Advancing subject-specific biomechanical models of soft tissues
Loading...
Authors
Lampen, Nathan, Jon
Issue Date
2024-12
Type
Electronic thesis
Thesis
Thesis
Language
en_US
Keywords
Biomedical engineering
Alternative Title
Abstract
Subject-specific modeling of soft tissue biomechanics is crucial for precise and effective medical interventions, particularly in simulating disease progression and surgical outcomes. Traditional finite element method (FEM) models, while accurate, often lack subject-specific material properties, are computationally intensive, and require labor-intensive manual mesh generation. This thesis addresses these limitations by integrating quantitative imaging, deep learning, and automated meshing techniques to enhance the accuracy, efficiency, and applicability of subject-specific models in clinical settings. The research in this thesis is centered around three specific aims: First, we develop a method to refine the mechanical properties of soft tissues in FEM models using quantitative magnetic resonance imaging (MRI) data. By using T2 relaxometry data to refine regional material properties in FEM models, we enhance their specificity and accuracy, enabling more precise simulations of tissue mechanics. Testing on subjects from the Osteoarthritis Initiative dataset shows that T2-refined models estimate more localized principal stresses and shear strains, with better correlation to MRI Osteoarthritis Knee Scores compared to homogeneous models. Second, we propose a deep learning framework designed to accelerate biomechanics simulations of soft tissue deformation. Utilizing the PointNet++ architecture, our network accepts detailed facial mesh point cloud data and input displacement to predict deformation efficiently. To further enhance simulation accuracy, we introduce a spatiotemporal incremental method, incorporating spatial and temporal information via a Graph Neural Network and a temporal memory mechanism. This approach significantly reduces simulation time compared to FEM while maintaining high accuracy, demonstrating the feasibility of rapid simulations in orthognathic surgical planning. Third, we present an automated method for generating subject-specific meshes of facial soft tissues from cone beam computed tomography images using Google MediaPipe for real-time facial landmark detection. This novel approach automates the creation of volumetric meshes, significantly reducing preparation time and enhancing the efficiency and scalability of personalized surgical planning. By addressing these critical issues, this thesis advances the field of subject-specific soft tissue biomechanics modeling, improving the accuracy, speed, and clinical utility of simulations. The methodologies developed in this work enhance patient outcomes, streamline clinical workflows, and broaden the application of personalized medicine, ultimately leading to more precise and effective medical interventions.
Description
December2024
School of Engineering
School of Engineering
Full Citation
Publisher
Rensselaer Polytechnic Institute, Troy, NY