Deep Learning-based Opportunistic CT Osteoporosis Screening and the Establishment of Normative Values

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Bone Health and Osteoporosis

Background:

• Osteoporosis remains underdiagnosed and undertreated globally.
• Opportunistic screening using computed tomography (CT) and artificial intelligence (AI) offers a potential solution.
• Current methods lack standardization across different CT scanners and protocols.

Purpose of the Study:

• To develop a reproducible deep learning model for automated identification of three-dimensional (3D) regions of interest (ROI) in trabecular bone.
• To create a correction method for normalizing CT attenuation values across diverse protocols and scanner models.
• To establish diagnostic thresholds for osteoporosis in a large, diverse patient population.

Main Methods:

• A retrospective study utilizing a deep learning-based method for automated quantification of trabecular attenuation in spinal CT images.
• Development of a statistical method to adjust for variations in tube voltage and scanner models.
• Establishment of normative values and diagnostic thresholds based on World Health Organization osteoporosis prevalence data.

Main Results:

• Analysis of 538,946 CT examinations from 283,499 patients across 43 scanner models and six tube voltages.
• Automated ROI placement achieved >99% agreement with manual radiologist review.
• Identified significant differences in trabecular attenuation based on age, sex, and race, with established normative data.

Conclusions:

• Deep learning-based automated opportunistic screening can effectively identify patients with low bone mineral density from routine CT scans.
• The developed method demonstrates reproducibility across different scanners and protocols.
• This approach facilitates opportunistic osteoporosis screening, potentially improving diagnosis and management.