Supervised learning for bone shape and cortical thickness estimation from CT images for finite element analysis.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computational Anatomy

Background:

• Accurate cortical bone thickness is crucial for fracture risk assessment.
• Standard CT image resolution limits the precision of finite element models.
• Existing methods often overlook detailed cortical bone geometry.

Purpose of the Study:

• To develop a pipeline for precise cortical bone thickness estimation from CT scans.
• To generate high-quality finite element meshes with patient-specific bone shapes.
• To improve fracture risk assessment by incorporating sub-voxel cortical bone data.

Main Methods:

• A three-step approach combining shape regression, supervised learning with QCT/HRpQCT data, and Gaussian process regularization.
• Generation of initial surface meshes using morphometric features and statistical shape models.
• Correction and regularization of meshes to achieve sub-voxel precision in cortical thickness estimation.

Main Results:

• The pipeline accurately estimates cortical thickness (error = 0.05 ± 0.40 mm) and produces robust outer surface meshes (RMSE = 0.36 ± 0.29 mm).
• Generated meshes exhibit high quality (aspect ratio = 1.4 ± 0.02), suitable for finite element simulations.
• Consistent mesh node and element numbering across different bone morphologies was achieved.

Conclusions:

• The proposed method enables the creation of patient-specific finite element models with accurate cortical bone thickness directly from CT scans.
• This technique overcomes the resolution limitations of CT imaging for bone analysis.
• The consistent mesh generation offers significant advantages for large-scale population studies in bone research.