Finite element models with automatic computed tomography bone segmentation for failure load computation.

Area of Science:

• Medical Imaging
• Biomechanical Engineering
• Artificial Intelligence

Background:

• Accurate bone segmentation is crucial for biomechanical failure load simulations using patient CT data.
• Manual segmentation is time-consuming, labor-intensive, and prone to variability.
• Limited availability of large annotated datasets hinders deep learning model generalization.

Purpose of the Study:

• To develop and evaluate an automated pipeline for segmenting in-vivo human femurs and vertebrae from CT scans.
• To assess the quality and effectiveness of deep learning-based segmentation for biomechanical simulations.
• To compare simulation outcomes using automated versus manual segmentations.

Main Methods:

• A dedicated pipeline involving preprocessing, U-Net based deep learning segmentation, and post-processing was implemented.
• Three U-Net architectures were experimented with for segmenting CT volumes of femurs and vertebrae.
• Biomechanical failure load simulations were performed using both automated and manual segmentations.

Main Results:

• Carefully trained U-Net models achieved high-quality, automatic volume segmentation of human femurs and vertebrae.
• Failure load simulations using automated segmentations yielded results comparable to those from manual expert segmentations.
• The sensitivity of simulations to minor variations in automated segmentation was analyzed.

Conclusions:

• The proposed automated segmentation approach is effective for biomechanical failure load simulations.
• Deep learning models, when carefully trained, can achieve high-quality bone segmentation from CT data.
• Automated segmentation offers a viable and efficient alternative to manual methods in clinical biomechanics.