Student Projects

Cardiac diffusion tensor imaging (cDTI), a specialized magnetic resonance imaging (MRI) method, provides information about the cardiac microstructure by measuring the diffusion of water molecules within the heart muscle. The cardiac muscle is constituted of four major intracellular compartments (myocytes, mural cells, collagen and blood vessels) that are connected to the extracellular space through permeable membranes. These compartment boundaries restrict the water diffusion in certain directions and thus change the measured MRI signal. The proposed compartment modelling approach utilizes different mesh shapes such as spheres and cylinders to model different cell types. Cell type specific parameters such as diffusivity and initial spin density are added as model properties resp. will be integrated into the partial differential equations describing the diffusion process. The cell type specific membrane permeabilities are added as PDE boundary conditions. The volume fractions of the four cardiac muscle compartments and the cell orientation will be extracted from literature, cardiac muscle histology and cDTI data. The compartment model can be extended with additional functionalities such as blood flow through the blood vessels.

The project aims to develop a compartment model for the cardiac muscle including the four major compartments (myocytes, mural cells, collagen and blood vessels). The compartment model is based on the SpinDoctor modelling approached used in brain DTI. The model will in the future be incorporated in CMRsim, an open-source python library for cardiac magnetic resonance simulation. It will be used for diffusion simulations by solving partial differential equations with finite elements.

Sandra Haltmeier (haltmeier@biomed.ee.ethz.ch) Isabel Margolis (margolis@biomed.ee.ethz.ch)

Cardiac diffusion tensor imaging (cDTI) provides information about the cardiac microstructure by measuring the diffusion of water molecules within the heart wall. Current imaging standards measure three slices distributed across the left ventricle. However, if not corrected, respiratory motion causes slice misalignments that obstruct microstructure inference. Yet, this motion might also allow us to estimate sample points between slices, thus adjusting for motion and increasing spatial coverage. The approach utilizes respiratory motion and 3D mesh mapping in cardiac diffusion tensor imaging (cDTI). By leveraging respiratory navigator data, you will map in-vivo cDTI to a 3D digital twin mesh and implement tensor estimation to discern sample points between slices. This not only corrects motion-induced misalignments but also expands spatial coverage. The method will then be rigorously evaluated for accuracy using simulated data to showcase its potential to advance cDTI. ** images from: - https://doi.org/10.1002/cnm.3178 - https://doi.org/10.1186/s12968-017-0342-x - https://doi.org/10.1371/journal.pone.0072795

The project aims to utilize respiratory motion to estimate sample points between slices and thus increase spatial coverage by applying spatial smoothness constraints on a digital twin mesh. Accuracy evaluation is performed on simulated data.

Sandra Haltmeier (haltmeier@biomed.ee.ethz.ch)

In (cardiovascular) MRI, a single voxel is often considered to be a homogeneous tissue. In reality however, voxels often contain a combination of tissues, each consisting of multiple inter- und extracellular compartments. With new methods like T1ρ imaging, slow molecular exchange processes between compartments can be probed. To disentangle signals from these compartments and subsequently microstructural information, optimized strategies for image acquisition (through simulation), reconstruction, and fitting need to be developed.

This project aims to develop a multi-exponential fitting technique that can derive relaxation times per compartment from a simulated sequence. To this end, sequence (acquisition) parameters also need to be optimized to obtain a good fit quality. Based on optimization results from simulation, validation can be done on data acquired with our state-of-the-art MRI scanner. For more information, please contact us. **Prerequisites** Basic knowledge of scientific programming and MRI physics is required, experience in Python is strongly recommended.

Vincent Vousten (vousten-at-biomed.ee.ethz.ch)

4D Flow MRI enables non-invasive, time-resolved imaging of blood flow in the cerebral vasculature but suffers from low spatial and temporal resolution, as well as noise and artifacts, limiting its accuracy. CFD simulations provide high-resolution flow data using patient-specific vascular geometries, offering a valuable benchmark for comparison. This project aims to compare hemodynamic parameters (such as pressure gradients) derived from CFD, 4D Flow MRI, and potential invasive pressure measurements, identifying the strengths and limitations of each method. The findings will improve the reliability of non-invasive techniques for cerebral hemodynamic assessment.

The goal of this project is to compare methods for computing pressure gradients in the cerebral vasculature. The student will perform high-resolution CFD simulations of patient-specific cerebral vasculature, focusing on pathologies such as stenosis. A method for computing pressure gradients from low-resolution 4D Flow MRI velocity fields will be developed and evaluated. Pressure gradients from 4D Flow MRI will be compared to those from CFD simulations and potentially validated with invasive in-vivo data. The project will make use of our existing numerical tools based on open-source libraries for Finite Volumes (OpenFOAM) and data processing (Python).

Supervisors: Jonathan Nestmann (nestmann@biomed.ee.ethz.ch) and Luuk Jacobs (jacobs@biomed.ee.ethz.ch). To apply for this project, please send a copy of your CV and transcripts of your Bachelor and Master studies.

4D flow MRI captures cerebrovascular blood flow, but achieving high-quality segmentation remains challenging due to noise, artifacts, and the limited availability of annotated data. Current 3D CNN-based methods for vascular segmentation struggle with the lack of diverse training datasets and ground truth segmentation.

In this project, the student will enhance an existing 3D Convolutional Neural Network (CNN) for the segmentation of 4D Flow MRI data. The aim is to improve the 3D segmentation of cerebral vasculature by expanding the training dataset to include diverse patient data with various pathologies and using data augmentation. Additionally, generative AI techniques will be used to create synthetic 4D Flow MRI data, providing a more varied and accurate training set. This will enhance the performance of 3D CNNs, particularly in challenging areas of the brain. The project will be implemented in Python, utilizing open-source libraries such as PyTorch and PyVista.

Supervisors: Jonathan Nestmann (nestmann@biomed.ee.ethz.ch) and Luuk Jacobs (jacobs@biomed.ee.ethz.ch). To apply for this project please send a copy of your CV and transcripts of your Bachelor and Master studies.

Phase Contrast Cardiovascular Magnetic Resonance MR Imaging (PC-MRI) allows for the subject specific quantitative assessment of blood flow. As abnormal flow patterns are related to the evolution of cardiovascular disease, automatic tools for hemodynamic and flow analysis are a fundamental step in diagnosing and monitoring disease progression. However, limited data availability and lack of ground truth information prohibit the development of generalized and robust models.

The aim of this project is to develop a method using physics-informed neural networks to learn aortic flow behaviour and identify aortic hemodynamic properties, which hold important information used for the assessment of aortic stenosis. The student will develop and expand existing physics-informed neural network approaches, as proposed in the literature, to analyse gemetrical and velocity field properties from in-silico generated aortae. Synthetic and subsequently real data will be used to test the approach. The project will be carried out in Python and make use of open-source libraries such as Pytorch and Pyvista. **We offer** - A warm start of the project with the state-of-the-art knowledge of the group in this field - A chance to connect state of the art research in MRI with modern ML tools

Gloria Wolkerstorfer (wolkerstorfer@biomed.ee.ethz.ch), Dr. Stefano Buoso (buoso@biomed.ee.ethz.ch). To apply for this project, please email a copy of your CV and transcripts of your Bachelor and/or Master studies.

Late gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR) images are the gold standard for myocardial scar imaging. Classifying healthy and scarred tissue is an essential clinical step for treatment planning. In the last years, many automatic approaches based on deep learning have been proposed for the task. However, they require large annotated datasets, which are expensive and time-consuming to generate. To overcome this limitation, we are interested in generating a synthetic dataset for the task.

In this project, the student will enhance an existing pipeline to generate synthetic LGE images with realistic background textures and details. These synthetic images will then be used to train and test models for myocardium and scar segmentation. Specifically, the segmentation performance of nnU-Net on the synthetic images will be evaluated and compared to the performance of nnU-Net models trained on real, in-vivo LGE data. A clinical dataset, which includes LGE images and their corresponding segmentation masks, has been made available for this project.

Isabel Margolis, margolis@biomed.ee.ethz.ch

Late gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR) imaging is the gold standard for visualizing myocardial scars, which is crucial for classifying healthy versus scarred tissue and guiding treatment decisions. However, current methods are often unreliable due to intra- and inter-observer variability, lack of consensus, and limited standardization. To address these challenges, this project aims to generate synthetic data based on high-resolution, ground truth scar patterns, which can be used to train deep learning models for automatic and reliable scar segmentation.

In this project, the student will explore existing computational models and deepen their understanding of cardiac fibrosis and myocardial scar formation. The student will then extend an existing pipeline to simulate high-resolution scar patterns. Starting points for the literature review will include an overview of computational models of cardiac electrophysiology and fibrosis analysis based on histology or high-resolution CMR imaging techniques.

Isabel Margolis, margolis@biomed.ee.ethz.ch

Phase Contrast Cardiovascular Magnetic Resonance MR Imaging (PC-MRI) allows for the subject specific quantitative assessment of blood flow. As abnormal flow patterns are related to the evolution of cardiovascular disease, automatic tools for flow analysis are a fundamental step for the clinical adoption of PC-MRI data in clinical practice. Before extracting meaningful biomarkers from patients’ flow measurements, data needs to be denoised, augmented and filtered. Digital twins, which are in-silico replica of patients, could be used for this task, as they provide a computational framework suitable for data processing.

The aim of this project is to develop an approach based on physics-based graph neural networks to generate digital twins from PC-MRI data. These will combine the flexibility of graph neural networks, which could easily adapt to each patient, with physical constraints defined by physiological priors injected to the network. The investigation could target blood flow in the aorta, in carotids or in the brain The student will develop and expand existing graph neural networks proposed in the literature to work with flow measurements from PC-MRI. Either real or synthetic data will be used to test the approach. The project will make use of our existing numerical tools based on open-source libraries for Finite Elements (FEniCS), Finite Volumes (openFOAM) and machine learning (Pytorch).

Supervisors: Dr. Stefano Buoso (buoso@biomed.ee.ethz.ch). To apply for this project please email a copy of your CV and transcripts of your Bachelor and/or Master studies.

Cardiovascular Magnetic Resonance MR Imaging (MRI) allows for the acquisition of subject-specific anatomical, functional as well as microstructural data of the heart. In the last years, several frameworks based on machine learning have been proposed to classify, label and segment MRI cardiac data. In a second step, personalised in-silico models are generated to analyse and enrich he available information from MRI. Conventionally, in these models, discretise the partial differential equations describing cardiac mechanics, electrophysiology and haemodynamics are solved relying on Finite Elements or Finite Volumes. In this project, we want the explore the challenges and requirements to extend these neural networks frameworks to the generation of personalised in-silico models. Particularly, we are interested in physics informed neural networks. The project will make use of our existing numerical tools based on open-source libraries for Finite Elements (FEniCS), Finite Volumes (openFOAM) and machine learning (Tensorflow, Pytorch, Keras).

Our investigation has multiple objectives and can accommodate for the student’s interests. Please contact us if you are interested and would like to know more. Potential projects include: - Modelling of cardiac mechanics, - Atlas based and reduced order modelling of cardiac mechanics, - Simulation of blood flow in arteries and left ventricles, - Reduced order modelling of blood flow, - Inverse problems in cardiac mechanics and fluid dynamics, - Generation of synthetic MRI images based on biophysical models and training of neural networks for functional evaluation of clinical images. **Requisites** Programming experience in Python is required. Familiarity with Tensorflow or Pythorch or Keras is a plus.

Supervisors: Dr. Stefano Buoso (buoso@biomed.ee.ethz.ch). To apply for this project please send a copy of your CV and transcripts of your Bachelor and Master studies.

Standardised performance assessment of image acquisition, reconstruction and processing methods is limited by the absence of images paired with ground truth reference values. Synthetic datasets could therefore be bridging this cap since they are associated with known parameters that could be used for computation of error metrics. However, these datasets are characterised by poor anatomical and functional variability which is usually limited to healthy conditions. With this project, we want to develop a generative model, based on deep-learning methods, for the generation of synthetic phantoms with sufficient anatomical variability to realistically represent population statistics. The project will be based on our recent work on generative models based and implemented in Pytorch.

The project aims at: - Building a generative model for left-ventricular 2D masks using convolutional variational autoencoders - Building a conditional generative model for the human torso tissues using convolutional variational autoencoders - Apply realistic tissue properties to the images using our texturizer network - Generating a dataset of representative cases including healthy and pathological left ventricles

Supervisors: Dr. Stefano Buoso (buoso@biomed.ee.ethz.ch). To apply for this project please send a copy of your CV and transcripts of your Bachelor and Master studies.