:

• Vertebra-level automated BMD predictions achieved a mean absolute error (MAE) of 0.079 and R2 of 0.714.
• Subject-level predictions demonstrated improved accuracy with MAE of 0.066 and R2 of 0.781.
• High correlation (Pearson's r=0.907) between predicted and actual BMD at the subject level.

Conclusions:

• CT scans from routine examinations can generate accurate DXA BMD predictions.
• This automated method bypasses the need for bone densitometry calibration.
• Enables opportunistic osteoporosis screening in diverse patient populations.