Can DXA image-based deep learning model predict the anisotropic elastic behavior of trabecular bone?

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computational Biology

Background:

• Assessing trabecular bone's anisotropic elastic behavior is crucial for bone fracture risk evaluation.
• Current methods like finite element (FE) and bone volume fraction (BV/TV)/fabric tensor modeling are complex and resource-intensive.
• Deep learning (DL) has shown promise in various image-based prediction tasks.

Purpose of the Study:

• To investigate the potential of DL techniques for predicting the apparent stiffness tensor of trabecular bone directly from dual-energy X-ray absorptiometry (DXA) images.
• To develop a convolutional neural network (CNN) model capable of this prediction.
• To compare the accuracy of the DL model against established methods.

Main Methods:

• Trabecular bone cubes from human cadaver proximal femurs were used.
• Simulated DXA images served as input for the CNN model.
• Micro-CT based FE simulations provided the ground truth for the apparent stiffness tensor.
• The DL model's predictions were compared with FE models and multiple linear regression models using histomorphometric parameters and BV/TV/fabric tensor.

Main Results:

• The DXA image-based DL model demonstrated high fidelity in predicting the apparent stiffness tensor of trabecular bone cubes.
• Prediction accuracy (R² values ranging from 0.905 to 0.973) was comparable to or surpassed traditional regression models.
• The study successfully supported the hypothesis that DL can predict bone stiffness from DXA images.

Conclusions:

• DL models, specifically CNNs, can accurately predict the apparent stiffness tensor of trabecular bone using DXA images.
• This DXA image-based DL approach offers a promising, potentially more accessible alternative to current complex modeling techniques.
• The findings pave the way for developing clinical DXA-based DL tools for improved bone fracture risk assessment.