Proximal femur segmentation and quantification in dual-energy subtraction tomosynthesis: A novel approach to fracture risk assessment.

This article focuses on improving bone density assessment in the proximal femur to enhance osteoporosis management by integrating tomosynthesis with dual-energy subtraction imaging and advanced segmentation models. Using a radiography/fluoroscopy system with dual-energy subtraction, the study optimized energy subtraction and scale factors to isolate bone from soft tissue in phantom models, and compared segmentation performance of convolutional neural networks, finding that a VGG19-based Faster Region-based Convolutional Neural Network (Faster R-CNN) outperformed U-Net with a mean intersection-over-union (IoU) of 0.865 versus 0.515. The findings suggest that this combined imaging and segmentation approach offers more accurate bone density quantification and segmentation than conventional methods, with potential clinical applications for fracture risk prediction and osteoporosis management. Limitations include the use of phantoms rather than clinical data and the need for further optimization and validation in diverse patient populations.