Patient-Specific Heart Mechanics and Biomarker Estimation from Medical Imaging
This project aims to bridge the gap between medical MRI images and clinical decision-making by developing a complete computational pipeline to estimate mechanical biomarkers such as tissue wall stress using FEniCSx, with the potential to explore surrogate/machine learning models to accelerate the workflow.
Biomechanical markers, such as tissue wall stress and myocardial stiffness, provide vital insights into cardiovascular disease progression, risk stratification, and patient outcome prediction. However, local tissue stress cannot be directly measured in vivo. Magnetic Resonance Imaging (MRI) provides rich anatomical and functional motion data.
To translate these images into clinical mechanical biomarkers, computational mechanics—specifically Finite Element Methods (FEM)—is used to solve the partial differential equations governing continuum mechanics. While accurate, patient-specific FEM simulations using FEniCSx can be computationally demanding and time-consuming. To make biomechanical modeling practical for clinical applications, the primary focus of this project is to build and evaluate a complete end-to-end pipeline: from medical image processing through finite element simulations in FEniCSx to tissue stress calculation.
Existing open-source tools, such as biv-me for image-to-mesh conversion and fenicsx-pulse for cardiac mechanics simulation, provide strong foundations for this computational pipeline. Developing this baseline workflow forms a robust core for the thesis. If time permits, the student can further explore modern computational techniques, machine learning, or surrogate modeling to accelerate the end-to-end pipeline while maintaining physical fidelity.
Goals
The main goal of this master's thesis is to build and evaluate a complete computational workflow that maps medical image data (e.g., MRI) -> finite element simulation (FEniCSx) -> clinical biomarkers (e.g., wall stress).
As the primary deliverable, the student will construct this full end-to-end pipeline, leveraging existing tools such as biv-me and fenicsx-pulse. Once the functional pipeline is established, and depending on thesis scope and time, the project can be extended to investigate methods for speeding up the workflow using data-driven surrogate models (e.g., neural networks or physics-informed architectures).
Learning outcomes
By completing this project, the student will gain hands-on experience in:
- End-to-end biomechanical modeling, continuum mechanics, and medical image processing
- Implementing finite element solutions using FEniCSx and specialized computational libraries such as fenicsx-pulse
- Automated mesh generation and image transformation workflows using tools such as biv-me
- Turning clinical MRI datasets into actionable mechanical biomarkers
- (Optional/Extension) Designing data-driven surrogate models and machine learning frameworks (e.g., PyTorch, JAX) for scientific computing and physics acceleration
- Conducting open, reproducible, and collaborative scientific research
Qualifications
Required (need-to-have):
- Solid programming skills in Python
- Basic understanding of numerical methods, linear algebra, or mechanics
Nice-to-have:
- Familiarity with finite element modeling in FEniCSx
- Experience or interest in medical image processing or mesh generation tools
- Familiarity with deep learning frameworks (e.g., PyTorch, JAX) or surrogate modeling techniques
- Knowledge of continuum mechanics or partial differential equations
Supervisors
- Nickolas Ivan Forsch
- Henrik Nicolay Finsberg