Enhanced opportunistic CT screening for osteoporosis using Machine learning derived volumetric vertebral and...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Bone Health Research

Background:

• Osteoporosis diagnosis relies on bone mineral density (BMD) assessment, often using dual-energy X-ray absorptiometry (DXA).
• CT imaging offers detailed anatomical information, but its utility for BMD prediction and osteoporosis classification is underexplored compared to traditional methods.
• Deep learning (DL) techniques show promise in extracting complex features from medical images.

Purpose of the Study:

• To evaluate if integrating volumetric vertebral and body composition features from CT images, using DL segmentation, improves BMD prediction and osteoporosis classification.
• To compare the performance of DL-based feature analysis against conventional single-slice lumbar vertebral attenuation.
• To determine the added value of body composition metrics and clinical data in these predictions.

Main Methods:

• A retrospective study of 383 adults with same-day CT scans and DXA measurements.
• Development of a 3DnnU-Net for segmenting thoracolumbar vertebrae and a 3D U-Net (DeepCatch) for segmenting muscle and fat.
• Building prediction models using vertebral features, combined features, and clinical data; comparison with linear regression on single-slice CT attenuation.

Main Results:

• Volumetric vertebral features significantly improved BMD prediction (lumbar spine R=0.92) and osteoporosis classification (AUROC=0.95) compared to single-slice attenuation.
• Adding body composition metrics enhanced hip BMD prediction and increased osteoporosis classification sensitivity (86%) while maintaining high specificity (95%).
• Clinical variables (age, sex, BMI) did not provide additional predictive benefit.

Conclusions:

• Deep learning segmentation of CT images enables accurate prediction of lumbar and femoral BMD.
• This approach significantly improves the sensitivity of osteoporosis detection.
• Integrating volumetric vertebral and body composition features from CT scans is a promising method for skeletal health assessment.