Studentenprojekte

Simulation-Based Design of Wearable Cardiac Bioimpedance Sensing

Wearable cardiac devices mainly measure electrical activity using ECG, while continuous information about mechanical cardiac function remains difficult to obtain. This Master’s thesis will investigate whether wearable thoracic bioimpedance measurements can recover relative cardiac volume variations. Using dynamic human torso models generated with TorsoGen, the student will develop finite-element and lead-field simulations, assess the observability of ventricular volume changes, and optimise electrode configurations under anatomical, physiological and measurement variability.

Schlagwörter

Wearable sensors, cardiac function, bioimpedance, electrical impedance tomography, finite-element modelling, inverse problems, digital twins, TorsoGen, electrode optimisation

Labels

Master Thesis

Description

Thoracic bioimpedance measurements apply small alternating currents through surface electrodes and measure the resulting voltage differences. Cardiac contraction, blood-volume redistribution and respiration modify the electrical properties of the thorax and therefore affect these measurements. Previous studies have demonstrated that cardiogenic impedance signals can be detected, but it remains unclear which cardiac components can be identified robustly across different human anatomies and realistic measurement conditions. This project will use TorsoGen to create a population of anatomically realistic human torso models with time-resolved cardiac geometries and known physiological ground truth. The student will: • Construct conductivity-aware torso models containing the ventricular blood pools, myocardium, lungs and major thoracic vessels. • Implement a three-dimensional finite-element forward model with surface electrodes and realistic current-injection and voltage-measurement protocols. • Derive phase-specific lead fields or sensitivity matrices to accelerate repeated simulations. • Quantify the separate contributions of LV and RV volume changes, myocardial deformation, heart motion and vascular blood-volume changes. • Investigate sensitivity to anatomical variability, respiration, tissue conductivity, electrode displacement, contact impedance and measurement noise. • Optimise electrode locations and measurement protocols for cardiac sensitivity, parameter separability and robustness. • Evaluate regularised inverse methods for estimating relative cardiac volume curves on previously unseen anatomies. The project is computational and will initially focus on establishing the physical observability and identifiability limits of wearable cardiac bioimpedance sensing. It will not assume that MRI-like anatomical images or absolute chamber volumes can be recovered without subject-specific calibration. Applicants should: • Be enrolled in a Master’s programme in biomedical engineering, electrical engineering, mechanical engineering, computational science, applied mathematics, physics or a closely related field. • Have strong programming skills in Python. • Have a solid background in numerical methods, linear algebra and scientific computing. • Be interested in biomedical modelling, wearable sensing and inverse problems. • Be able to work independently and document computational methods clearly. Experience with finite-element methods, electrical impedance tomography, signal processing, optimisation, machine learning, medical-image processing or anatomical modelling is advantageous but not mandatory.

Goal

Contact Details

Mehr Informationen

Dieses Projekt öffnen...

Publiziert seit: 2026-07-27

Organisation(en) Cardiovascular Magnetic Resonance

Host(s) Buoso Stefano

Themen Engineering and Technology

Deep Learning and Sequence-Informed Adaptive Filtering for ECG Gradient Artifact Reduction in MRI

Gradient switching during MRI creates severe noise on ECG signals, often causing false triggers and disrupting cardiac gating. This project focuses on developing sequence-informed deep learning filters to suppress these artifacts. Working with real 0.6T scanner data, you will build novel models and test their real-world impact using automated QRS detection.

Schlagwörter

Electrocardiogram (ECG), Magnetic Resonance Imaging (MRI), Gradient Artifacts, Signal Processing, Adaptive Filtering, Deep Learning, QRS Detection

Labels

Semester Project

Description

Reliable Electrocardiogram (ECG) acquisition during Magnetic Resonance Imaging (MRI) remains a major clinical challenge. Fast-switching magnetic field gradients (Gx, Gy, Gz) superimpose severe artifacts onto the electrophysiological signal. Because the shape of these gradient-induced artifacts often mimics the QRS-complex, they can cause misinterpretation during patient monitoring and lead to false triggering or gating of the MRI scanner [1,3]. To suppress these artifacts in real time, adaptive filtering techniques, such as Least Mean Square (LMS) or Kalman filters, are commonly used. Existent solutions integrate hardware-based Hall probes to measure magnetic gradients directly for adaptive LMS filtering [1,2,3]. However, magnetic field gradient information is also available from the MR pulse sequence description during scanning, providing a possible alternative to additional sensors and specialized hardware. Incorporating the known gradient waveforms into a sequence-informed, adaptive or deep learning-based filter design is therefore a promising approach to be investigated further. The student will implement a QRS-peak detection algorithm as a downstream analysis task to evaluate the developed models. The novel filters will be benchmarked against the integrated LMS filter and offline reference methods such as Wiener-Hopf. Evaluation metrics will include the noise reduction as well as the downstream accuracy of the QRS-detector for reliable MRI gating.[4] The project offers hands-on experience in biological signal processing, data acquisition, and the application of classical and deep-learning-based filtering techniques to real-world clinical hardware challenges. **Prerequisites:** - Strong programming skills (Python, PyTorch) - Solid background in signal processing and digital filtering - Solid understanding of machine learning/deep learning concepts - Interest in medical imaging and electrophysiology **References:** 1. Dos Reis JE, Soullié P, Oster J, et al. Reconstruction of the 12-lead ECG using a novel MR-compatible ECG sensor network. _Magn Reson Med_. 2019;82(5):1929-1945. doi:10.1002/mrm.27854 2. Abächerli R, Pasquier C, Odille F, Kraemer M, Schmid JJ, Felblinger J. Suppression of MR gradient artefacts on electrophysiological signals based on an adaptive real-time filter with LMS coefficient updates. _MAGMA_. 2005;18(1):41-50. doi:10.1007/s10334-004-0093-1 3. Guillou A, Sellal JM, Ménétré S, Petitmangin G, Felblinger J, Bonnemains L. Adaptive step size LMS improves ECG detection during MRI at 1.5 and 3 T. _MAGMA_. 2017;30(6):567-577. doi:10.1007/s10334-017-0638-8 4. Mahmud S, Chowdhury et al. Restoration of magnetohydrodynamic-corrupted 12-lead electrocardiogram to enhance cardiac monitoring during magnetic resonance imaging, *Engineering Applications of Artificial Intelligence,* Volume 133, Part E, 2024, doi:10.1016/j.engappai.2024.108483.

Goal

Contact Details

Mehr Informationen

Dieses Projekt öffnen...

Publiziert seit: 2026-07-24 , Frühester Start: 2026-09-14 , Spätestes Ende: 2027-04-30

Bewerbungen eingeschränkt auf ETH Zurich

Organisation(en) Cardiovascular Magnetic Resonance

Host(s) Assmann Cederic

Themen Medical and Health Sciences , Information, Computing and Communication Sciences , Engineering and Technology

Learning Neural Surrogate Models for Patient-Specific Cardiac Mechanics

Finite-element (FE) models provide the most physiologically realistic simulations of cardiac mechanics but require several minutes to hours for a single simulation, limiting their use for uncertainty quantification, inverse modelling and digital twins. This project aims to develop a graph-based neural surrogate that accurately approximates nonlinear biventricular FE simulations. Using a large database of more simulated hearts spanning anatomical variability, loading conditions and myocardial properties, the student will investigate modern graph neural networks and neural operators capable of predicting three-dimensional cardiac deformation directly from anatomy, fibre architecture, active tension and pressure fields. The resulting surrogate should reproduce FE-quality solutions several orders of magnitude faster while naturally supporting heterogeneous myocardial properties.

Schlagwörter

Scientific Machine Learning, Graph Neural Networks, Neural Operators, Computational Biomechanics, Cardiac Mechanics, Finite Elements, Digital Twins, Deep Learning, High-Performance Computing

Labels

Master Thesis , ETH Zurich (ETHZ)

Description

Goal

Contact Details

Mehr Informationen

Dieses Projekt öffnen...

Publiziert seit: 2026-07-24

Organisation(en) Cardiovascular Magnetic Resonance

Host(s) Buoso Stefano

Themen Mathematical Sciences , Information, Computing and Communication Sciences , Engineering and Technology

(Joint) Segmentation and Registration for Quantitative Perfusion CMR

This project aims to develop and validate (joint) segmentation and registration methods for quantitative first-pass perfusion cardiac magnetic resonance (CMR). To derive the myocardial and blood-pool concentration–time curves required for tracer-kinetic model fitting, the dynamic image series — which are characterized by strong, rapid changes in contrast and few stable landmark features — must be segmented and corrected for respiratory motion. The student will investigate howe state-of-the-art segmentation and motion compensation techniques can be applied, optimized and combined to improve robustness and workflow of myocardial blood flow (MBF) quantification.

Schlagwörter

cardiac magnetic resonance imaging, quantitative perfusion, first-pass perfusion, image registration, motion correction, image segmentation, deep learning, tracer-kinetic modeling

Labels

Semester Project , Master Thesis

PLEASE LOG IN TO SEE DESCRIPTION

Mehr Informationen

Dieses Projekt öffnen...

Publiziert seit: 2026-06-29 , Frühester Start: 2026-09-14 , Spätestes Ende: 2027-04-30

Bewerbungen eingeschränkt auf ETH Zurich , EPFL - Ecole Polytechnique Fédérale de Lausanne , University of Zurich , Paul Scherrer Institute

Organisation(en) Cardiovascular Magnetic Resonance

Host(s) Fütterer Maximilian

Themen Medical and Health Sciences , Information, Computing and Communication Sciences , Engineering and Technology , Physics

Predicting Cardiomyopathy Genotypes from Cardiac MRI Scar Patterns Using Deep Learning

This project aims to develop machine learning methods for predicting selected cardiomyopathy-associated genetic variants from cardiac magnetic resonance (CMR) images. Using late gadolinium enhancement (LGE) imaging and myocardial scar segmentations, the student will investigate whether imaging-derived scar patterns can be used to identify the underlying genetic cause of disease.

Schlagwörter

cardiac magnetic resonance imaging, medical imaging, machine learning, deep learning, neural networks, image classification, medical image analysis

Labels

Semester Project , Master Thesis

Description

LGE CMR imaging is the clinical gold standard for visualizing myocardial fibrosis. The distribution and extent of scar provide important diagnostic and prognostic information in patients with cardiomyopathy. Cardiomyopathies are a heterogeneous group of myocardial diseases, many of which have a genetic basis. Pathogenic variants in genes such as _TTN_, _DSP_, and _MYH7_ are associated with distinct cardiomyopathy phenotypes, including dilated and hypertrophic forms [1-5]. Emerging evidence suggests that different genetic variants may be associated with characteristic scar patterns visible on LGE imaging. However, the extent to which these patterns can be used to identify genotypes remains unclear. In this project, the student will investigate genotype-phenotype relationships using a dataset of LGE CMR images, scar segmentations, and genetic information. After reviewing the relevant literature, the student will develop and evaluate machine learning models to predict the presence of selected disease-causing genetic variants from imaging data. Potential approaches include convolutional neural networks (CNNs), graph neural networks (GNNs), autoencoders, or other deep learning architectures. The project offers hands-on experience in medical image analysis, machine learning, and cardiovascular imaging. **Prerequisites:** - Strong interest in machine learning and medical data analysis - Strong Python programming skills - Familiarity with deep learning frameworks (preferably PyTorch) - Basic knowledge of image processing **References:** [1] de Frutos F, Ochoa JP, Fernández AI, et al. Late gadolinium enhancement distribution patterns in non-ischaemic dilated cardiomyopathy: genotype-phenotype correlation. Eur Heart J Cardiovasc Imaging. 2023;25(1):75-85. doi:10.1093/ehjci/jead184 [2] Castrichini M, De Luca A, De Angelis G, et al. Magnetic resonance imaging characterization and clinical outcomes of dilated and arrhythmogenic left ventricular cardiomyopathies. J Am Coll Cardiol. 2024;83(19):1841-1851. doi:10.1016/j.jacc.2024.02.041 [3] Akhtar MM, Lorenzini M, Cicerchia M, Ochoa JP, Morris Hey T, Sabater Molina M, et al. Clinical phenotypes and prognosis of dilated cardiomyopathy caused by truncating variants in the TTN gene. Circ Heart Fail. 2020;13(10):e006832. doi:10.1161/CIRCHEARTFAILURE.119.006832 [4] Smith ED, Lakdawala NK, Papoutsidakis N, Aubert G, Mazzanti A, McCanta AC, et al. Desmoplakin cardiomyopathy, a fibrotic and inflammatory form of cardiomyopathy distinct from typical dilated or arrhythmogenic right ventricular cardiomyopathy. Circulation. 2020;141(23):1872-1884. doi:10.1161/CIRCULATIONAHA.119.044934 [5] Domínguez F, Cuenca S, Bilińska Z, Toro R, Villard E, Barriales-Villa R, et al. Dilated cardiomyopathy due to BLC2-associated athanogene 3 (BAG3) mutations. J Am Coll Cardiol. 2018;72(20):2471-2481. doi:10.1016/j.jacc.2018.08.2181

Contact Details

Mehr Informationen

Dieses Projekt öffnen...

Publiziert seit: 2026-06-25 , Frühester Start: 2026-09-14 , Spätestes Ende: 2027-06-30

Organisation(en) Cardiovascular Magnetic Resonance

Host(s) Margolis Isabel

Themen Medical and Health Sciences , Information, Computing and Communication Sciences , Engineering and Technology