COLIBRI Focus Workshop on Computational Medicine

Colibri Focus Workshop 2025 on Computational Medicine

Program

Day 1

9h00 opening 

Sandra Holasek (Styrian State Parliament), Klemens Fellner (Dean of the Faculty of Natural Sciences, University of Graz), Gundolf Haase (Co-Speaker (COLIBRI), Thomas Pock (Co-Director BioTechMed-Graz), Gunther Leobacher (Head of Institute of Mathematics and Scientific Computing University of Graz) 

Session 1: Cardiac Modelling

• 9h45 Keynote: Christoph Augustin: Physiologically Accurate Electromechanical Models of the Heart and Circulation for Clinical Applications 

• 10h15 Contributed Talks

  • Ester Bergantin: The rhythm of our heart: a cardiovascular model of dyssynchrony

  • Francesco Regazzoni: Improving computational efficiency in cardiac simulation: stress-free configuration recovery and multiphysics coupling

• 10h45 Coffee break 

• 11h30 Contributed Talks

  • Ludovica Cicci: Linking Electrophysiology Parameters and Simulated 12-lead ECGs to Enable Patient-Specific Model Calibration

  • Simone Pezzuto: Cardiac Digital Twin from surface ECGs: Insights into Identifiability

  • Karli Gillette: Generation of cardiac digital twins of whole-heart electrophysiology under normal sinus rhythm

  • Argyrios Petras: Towards in-silico cardiac ablation procedures on virtual patients

 

12h30 Lunch break 

Session 2: Machine learning, image processing & scientific computing

• 14h00 Keynote: Simone Saitta: Deep learning approaches for image analysis in interventional cardiology applications 

• 14h30 Contributed Talks 

  • Matthias Höfler: Estimating active stress in cardiac biomechanical models based on physics-informed neural networks

  • Stefano Pagani: Patient-specific electrophysiological models and machine learning methods for the numerical assessment of proarrhythmic propensity

  • Bruno Viti: Gaussian Process Emulators for Few-Shot Segmentation in Cardiac MRI

• 15h15 Coffee break 

• 16h00 Contributed Talks
  • Giulia Merlini: Numerical modelling of dynamic elastographic measurements. Application to the cornea
  • Francesco Romor: Registration-based data assimilation from medical images
  • Francesco Mantegazza: Data-Driven Cardiac Imaging Reconstruction: A PBDW Framework for Cardiac Mechanics

16h45 Break 

17h30 Public Lecture – Gernot Plank (Medical University of Graz): Digital Medicine - Medicine of the Future

18h45 Networking event

 

Day 2

Session 3: Brain imaging & modelling

• 9h00 Keynote: Paola Antonietti: Mathematical and numerical modeling of neurodegenerative diseases 

• 9h30 Contributed Talks 

  • Ivan Fumagalli: Computational modeling of brain multi-physics flows by polytopal methods 

  • Mattia Corti: Polytopal discontinuous Galerkin methods in Alzheimer's disease 

  • Johannes Haubner: Image Registration Using Optimal Control of a Linear Hyperbolic Transport Equation 

• 10h15 Poster pitches (2 min. / poster) 

• 10h45 Poster presentation & Coffee break 

Session 4: Industry session

• 11h45: Contributed Talks 

  • NumeriCor GmbH: State of the Art in Cardiac Electrophysiology Simulation

  • Computational Life: Multiscale computational modeling of cardiovascular system and drug interactions through cloud-based simulation

  • Know Center GmbH: Physics-Informed Machine Learning and Hybrid Modeling for Simulating Sustainable Systems

 

12h30 Lunch break 

Session 5: Blood flow modeling

• 14h00 Keynote: Lucas Müller: Assessment of cerebral reperfusion during normothermic regional perfusion for organ transplant using a closed/loop anatomically detailed arterio-venous blood flow model 

• 14h30 Contributed Talks 

  • Bianca Maria Laudenzi: In-silico study to predict in-hospital indicators from wearable-derived signals for cardiovascular and cardiorespiratory disease monitoring 

  • Bernhard Hametner: Mathematical Models for Arterial Pulse Wave Analysis: From Device-based Solutions to Virtual Twins 

  • Lukas Obermeier: Intracardiac Hemodynamics before and after Surgical Ventricular Restoration using image-based CFD 

• 15h15 Coffee break 

• 16h00 Contributed Talks 

  • Caterina Dalmaso: Cardiopulmonary mechanical interactions. Insights from an anatomically detailed arterial-venous network model 

  • Francesca Renzi: Image-based computational hemodynamics in the right heart
     

16h30 Closing session 

Abstracts

Public lecture

Plank, Gernot: Digital Medicine - Medicine of the Future
Digital medicine is an interdisciplinary field that uses advanced digital technologies such as artificial intelligence (AI), machine learning (ML), big data analytics, and mobile health platforms to improve patient care. These digital technologies hold high promise to revolutionize medicine by making medical treatments more precise, efficient personalized and accessible, but they also pose various risks such as privacy concerns, diagnostic and therapeutic malpractice and related liability issues. This talk aims to provide a big picture view on these important trends, with a more detailed glimpse into the specific application of using “digital twins”- these are simulation-based virtual replicas of patients - for stratification and therapy planning.

 

Keynotes talks

Antonietti, Paola F.: Mathematical and numerical modeling of neurodegenerative diseases
Neurodegenerative diseases (NDs) are complex disorders that primarily affect neurons in the brain and nervous system, leading to progressive deterioration and loss of function over time. A common pathological hallmark among various NDs is the accumulation of disease-specific misfolded proteins, such as amyloid-beta and tau in Alzheimer's disease and alpha-synuclein in Parkinson's disease. In this talk, we discuss the mathematical and numerical modelling of misfolded protein dynamics in neurodegenerative diseases, employing mathematical models of increasing complexity. To tame complexity, we propose and analyse high-order discontinuous Galerkin methods on polytopal grids for numerical discretisation, and discuss suitable machine learning enhancement and acceleration techniques. In the second part of the talk, we discuss the computational modelling of mechanisms for waste removal (clearance) from the brain, which plays a crucial role in the onset and progression of NDs. Patient-specific numerical simulations based on clinical data are also presented.

Augustin, Christoph: Physiologically Accurate Electromechanical Models of the Heart and Circulation for Clinical Applications
Image-based computational models of cardiac electromechanics are powerful tools for understanding the mechanisms underlying physiological and pathological cardiac function. However, their application in clinical practice has been limited. Advancing their utility in the clinic requires addressing key methodological challenges. First, computational efficiency and robustness are essential to enable model personalization and simulate prolonged observation periods under diverse conditions. Second, achieving physiological completeness is necessary to predict therapy-relevant mechanisms, moving beyond merely replicating observed data.
                                                                               
This talk presents a universal cardiac electromechanic modeling framework that incorporates advanced coupling methods across relevant physical processes. Recent progress in model personalization and validation is also discussed.
                                                                               
The framework demonstrates computational efficiency, robustness, and accuracy, effectively replicating physiological behaviors, such as responses to changes in loading conditions and contractility. Additionally, its potential applicability to clinically relevant problems is illustrated, providing a pathway toward broader clinical adoption.

Müller, Lucas Omar: Assessment of cerebral reperfusion during normothermic regional perfusion for organ transplant using a closed/loop anatomically detailed arterio-venous blood flow model
As the application presented in this talk will demonstrate, a global view of the cardiovascular system is sometimes necessary, even if one is interested in something organ-specific, like cerebral circulation.

In this presentation, we will briefly describe an anatomically detailed closed-loop model of the cardiovascular system (Müller et al., 2023) featuring a one-dimensional description of systemic arteries and veins and the interaction of cerebral blood flow with the craniospinal cavity. We will also assess the model output's physiological correctness concerning well-known haemodynamic indexes and waveform features.

Next, we will present an application in which a closed-loop description of the cardiovascular system is essential. In particular, we will show current work on potential brain reperfusion during normothermic regional perfusion (NRP). Normothermic Regional Perfusion (NRP) is a rapidly growing organ recovery technique. NRP doubles liver utilisation (29% v 66%) and improves 1-year eGFR by 6.3 ml/min/1.73m2 (Oniscu et al., 2023). However, there remain concerns that NRP could result in meaningful cerebral perfusion.
We model NRP in a male subject (height 170 cm; weight 56 kg) as follows: 1) Reference model (living); 2) NRP (distal aortic and caval cannulation, thoracic clamp and distal ligation of cava and aorta); 3) NRP with aortic arch drainage to reservoir (0mmHg). Results will be discussed in detail, providing insight into best NRP practices for brain reperfusion risk minimisation.

Müller et al., 2023. Front Physiol. doi: 10.3389/fphys.2023.1162391.
Oniscu et al., 2023. Transplantation. 107(2):438–448. doi: 10.1097/TP.0000000000004280.

Saitta, Simone: Deep learning approaches for image analysis in interventional cardiology applications
Deep learning has revolutionized the field of medical image analysis, offering unprecedented opportunities to enhance precision and efficiency in interventional cardiology. This keynote will explore how advanced deep learning methodologies can address critical challenges in this domain, spanning from pre-procedural planning to real-time intraoperative decision support. By leveraging the power of data-driven models, we can extract clinically meaningful insights, optimize workflows, and improve patient outcomes.

Key highlights will include an overview of approaches for analyzing computed tomography imaging data, enabling accurate characterization of cardiac structures and vascular dynamics. Examples will illustrate how innovative techniques, such as semi-supervised learning, surrogate modeling, and comprehensive segmentation algorithms, are transforming the evaluation of valvular and coronary and valvular pathologies. Additionally, we will discuss the integration of artificial intelligence into procedural pipelines, emphasizing its potential to enhance both diagnostic and therapeutic precision.

The lecture will also address practical considerations, such as handling variability in intracoronary optical coherence tomography image quality, integrating AI tools into clinical practice, and overcoming data scarcity. These insights aim to bridge the gap between technical innovation and clinical application, showcasing how deep learning can empower cardiologists to deliver personalized, data-driven care in interventional settings.

 

Presentations

Bergantin, Ester: The rhythm of our heart: a cardiovascular model of dyssynchrony
Under physiological conditions, the heart’s ventricles experience rapid electrical activation through the cardiac conduction system, resulting in nearly synchronous mechanical contraction. Cardiac electrical asynchrony occurs due to conduction disorders such as left bundle-branch block (LBBB). Electrically asynchronous activation leads to heterogeneous myocardial contraction, potentially causing reduced pump function. Cardiac resynchronization therapy has emerged as an important therapy to improve pump function in patients with LBBB. This therapy aims to restore ventricular electrical synchrony, by pacing both ventricles either simultaneously or sequentially. In this context, computational cardiovascular modelling is increasingly recognized and used as a valuable tool for investigating pathological conditions and evaluating treatment strategies.
The main goals of our research involve modelling the effects of dyssynchrony on cardiac function and the vascular system, as well as investigating the effect of different pacing strategies. To this end, we used a cardiac electrophysiology model to calculate activation maps, which were then used as inputs for a lumped mechanical and hemodynamic model to predict the effects of dyssynchrony on regional tissue mechanics and global cardiac pump function. 
Using patient-specific cardiac geometries, electrical activation maps were calculated using an Eikonal model on a 3D biventricular geometry. The Purkinje network was identified using a probabilistic approach based on non-invasive clinical data such as the standard ECG. The activation times were then averaged within the AHA regions and used as inputs in CircAdapt, a cardiovascular model capable of describing heterogeneous mechanical activation. The CircAdapt model was calibrated to match MRI-derived variables such as end-systolic and end-diastolic volumes. We then compared simulated strain, regional myocardial work, and hemodynamic function for each pacing strategy.

Christian Contarino: Multiscale computational modeling of cardiovascular system and drug interactions through cloud-based simulation
The Computational Life (CL) platform is a multiscale computational modeling framework designed to simulate the cardiovascular system, whole-body physiology, mechanical circulatory support (MCS) devices, and drug interactions. Integrating a whole-body physiology model based on the open-source Pulse framework by Kitware, the platform combines one-dimensional (1D) representations of major arteries and veins with zero-dimensional (0D) lumped parameter models covering cardiac chambers, microcirculation, and organ systems. Local time-stepping methods optimize the computational performance of the 1D components, enabling accurate and efficient simulation of dynamic hemodynamic responses.
To accelerate analysis, the platform incorporates machine learning techniques, such as Gaussian Process Emulators (GPEs), for rapid exploration of parameter spaces and patient-specific simulations. This approach supports global sensitivity analysis and improves the understanding of treatment variability. Cloud-based deployment ensures scalability, flexibility, and seamless integration of complex physiological, pharmacological, and device interactions, advancing personalized medicine and clinical decision-making.

Cicci, Ludovica: Linking Electrophysiology Parameters and Simulated 12-lead ECGs to Enable Patient-Specific Model Calibration
Computational cardiac models are a powerful tool for investigating in-silico electrophysiological changes at cellular and organ levels. While patient-specific geometries can be reconstructed from medical images, other model parameters must be personalised to represent the electrophysiological characteristics of an individual.
This study investigates the interactions between cellular-scale parameters and organ-scale measurements to identify which model inputs can be inferred from non-invasive clinical data, such as the 12-lead ECG.
Using N=5 whole-torso meshes reconstructed from MRI data, we simulated ventricular depolarisation and repolarisation with a reaction-eikonal model and a fascicular-based representation of the His-Purkinje system. We rely on a linear screening method to rank the input parameters – including ion channel densities, conductivity gradients, and cardiac conduction system anatomy -- by their influence on scalar ECG-related quantities. We repeated the analysis after adapting the geometries to represent hypertrophic and dilated cardiomyopathy patients. Inter- and intra-patient comparisons assessed the relative contributions of anatomical and disease-specific changes.
The results demonstrated consistent rankings of the most important parameters across geometries, as well as the least influential ones. The latter can be arbitrarily fixed in their range of variability without affecting the ECG signal morphology. Moderately influential inputs were instead notably affected by anatomical variability, particularly in the diseased cases. These findings highlight the potential of leveraging standard 12-lead ECG data to estimate cellular and tissue properties, facilitating model personalisation.

Corti, Mattia: Polytopal discontinuous Galerkin methods in Alzheimer's disease
Neurodegenerative diseases have a significant global impact, affecting millions of individuals worldwide. Some of them, known as proteinopathies (for example, Alzheimer's and Parkinson's diseases), are characterized by the accumulation and propagation of toxic proteins known as prions. Mathematical models of prion dynamics play a crucial role in understanding disease progression. Several models have been proposed to describe the misfolding process, with various levels of detail. In this context, we introduce a discontinuous Galerkin method on polytopal grids (PolyDG) for the semi-discrete approximation of different nonlinear population dynamics models, such as the Fisher-Kolmogorov and the heterodimer model.

Calibrating model parameters is vital for accurate simulations. We perform a sensitivity analysis of these models to assess how equilibrium protein concentration values influence solution patterns, using biological measurements from the brain cortex of Alzheimer's patients and healthy controls for the tau protein and amyloid-beta. Finally, using the sensitivity analysis results, we discuss the applicability of both models in the correct simulation of the spreading of the two proteins.

Dalmaso, Caterina: Cardiopulmonary Mechanical Interactions: Insights from an Anatomically Detailed Arterial-Venous Network Model
The cardiovascular and respiratory systems guarantee optimal organ and tissue perfusion by continuously adjusting their functions in response to stimuli. 
Numerous models have been proposed since the 1970s to describe the cardiovascular and respiratory systems at rest and during exercise, to improve the understanding of the physiological mechanisms that regulate them and their interactions. Most of these models are 0D and focus primarily on the cardiovascular system and cardiovascular control, or on cardiopulmonary interactions. Additionally, numerous 1D models have been proposed to study hemodynamics in large arteries and veins, and to assess the impact of stenoses and other pathological conditions. Nonetheless, no models are currently available in the literature that couple a 1D description of the cardiovascular system to a 0D description of pulmonary mechanics. As a consequence, we propose here a 1D-0D model that couples a 0D description of lung mechanics to the closed-loop Anatomically-Detailed Arterial-Venous Network (ADAVN) model, which includes over 2000 vascular segments.
After showing that we can satisfactorily reproduce a set of cardiovascular indices of interest observed in young healthy males at rest, we use our model to assess the impact of respiration on cardiac performance and on the periodicity and average values of pressure and flow waveforms in different vascular districts. In particular, we conduct a spectral analysis to characterize the main determinants of the oscillatory patterns in pressure and flow rate tracings, and a nonlinear wave-intensity analysis. Consistently with physiological literature, we observe that respiratory mechanics affects waveforms in arterial segments mostly in terms of average values, while in veins it has a marked impact also on wave periodicity. Finally, we assess the sensitivity of model predictions to variations in model parameters through a local sensitivity analysis, both in the presence and absence of respiration.

Fumagalli, Ivan: Computational modeling of brain multi-physics flows by polytopal methods
The dynamics of cerebrospinal fluid (CSF) and blood flow are critical to understanding brain physiology, particularly in their roles in clearing waste products, transporting nutrients and ions, regulating intracranial pressure, and sustaining neuronal health. The coupling of CSF flow with blood pulsation, respiration, and flow-tissue interaction occurs within the brain's geometry. Disruptions in these fluid dynamics and circulatory pathways are closely associated with pathologies such as hydrocephalus, neurodegenerative diseases and ischemic conditions, highlighting their clinical significance.
To investigate these processes, we develop a computational framework that integrates Multiple-network Poro-Elasticity (MPE) equations for tissue perfusion with Stokes equations governing CSF flow in the brain ventricles. A polytopal method is introduced for the efficient discretization of the brain’s complex geometry, enabling the study of pulsatility, interface conditions, and fluid inertia in both 2D and 3D geometries. Numerical experiments demonstrate the computational efficiency of the method in dealing with detailed brain geometries. Furthermore, a posteriori analysis of a fluid-poroelastic structure interaction model offers insights into mesh adaptivity strategies to further enhance computational efficiency for patient-specific simulations.
Acknowledgments: This work has been partially supported by ICSC-Centro Nazionale di Ricerca in High Performance Computing, Big Data, and Quantum Computing funded by European Union-NextGenerationEU. It is also part of the activities of “Dipartimento di Eccellenza 2023-2027”, Dipartimento di Matematica, Politecnico di Milano. All contributors are members of GNCS-INdAM and we acknowledge the support of the GNCS project CUP E53C23001670001.

Geiger, Bernhard : Physics-Informed Machine Learning and Hybrid Modeling for Simulating Sustainable Systems
Physics-informed machine learning and hybrid modeling are paradigms that recently emerged from the wish to combine the rigor and precision of first-principles models with the adaptability and efficiency of data-driven methods. After situating physics-informed machine learning and hybrid modeling in the wider research landscape, we will present two applications in detail: The first application concerns robust optimization of multi-stage forging processes. Based on Gaussian process surrogate models, a Bayesian optimization scheme is presented, ensuring that the forging process is not only accurate on average, but that the forged part also stays within specifications with high probability. The second application concerns solving partial differential equations using physics-informed neural networks. We show how to overcome well-known training problems for a reaction-diffusion system and present a neural model that solves this system for a large range of reaction rate coefficients.

Gillette, Karli: Generation of cardiac digital twins of whole-heart electrophysiology under normal sinus rhythm
Personalized medicine using cardiac digital twins of cardiac electrophysiology has shown great promise for enhancing diagnostics and ther- apy planning for cardiac arrhythmias. Whole-heart cardiac digital twins, how- ever, are challenging to personalize in terms of both anatomy and function. We present a novel computational pipeline for generating single snapshots of car- diac digital twins of whole-heart electrophysiology based on non-invasive clinical imaging and 12 lead electrocardiogram (ECG) data.
Our computational pipeline produces anatomically highly detailed heart-torso models of patient hearts from clinical cardiac magnetic resonance images and calibrates their electrophysiological model properties to replicate the measured 12 lead ECGs. Efficient modeling pipelines in the atria and ven- tricles are deployed with modifications for atrioventricular entities. We utilize a novel optimization approach termed Geodesic-BP to infer ventricular activation during normal sinus rhythm based on the QRS complex. T-wave morphology is based on ventricular repolarization gradients related to activation, and the P- wave depends on modified atrial electrophysiology from generic parameters. The method is demonstrated for two healthy subjects under normal sinus rhythm.
The novel computational pipeline can generate cardiac digital twins of whole-heart electrophysiology at scale within clinical time frames under 10 hours. Segmentation and optimization of the ventricular activation constituted the highest temporal costs. For the two subjects, simulated 12 lead ECGs are high fidelity, especially in the QRS complex.
Our robust and non-invasive computational pipeline facilitates the generation of cardiac digital twins based on non-invasive clinical data. The method is scalable for additional subjects. In future work, we aim to generate time-integrated cardiac digital twins and apply the cardiac digital twins across various cardiac arrhythmias. Depending on the application, a detailed His-Purkinje system must be incorporated, and further optimization of atrial parameters may be needed.

Hametner, Bernhard: Mathematical Models for Arterial Pulse Wave Analysis: From Device-based Solutions to Virtual Twins
Arterial pulse wave analysis (PWA) is a key method for understanding cardiovascular health. It offers insights into vascular function but also to related diseases of other organs such as the heart, brain, or kidneys. The shape of the arterial pulse wave is influenced by factors such as arterial stiffness, wave reflection, blood pressure, and ventricular function. Regulatory mechanisms, including neural and hormonal controls, modulate wave propagation and reflection, shaping the overall waveform.
For a physiological interpretation of the measured waveform, mathematical models can be applied for PWA. This involves both physics-based and data-driven models. Several models are incorporated in the ARCSolver algorithms for PWA from the AIT Austrian Institute of Technology, intended to be used together with specific medical devices. Model-based analyses provide markers of arterial stiffness, wave reflection and ventricular function. In clinical studies, the obtained parameters have been evaluated against gold-standard methods and the predictive value of central blood pressure, pulse wave velocity and wave reflection indices was shown in several cohorts. 
Currently, the research field is working on virtual twins for the cardiovascular system. AIT is involved in the project VITAL (vital-horizoneurope.eu), which aims to create digital twins to allow personalized treatment of cardiovascular diseases. This solution involves the integration of data from several sensors, such as a wearable capturing arterial pulse waves, and image-based techniques. Besides the complex interplay of existing and newly developed models, the patient-specific calibration of the models will be a main challenge. 
Thus, while the concept of virtual twins for cardiovascular assessment is very appealing, personalised risk prediction based on arterial pulse wave measurements is currently mainly feasible with specific and problem-tailored methods.

Haubner, Johannes: Image Registration Using Optimal Control of a Linear Hyperbolic Transport Equation
Image registration is crucial in many imaging applications such as medical imaging or computer vision. The goal of finding a suitable transformation between two images poses similar restrictions and regularity requirements on the set of admissible transformations as shape optimization problems. In the scope of this talk, we build on an approach that models image registration as an optimization problem that is constrained by a linear hyperbolic transport equation. We use a higher-order discontinuous Galerkin finite element method for discretization and motivate the numerical upwind scheme and its limitations from the continuous weak space-time formulation of the transport equation. Moreover, we build on recent theoretical results to model the optimization problem. To discuss the potential of the proposed algorithm, we apply it to patient specific brain mesh generation from magnetic resonance (MR) images. This can be a time consuming task and require manual corrections, e.g., for meshing the ventricular system or defining subdomains. The idea is to use the registration of an input MR image to a respective target in order to obtain a new mesh from a high-quality template mesh.

Höfler, Matthias: Estimating active stress in cardiac biomechanical models based on physics-informed neural networks
Active stress models in cardiac biomechanics account for the mechanical deformation caused by muscle activity, thus providing a link between the electrophysiological and mechanical properties of the tissue. The accurate assessment of active stress parameters is fundamental for a precise understanding of myocardial function but remains difficult to achieve in a clinical setting, especially when only displacement and strain data from medical imaging modalities are available.  This poster investigates, through an in-silico study, the application of physics-informed neural networks for inferring active contractility parameters from these types of imaging data. This approach opens a new pathway to enable the detection and characterisation of tissue inhomogeneities, such as fibrotic regions, and could significantly impact the diagnosis, treatment planning, and management of heart conditions associated with cardiac fibrosis.

Laudenzi, Bianca Maria: In-silico study to predict in-hospital indicators from wearable-derived signals for cardiovascular and cardiorespiratory disease monitoring
Cardiovascular disease (CVD) is the leading cause of death worldwide. The construction of health digital twins for patient monitoring is becoming a fundamental tool to reduce invasive procedures, minimize patient hospitalization, design clinical trials and personalize therapies. 
The aim of this study is to investigate the feasibility of machine learning-based bio-signals analysis for the monitoring of patients with CVD, using a database of in-silico patient data.

In particular, a population of virtual subjects, both healthy and with CVD, is created using a comprehensive zero-dimensional global closed-loop model, comprising major elements characterizing cardio-respiratory functions, such as the cardiovascular system, the lungs mechanics, the gas exchange and transport and the main short term regulatory mechanism.

Our database allows us to train Gaussian process regressors, informed by wearable-acquired data, to predict variables normally acquired only during in-hospital exams. Simulated results provides theoretical assessment of accuracy for predictions of stroke volume, cardiac output, ejection fraction, central venous pressure, and partial pressure of oxygen and carbon dioxide, using wearable-derived indices, i.e. systolic/diastolic blood pressure and heart rate.

Mantegazza, Francesco: Data-Driven Cardiac Imaging Reconstruction: A PBDW Framework for Cardiac Mechanics
Personalizing cardiac modeling through data assimilation for state estimation is vital, yet it often demands significant computational resources. Traditional variational methods encounter challenges due to their high computational overhead, limiting their practical use. In response, reduced-order models (ROMs) [1] have emerged as efficient substitutes for full-order models. Moreover, the concept of state estimation can be reframed as an optimal recovery problem, offering a non-intrusive data-driven perspective that complements traditional model-based approaches.

Within this context, we introduce a computational framework aimed at enhancing the image quality of cardiac magnetic resonance imaging (MRI) through advanced mathematical modeling and state estimation. Our method integrates the Parameterized-Background Data-Weak (PBDW) [2] technique with an innovative sensor selection approach, addressing the critical need to maximize the value extracted from medical images.

Our methodology employs a computationally efficient reduced basis method and a novel mini-batch worst-case orthogonal matching pursuit (wOMP) algorithm for optimal sensor selection. The implementation processes MRI voxel data using carefully designed measurement functionals representing volumetric averages, aligning with standard clinical imaging protocols.

To generate synthetic data, we use our in-house finite element solver, CARPentry, in order to simulate nonlinear orthotropic mechanics on an ellipsoidal geometry representing a left ventricle. Our approach demonstrates robust reconstruction of high-resolution 3D displacement fields from limited-resolution measurements, assessed through relative H^1 norm comparisons. In particular, the employed measurements are noisy as they are enriched with Gaussian noise.

[1] A. Quarteroni, A. Manzoni, et al.  Reduced Basis Methods for Partial Differential Equations, Springer International Publishing, 2016.
[2] T. Taddei. Model order reduction methods for data assimilation: state estimation and structural health monitoring, MIT, 2016.

Merlini Giulia: Numerical modelling of dynamic elastographic measurements. Application to the cornea.
This work concerns the simulation of the shear waves propagation in the cornea. The underlying application is the detection, through elastography techniques, of pathologies that are characterized by changes in the mechanical properties of the tissues. Due to its collagen network and its high water content, the cornea is modeled as a non-linear anisotropic hyperelastic nearly incompressible tissue. Moreover, the cornea is subject to a stationary pressure applied on the internal surface. As a first step, we define the biomechanical model of the cornea, comprised of the aforementioned characteristics, with the anisotropic behavior described by a micro-sphere model. By linearization, we derive the governing equation of the linear elastic wave propagation superimposed on a finite deformation of the cornea. The non-linear elasto-static problem is solved through a preconditioned gradient descent algorithm. Then, in order to compute the shear wave propagation resulting from an external excitation, we use mass-lumped spectral finite elements, combined with an inf-sup stable mixed formulation to deal with the nearly incompressibility, and finally, an energy-preserving explicit time discretization. Furthermore, we address the case where the assumption of linear wave propagation may not be sufficient. We propose the application of a novel numerical method for solving nonlinear elastodynamic problems. This fully explicit, three-step method ensures stability through a posteriori energy criterion. The nonlinear static model of the cornea is then coupled with the nonlinear wave propagation model.

Neic, Aurel: State of the Art in Cardiac Electrophysiology Simulation
NumeriCor is a young company developing high-performance cardiac electrophysiology simulation tools. Cardiac device manufacturers use our software to predict cardiac response to therapies and for generating virtual input for their algorithms. We will present the problem setting in electrophysiology simulation as well as explore specific applications, such as defibrillator performance prediction and cardiac rhythm management.

Obermeier, Lukas: Intracardiac Hemodynamics before and after Surgical Ventricular Restoration using image-based CFD
Introduction: Surgical ventricular restoration (SVR) excludes scarred myocardium after myocardial infarction to restore shape and contractility of dilated, aneurysmal left ventricles (LVs). In this study, virtual digital twins of the LV are used to study the hemodynamic impact of successful SVR.
Methods: The digital replicas were built on pre-operative and post-operative cardiac computed tomography data of nine patients (3 females, 60±13 years) who underwent successful SVR (significant New York Heart Association class improvement). The computational framework was used to calculate LV morphology, dynamics, as well as intracardiac hemodynamics using image-based computational fluid dynamics (CFD).
Results: SVR successfully reduced the LV volumes by 94.0±61.5 (end-diastole) and 99.8±59.0 ml (end-systole). Morphological analysis showed restored myocardial wall thickness in aneurysmal regions (5.5±2.0 vs. 8.6±3.0 mm) and an increased end-diastolic sphericity (sphericity index 0.39±0.07 vs. 0.46±0.07). No distinct flow alterations could be linked thereto. CFD revealed a higher post-operative energetic level (diastolic maximum 9.8±7.5 vs. 16.6±9.1 mJ) and an improved global washout (29.5±9.7 vs. 10.3±6.4 % still present after five cycles), which could be closely linked to increases in volume-curve-derived diastolic energy gain and ejection fraction, respectively. Flow efficiency improved by means of an increased end-diastolic surface-averaged vortex strength (15.8[10.1;17.9] vs. 23.6[21.0;24.7] 1/s) and a decreased normed diastolic energy loss (15.5±4.1 vs. 12.0±3.1 %), both of which correlated to ventricular contractility.
Conclusion: The digital patient replicas facilitated a detailed analysis and showed favorable flow changes with successful SVR. It also demonstrated the ability of virtual digital twins to identify complex functional relations and derive simplified, model-based biomarkers that may support clinical diagnosis and treatment planning.

Pagani, Stefano: Patient-specific electrophysiological models and machine learning methods for the numerical assessment of proarrhythmic propensity
One of the major challenges in computational electrophysiology is to predict the risk of developing an arrhythmic event and to identify the potential areas involved. This requires the accurate personalization of electrophysiological models from clinical data and the development of machine learning approaches that enhance the identification of effective biomarkers to support clinical decision-making.

In this talk, we will first present a personalized electrophysiological model based on myocardial blood flow maps that has been developed to assess the proarrhythmic risk associated with acute myocardial ischemia [1]. Myocardial blood flow maps allow the construction of patient-specific parameterizations of the differential model informed by realistic distributions of less perfused regions. Subsequently, numerical simulations of the electrophysiological response to ectopic beats provide a quantitative assessment of the proarrhythmic risk. Our results highlight the impact of the electrical substrate on the induction and sustainment of re-entrant drivers.  

In addition, we will present three machine learning approaches based on deep anomaly detection to characterize electrogram signals associated with pro-arrhythmic substrates [2]. In an unsupervised manner, these methods identify an all-in-one electrophysiological indicator of electrical dysfunction capable of surrogating all the standard biomarkers.

[1] Corda, Pagani, Vergara. Influence of acute myocardial ischemia on arrhythmogenesis: a computational study, medRxiv 2024.11.20.24317476
[2] Bindini, Pagani, Bernardini, Grossi, Giomi, Frontera, Frasconi. All-in-one electrical atrial substrate indicators with deep anomaly detection, Biomedical Signal Processing and Control, 98, 2024.

Petras, Argyrios: Towards in-silico cardiac ablation procedures on virtual patients
Catheter ablation is a common treatment for severe cardiac arrhythmias. During the procedure, a catheter is typically inserted through the patient’s groin and guided into the cardiac chamber, where it inflicts targeted damage to arrhythmogenic tissue in order to restore normal sinus rhythm. Two primary ablation technologies are used: thermal-based techniques like radiofrequency ablation (RFA) and electroporation-based approaches such as pulsed field ablation (PFA). While these methods are generally safe and effective, further research is required to improve lesion durability and reduce procedural complications.

In this presentation, we introduce our computational framework that uses virtual patients (image reconstructed geometries) for simulating the ablation procedure. We present our multiphysics model for capturing the lesion evolution during RFA, and also our strategies for extending these models to support PFA, personalized medicine and in-silico clinical trials.

Pezzuto, Simone: Cardiac Digital Twin from surface ECGs: Insights into Identifiability
Digital twins for cardiac electrophysiology are an enabling technology for precision cardiology. Current forward models are advanced enough to simulate the cardiac electric activity under different pathophysiological conditions and accurately replicate clinical signals like torso electrocardiograms (ECGs). In this work, we address the challenge of matching subject-specific QRS complexes using anatomically accurate, physiologically grounded cardiac digital twins.
By fitting the initial conditions of a cardiac propagation model, our non-invasive method predicts activation patterns during sinus rhythm. 
For the first time, we demonstrate that distinct activation maps can generate identical surface ECGs. To address this non-uniqueness, we introduce a physiological prior based on the distribution of Purkinje-muscle junctions. Additionally, we develop a digital twin  ensemble for probabilistic inference of cardiac activation.
Our approach marks a significant advancement in the calibration of cardiac digital twins and enhances their credibility for clinical application.

Regazzoni, Francesco: Improving computational efficiency in cardiac simulation: stress-free configuration recovery and multiphysics coupling
One of the main obstacles toward clinical translation of cardiac computational models is their elevated computational burden, which is often not compatible with the constraints dictated by clinical practice. In this talk, we present some recent contributions aimed at accelerating the numerical approximation of cardiovascular models. Specifically, we present a formulation that significantly accelerates the simulation setup, allowing the stress-free configuration of the myocardium to be reconstructed from a deformed geometry. Our results show that this approach is more robust and efficient than the prevailing point-fixed scheme commonly employed. Next, we present a novel algorithm for transferring information between different meshes, which is particularly useful in allowing the use of different resolutions for mechanical and electrophysiological variables. The proposed algorithm allows for the use of independent meshes, ensuring the physical consistency of the transferred tensors through a combined use of Radial Basis Function (RBF) interpolation and Singular Value Decomposition (SVD).

Renzi, Francesca: Image-based computational hemodynamics in the right heart
Characterizing flow within the right heart (RH) is particularly challenging due to its complex geometries. However, gaining insight into RH fluid dynamics is of extreme diagnostic importance, given the high prevalence of acquired and congenital heart diseases with impaired RH function. 
In this proof-of-concept study, we propose a pipeline for patient-specific simulations of RH hemodynamics. We reconstruct the geometry and motion of the patient’s right atrium, ventricle, and pulmonary and tricuspid valves, from multi-series cine-MRI. For this purpose, we develop a novel and flexible reconstruction procedure that integrates patient-specific tricuspid valve dynamics into a computational model, enhancing the accuracy of our RH blood flow simulations. We apply this approach to study the hemodynamics in a healthy RH and a repaired Tetralogy of Fallot RH with severe pulmonary regurgitation, as well as to assess the hemodynamic changes induced by the pulmonary valve replacement intervention.
Modeling the entire RH enables us to understand the contribution of the superior and inferior vena cava inflows to the ventricular filling, as well as the impact of the impaired right atrial function on the ventricular diastole. To analyze the turbulent and transitional behaviour, we include the Large-Eddies Simulation sigma model in our computational framework, which reveals how the contribution of the smallest scales in the dissipation of the turbulent energy changes among health and disease.
Finally, we prove the reliability of the proposed modeling pipeline by comparing our numerical results with reference literature values and, for the diseased scenario, with available phase-contrast measurements at the pulmonary outflow tract.

Romor, Francesco: Registration-based data assimilation from medical images
Image-based, patient-specific modeling of hemodynamics can improve diagnostic capabilities and provide complementary insights to better understand the after-effects of treatments. However, computational fluid dynamics simulations remain relatively costly in a clinical context. Moreover, projection-based reduced-order models and purely data-driven surrogate models struggle due to the high geometric variability of biomedical datasets. A possible solution is shape registration: a reference template geometry is designed from a cohort of available geometries, which can then be diffeomorphically mapped onto it. This provides a natural encoding that can be exploited by machine learning architectures and, at the same time, a reference computational domain in which efficient dimension-reduction strategies can be performed (Guibert, Caiazzo et al. 2014). We compare state-of-the-art graph neural network models with recent data assimilation strategies (Galarce, Lombardi, Mula, 2022) for the prediction of physical quantities and clinically relevant biomarkers in the context of aortic coarctation.

Viti, Bruno: Gaussian Process Emulators for Few-Shot Segmentation in Cardiac MRI
Segmentation of cardiac magnetic resonance images (MRI) is crucial for the analysis and assessment of cardiac function, helping to diagnose and treat various cardiovascular diseases.
Most recent techniques rely on deep learning and usually require an extensive amount of labeled data.
To overcome this problem, few-shot learning has the capability of reducing data dependency on labeled data.
In this work, we introduce a new method that merges few-shot learning with a U-Net architecture and Gaussian Process Emulators (GPEs), enhancing data integration from a support set for improved performance.
GPEs are trained to learn the relation between the support images and the corresponding masks in latent space, facilitating the segmentation of unseen query images given only a small labeled support set at inference.
We test our model with the M&Ms-2 public dataset to assess its ability to segment the heart in cardiac magnetic resonance imaging from different orientations, and compare it with state-of-the-art unsupervised and few-shot methods.
Our architecture shows higher DICE coefficients compared to these methods, especially in the more challenging setups where the size of the support set is considerably small.

 

Poster presentations

Buonocunto, Melania: Influence of stretch-activated ion channels (SACs) on SQT1 and LQT2 electrophysiology: a cellular and tissue level in-silico study
Short and Long QT syndromes (SQTS, LQTS) are inherited cardiac channelopathies causing ventricular arrhythmia and sudden cardiac death. SQTS type 1 (SQT1) and LQTS type 2 (LQT2) involve gain and loss of function of the IKr current, resulting in shortened and lengthened QT intervals/action potential (AP) durations, respectively.
Recent studies emphasize the electro-mechanical nature of these diseases, suggesting that acute mechano-electric feedback (MEF) mechanisms may contribute to arrhythmogenesis. However, comprehensively characterizing MEF in relevant animal models is experimentally challenging. Here, we aim to characterize acute stretch arrhythmogenic effects on ventricular electrophysiology reflecting SQT1 and LQT2 conditions, using an in-silico model.
We used our prior in-silico model of human ventricular cardiomyocyte electrophysiology incorporating MEF via 3 stretch-activated ion channels (SACs), to simulate AP duration differences observed in experimental data from SQT1 and LQT2 transgenic rabbit cardiomyocytes versus wild-type.
To simulate acute stretch, we applied a step signal and varied amplitude and timing relative to AP onset. The model was then implemented in Myokit for tissue simulations, inducing myocardial stretch by activating SACs in a circular region while varying stretch stimulus size and timing.
Our simulations revealed that at cellular level, a 10-ms stretch triggered APs during phase 4 in all cases. However, with stretch stimuli falling in the AP plateau phase, early afterdepolarizations occurred only in LQT2. At tissue level, these effects were more pronounced: stretch could initiate more re-entries in SQT1 compared to the other variants, while only LQT2 could prevent the propagation of subsequent paced waves.
In conclusion, our simulations showed increased pro-arrhythmic effects of stretch in SQT1 and LQT2, while highlighting genotype-specific mechanisms.

Jung, Alexander: Neural network emulation of the human ventricular cardiomyocyte action potential for more efficient computations in pharmacological studies
Interfacing computer models of the human ventricular cardiomyocyte action potential (AP) with experimental data can become a substantial computational burden when considering that each simulation includes up to thousands of beats to reach the limit cycle. To address this issue, an emulator based on a neural network was developed that rapidly predicts the limit cycle AP for given maximum conductances of crucial ion channels, pumps, and exchangers. The training was performed on simulation data produced by the ToR-ORd-dynCl model and the evaluation was done for forward (find drugged AP for given pharmacological scaling factors of maximum conductances) and inverse problems (find pharmacological scaling factors for given APs before and after drug administration) of pharmacology studies on synthetic and experimental data. First and foremost, it has been shown that the NN emulator potentially enables massive speed-ups compared to regular simulations simulations (e.g. 171x when using Myokit for the simulation of 100 beats on a CPU). Furthermore, the forward problem on synthetic data could be solved with average root-mean-square errors (RMSE) of 0.47 mV in normal APs and of 14.5 mV in abnormal APs exhibiting early afterdepolarisations (drug-induced abnormal repolarization). The inverse problem on synthetic data could be solved with a maximum RMSE of 0.22 in the inferred pharmacological scaling factors. However, notable mismatches were observed between scaling factors inferred from experimental APs, and corresponding measured scaling factors. These mismatches can be attributed particularly to the fact that AP recordings were performed in small tissue preparations while the emulator was trained on single cardiomyocyte data. Overall, our study highlights the potential of neural network emulators as tool for an increased efficiency in pharmacology studies.

Lainscsek, Xenia: Predicting 3D chromosome architecture with delay differential analysis
The 3D conformation of chromatin is crucial for the precise regulation of gene expression. Enhancers regulate the transcription of genes by interacting with their respective promoters in what is known as enhancer-promoter interaction (EPI). Enhancers can be located thousands to millions of base-pairs away from genes they regulate. Experimental techniques such as HiC and ChIA-PET enable the identification of EPI pairs, many of which are cell-type specific. The high demands and costs of such experiments limit the availability of EPI data across cell-types. This has driven the development of computational learning models which use sequence and/or epigenomic data for predicting EPIs. Purely sequence-based models are advantageous for studying the effects of genetic variants, but face challenges in achieving satisfactory predictive performance. This naturally raises the question to what extent the information underlying valid EPI pairs is encoded within the DNA sequence itself? 
We analyze sequences of EPI pairs in the context of delay differential analysis (DDA), a nonlinear time-series classification framework which we have extended for analyzing DNA sequences (DNA-DDA). DDA employs a sparse functional delay differential embedding in a fundamentally different approach to traditional machine learning techniques. DDA model parameters are fixed and not iteratively updated to learn through repeated training cycles. DDA can achieve high classification performance using a low dimensional feature set rendering it insensitive to overfitting. DNA-DDA has been used to predict genome-wide A/B compartmentalization, a fundamental feature of 3D chromatin architecture. It competed well with alternative methods while requiring a fraction of the data used for training. This highlights the method's ability to extract key sequence features making it a promising alternative computational tool for studying various genomic contexts such as the EPI pair classification task.

Ottavi, Enia: Rule-Based Modelling of Cardiac Fibers in Complex Atrial Geometries with Applications to Electrophysiological Simulations
The human heart functions through a delicate interplay between its structural and electrophysiological properties, with myocardial fibers playing a key role in coordinating contraction and electrical propagation. While ventricular fiber organization is well-documented, the complex atrial geometry and variable orientations of atrial fibers present significant challenges for accurate modeling. This research project addresses these challenges by refining a Laplace-Dirichlet rule-based method for generating fiber orientations tailored to a specific atrial geometry and integrating these results into electrophysiological simulations  within Propag-5, a specialized cardiac modeling software.
First, the heart's anatomy is explored, connecting microscopic cellular structures to the gross anatomy and highlighting the role of myocardial fibers in cardiac function. To validate the method, a simplified ventricular geometry is employed to replicate key findings from Streeter’s studies on fiber organization. The rule-based method is then applied to a complex left atrial geometry, derived from the biatrial Lugano-Maastricht model. This application demonstrates good adherence to histological data, capturing the main characteristics of atrial fiber architecture. By employing this method, the fiber orientations are determined, and the corresponding fiber angles are computed. Finally, the mathematical setting for electrophysiological simulations based on the mondomain model is presented, incorporating the fiber data into the Propag software. This work enhances the understanding of atrial fiber organization and provides a framework for its replication in computational models, improving the accuracy of electrophysiological simulations and providing new insights into atrial function in health and disease.

Pera, Camilla: Towards a Sex-Specific Action Potential Model for Rabbit Atrial Cells: Mathematical Modeling and Analysis
This thesis presents the analysis of two mathematical models of the action potentials of rabbit atrial cells, with a focus on gender differences. Gender medicine is increasingly becoming a necessity, recognizing how gender and hormones influence the symptomatology and treatment of diseases. In particular, we focused on sex differences in cardiac electrophysiology, thanks to experimental data provided by Katja Odening's laboratory at the University of Bern. We therefore started by analyzing two pre-existing models for atrial electrophysiology and performed sensitivity analysis using Sobol' indices in order to identify the ionic conductances most influential on the properties of the action potential. To cope with computational complexity, we developed a Gaussian Process emulator, significantly reducing the computational time required for sensitivity analysis while maintaining high accuracy. Parameter optimization, guided by sensitivity analysis results, successfully reproduced experimental action potentials. We employed Approximate Bayesian Computation with Sequential Monte Carlo methods (SMC-ABC) to estimate model parameters and quantify uncertainties, revealing varying degrees of parameter identifiability. While our analysis did not identify significant sex-based differences in ion channels conductances, possibly due to limited sample size, this work establishes a framework for investigating sex-specific cardiac electrophysiology through mathematical modeling.

Picó Cabiró, Sergi: A comprehensive model of atrial electromechanics
Atrial Fibrillation (AF) is a common heart rhythm disorder characterized by fast and irregular beats in the upper heart chambers (atria). Given its association with severe conditions such as stroke, heart failure, and other cardiovascular issues, understanding and addressing AF is crucial for advancing treatment strategies.

We present a 3D multiscale electromechanical framework for atrial simulations, providing a comprehensive tool for gaining insights into AF mechanisms. Our approach couples an electrophysiological model for atrial cells with an orthotropic model for passive mechanics through calcium-driven excitation-contraction dynamics, linking electrical and mechanical processes. Additionally, the framework incorporates a 0D closed-loop hemodynamic model representing the human circulatory system.

In this study, we demonstrate our framework's ability to simulate atrial physiology across multiple scales, including cellular, tissue, and whole-organ levels, using biatrial geometry. Electrophysiological and mechanical biomarkers are compared with literature reported data. Furthermore, we explore the impact of mechanical coupling on electrophysiological simulations, showing the interplay between electrical and mechanical processes. To simulate AF, we incorporate structural remodeling into our electrophysiological models and replicate arrhythmia generation through ectopic activity mechanisms. Future work will extend this approach to simulate arrhythmias in electromechanical simulations and validate the AF model at this level.

Piersanti, Roberto: Comprehensive multiscale electromechanical modeling of the atrial function
Atrial fibrillation (AF) is a prevalent disease condition that disrupts atrial structure, electrophysiology, and biomechanical function. However, the intricate interplay between these structural, electrical, and mechanical factors in AF remains poorly understood. To better understand these mechanisms, we present a biophysically detailed, fully coupled multiscale mathematical model of cardiac electromechanics (EM) tailored to the human atria. Our atrial EM model integrates several features: (i) a detailed myocardial fiber architecture derived using a rule-based method validated against diffusion tensor magnetic resonance imaging data; (ii) a spatially heterogeneous fibrosis model accounting for scar-fibrotic regions; (iii) state-of-the-art electrophysiological and mechanical models; (iv) a microscale active force-generation model specifically calibrated for atrial morphologies; (v) key interactions between electrical and mechanical systems, including mechano-electric feedback; (vi) a 0D closed-loop model of the whole cardiovascular system that dynamically interacts with the atrial mechanics;  Our EM framework is versatile, supporting simulations in full biatrial geometries or individual left/right atria. We demonstrate the effectiveness of our model through EM simulations of sinus rhythm (SR) in a realistic biatrial geometry, accurately reproducing healthy atrial functions, including electrical activation, eight-shape pressure-volume loops, time-resolved pressures, and three-dimensional deformation. Additionally, we simulate AF scenarios, highlighting reentrant electrical activity around fibrotic patches in real left atrial geometry. These reentrant patterns result in marked reductions in atrial displacement at peak contraction compared to SR, reflecting characteristic dysfunction in AF. Our model offers a comprehensive tool for exploring the electromechanical mechanisms underlying AF and provides a foundation for future investigations into atrial pathophysiology.

Ranno, Anna: Hemodynamics in stented arteries: full and reduced computational models
We present a computational model for arteries with drug-eluting stents, where in-stent restenosis (ISR) occurs.
The multiphysics aspect encompasses interactions between blood and arterial wall. We include hemodynamics, drug absorption in the wall and drug elution in the lumen, hydrophobic drug characteristics, and cell species interactions governing ISR growth. Furthermore, blood flow dynamics operate on a scale of seconds while ISR and drug elution occur over weeks. We address the multiscale nature of the model by means of quasi-steady assumptions and homogenization.
In this work, we focus on drug elution and on the mutual influence of disturbed hemodynamics and ISR. We test the model on an idealized artery with a ring stent, followed by a geometry obtained from patient data, incorporating virtual implantation of a single-crown stent. To further reduce the computational costs, we introduce a fictitious domain formulation based on hierarchical model reduction.

Schrotter, Thomas: A Full Physics Real-Time Solver for Simulating Arrhythmias in the Human Heart
For simulating human cardiac electrophysiology, the bidomain model is considered the gold standard owing to its ability to replicate experimental observations with high fidelity. However, its reaction-diffusion nature and associated constraints on spatiotemporal discretization result in significant computational costs, limiting its adoptions in industrial and clinical applications. Eikonal-based models are computationally more efficient but lack the capability to accurately simulate cardiac arrhythmias. We present a real-time capable reentrant eikonal (REK) model that combines an eikonal solver with a finite state machine to track action potential phases and incorporates physiological restitution, curvature and diffusion effects, as measured in a high fidelity reference reaction-diffusion model. Local activation times, conduction velocities and local repolarization times measured in infarct simulations were compared against baseline reaction-diffusion monodomain solutions, where low root mean square errors of 2.4 ms, 1.6 mm/s and 6.8 ms were reported, respectively. Reconstructed transmembrane voltages and electro-cardiograms qualitatively agree closely with the baseline solution, while offering a speedup of ~420. Our results demonstrate that the REK model is capable of simulating cardiac arrhythmias in real-time, matching monodomain solutions with minor deviations.

Stengel, Laura: Enriched and Discontinuous Galerkin Discretizations for a Cardiac Mechanics Benchmark Problem
Computer models of the human heart can lead to a better understanding of cardiac function. Since the objective of many of these models is to be used in a clinical setting, a compromise between computational cost and numerical accuracy is needed. Due to the high mathematical complexity of the underlying model, the finite element discretization commonly used may not be the optimal balance between efficiency, reliability, and accuracy.
To investigate the impact of different spatial discretization schemes on cardiac mechanics, we realized a benchmark configuration that considers the hyper-elastic problem of inflating and actively contracting an idealized left ventricle with transversely isotropic and nearly incompressible properties. In this study, we examined the influence of three different finite elements — conforming Galerkin (cG), discontinuous Galerkin (dG), and enriched Galerkin (eG) elements — by investigating the cavity volume and apex shortening for four mesh refinements. Furthermore, we compare the various spatial discretizations concerning the number of degrees of freedom and computational time. All simulations were conducted using both linear and quadratic elements for all methods.
We demonstrate that the cG scheme leads to the occurrence of locking phenomena for coarse mesh resolutions using linear elements. However, locking can be mitigated by using finer mesh resolutions, higher-order elements, or by adopting the dG or eG elements. 
DG elements have notably more degrees of freedom compared to the cG method, while eG discretization has only one additional per element. However, both eG and dG schemes cause higher computational costs, particularly the dG method. Furthermore, simulations utilizing the eG and dG schemes demonstrate enhanced robustness and stability compared to the conforming method.
In conclusion, the eG approach offers a favorable balance between computational efficiency and numerical robustness in cardiac modeling applications.

Thaler, Franz: Augmentation-based Domain Generalization and Joint Training from Multiple Source Domains for Whole Heart Segmentation
As the leading cause of death worldwide, cardiovascular diseases motivate the development of more sophisticated methods to analyze the heart and its substructures from medical images like Computed Tomography (CT) and Magnetic Resonance (MR). Semantic segmentations of important cardiac structures that represent the whole heart are useful to assess patient-specific cardiac morphology and pathology. Furthermore, accurate semantic segmentations can be used to generate cardiac digital twin models which allows e.g. electrophysiological simulation and personalized therapy planning. Even though deep learning-based methods for medical image segmentation achieved great advancements over the last decade, retaining good performance under domain shift – i.e. when training and test data are sampled from different data distributions – remains challenging. In order to perform well on domains known at training-time, we employ a (1) balanced joint training approach that utilizes CT and MR data in equal amounts from different source domains. Further, aiming to alleviate domain shift towards domains only encountered at test-time, we rely on (2) strong intensity and spatial augmentation techniques to greatly diversify the available training data. Our proposed whole heart segmentation method, a 5-fold ensemble with our contributions, achieves the best performance for MR data overall and a performance similar to the best performance for CT data when compared to a model trained solely on CT. With 93.33% DSC and 0.8388 mm ASSD for CT and 89.30% DSC and 1.2411 mm ASSD for MR data, our method demonstrates great potential to efficiently obtain accurate semantic segmentations from which patient-specific cardiac twin models can be generated.