Deep Learning-Based Surrogate Model of Subject-Specific Finite-Element Analysis for Vertebrae.

Area of Science:

• Computational biomechanics
• Medical imaging
• Machine learning applications in healthcare

Background:

• Subject-specific finite-element analysis (FEA) models are crucial for simulating vertebral biomechanics.
• Traditional FEA methods are computationally intensive, limiting their clinical application.
• Need for efficient tools to analyze vertebral body stress distributions.

Purpose of the Study:

• To develop and validate a novel deep learning (DL)/machine learning (ML) surrogate model for predicting stress in vertebral bodies.
• To significantly decrease the time required for subject-specific biomechanical assessments.
• To create an automated pipeline for rapid clinical integration.

Main Methods:

• Developed a DL/ML surrogate model integrating vertebral shape encoding with separate decoding for surface and internal nodes.
• Trained the model on 3,960 synthetic L1 vertebrae derived from 42 real computed tomography (CT) scans using data augmentation.
• Evaluated model accuracy using mean absolute error (MAE) and R-squared (R²) for von Mises stress on independent test data.

Main Results:

• The surrogate model achieved a mean absolute error (MAE) of 0.0596 MPa and an R² of 0.864 for von Mises stress.
• Predicted stress patterns showed strong agreement with FEA-computed results, with minor discrepancies at specific anatomical locations.
• An automated pipeline reduced processing time from 90-120 minutes to approximately 134-154 seconds per subject.

Conclusions:

• The proposed DL/ML surrogate model offers a highly efficient alternative to traditional FEA for vertebral biomechanics.
• The model demonstrates potential for facilitating rapid, subject-specific biomechanical assessments in clinical workflows.
• This approach can accelerate diagnosis and treatment planning for spinal conditions.