Abstract:

Cone-beam computed tomography (CBCT) has become a widely adopted modality for image-guided radiotherapy (IGRT). However, CBCT is characterized by increased noise, limited soft-tissue contrast, and artifacts. These issues result in unreliable Hounsfield unit (HU) values, which limits direct dose calculation. These issues were addressed by generating synthetic CT (sCT) from CBCT, particularly by adopting deep learning (DL) methods. However, existing DL approaches were hindered by institutional heterogeneity, scanner-dependent variations, and data privacy regulations that prevented multi-center data sharing.To overcome these challenges, we proposed a cross-silo federated learning approach for CBCT-to-sCT synthesis in the head and neck region. This approach extended our original FedSynthCT framework to a different image modality and anatomical region. A conditional generative adversarial network (cGAN) was trained using data from three European medical centers within the SynthRAD2025 public challenge dataset while maintaining data privacy at each institution. A combination of the FedAvg and FedProx aggregation strategies, alongside a standardized preprocessing pipeline, was adopted to federate the DL model.The federated model effectively generalized across participating centers, as evidenced by the mean absolute error (MAE) ranging from $64.38\pm13.63$ to $85.90\pm7.10$ HU, the structural similarity index (SSIM) ranging from $0.882\pm0.022$ to $0.922\pm0.039$, and the peak signal-to-noise ratio (PSNR) ranging from $32.86\pm0.94$ to $34.91\pm1.04$ dB. Notably, performance on an external validation dataset of 60 patients yielded comparable metrics: a MAE of $75.22\pm11.81$ HU, an SSIM of $0.904\pm0.034$ and a PSNR of $33.52\pm2.06$, confirming robust cross-center generalization despite differences in imaging protocols and scanner types, without additional training. Furthermore, a visual analysis of the results revealed that the obtained metrics were influenced by registration errors.Our findings demonstrated the technical feasibility of FL for CBCT-to-sCT synthesis task while preserving data privacy, offering a collaborative solution for developing generalizable models across institutions without requiring data sharing or center-specific models.

Abstract:

The generation of Synthetic Computed Tomography (sCT) images has become a pivotal methodology in modern clinical practice, particularly in the context of Radiotherapy (RT) treatment planning. The use of sCT enables the calculation of doses, pushing towards Magnetic Resonance Imaging (MRI) guided radiotherapy treatments. Moreover, with the introduction of MRI-Positron Emission Tomography (PET) hybrid scanners, the derivation of sCT from MRI can improve the attenuation correction of PET images.Deep learning methods for MRI-to-sCT have shown promising results, but their reliance on single-centre training dataset limits generalisation capabilities to diverse clinical settings. Moreover, creating centralised multi-centre datasets may pose privacy concerns. To address the aforementioned issues, we introduced FedSynthCT-Brain, an approach based on the Federated Learning (FL) paradigm for MRI-to-sCT in brain imaging. This is among the first applications of FL for MRI-to-sCT, employing a cross-silo horizontal FL approach that allows multiple centres to collaboratively train a U-Net-based deep learning model. We validated our method using real multicentre data from four European and American centres, simulating heterogeneous scanner types and acquisition modalities, and tested its performance on an independent dataset from a centre outside the federation.In the case of the unseen centre, the federated model achieved a median Mean Absolute Error (MAE) of 102.0 HU across 23 patients, with an interquartile range of 96.7–110.5 HU. The median (interquartile range) for the Structural Similarity Index (SSIM) and the Peak Signal to Noise Ratio (PNSR) were 0.89 (0.86–0.89) and 26.58 (25.52–27.42), respectively.The analysis of the results showed acceptable performances of the federated approach, thus highlighting the potential of FL to enhance MRI-to-sCT to improve generalisability and advancing safe and equitable clinical applications while fostering collaboration and preserving data privacy.

Abstract:

This review focuses on the computerized modeling of the electrophysiology of the human atria, emphasizing the simulation of common arrhythmias such as atrial flutter (AFlut) and atrial fibrillation (AFib). Which components of the model are necessary to accurately model arrhythmogenic tissue modifications, including remodeling, cardiomyopathy, and fibrosis, to ensure reliable simulations? The central question explored is the level of detail required for trustworthy simulations for a specific context of use. The review discusses the balance between model complexity and computational efficiency, highlighting the risks of oversimplification and excessive detail. It covers various aspects of atrial modeling, from cellular to whole atria levels, including the influence of atrial geometry, fiber direction, anisotropy, and wall thickness on simulation outcomes. The article also examines the impact of different modeling approaches, such as volumetric 3D models, bilayer models, and single surface models, on the realism of simulations. In addition, it reviews the latest advances in the modeling of fibrotic tissue and the verification and validation of atrial models. The intended use of these models in planning and optimization of atrial ablation strategies is discussed, with a focus on personalized modeling for individual patients and cohort-based approaches for broader applications. The review concludes by emphasizing the importance of integrating experimental data and clinical validation to enhance the utility of computerized atrial models to improve patient outcomes.

Abstract:

Since the turn of the millennium, computational modelling of biological systems has evolved remarkably and sees matured use spanning basic and clinical research. While the topic of the peri-millennial debate about the virtues and limitations of 'reductionism and integrationism' seems less controversial today, a new apparent dichotomy dominates discussions: mechanistic vs. data-driven modelling. In light of this distinction, we provide an overview of recent achievements and new challenges with a focus on the cardiovascular system. Attention has shifted from generating a universal model of the human to either models of individual humans (digital twins) or entire cohorts of models representative of clinical populations to enable in silico clinical trials. Disease-specific parameterisation, inter-individual and intra-individual variability, uncertainty quantification as well as interoperable, standardised, and quality-controlled data are important issues today, which call for open tools, data and metadata standards, as well as strong community interactions. The quantitative, biophysical, and highly controlled approach provided by in silico methods has become an integral part of physiological and medical research. In silico methods have the potential to accelerate future progress also in the fields of integrated multi-physics modelling, multi-scale models, virtual cohort studies, and machine learning beyond what is feasible today. In fact, mechanistic and data-driven modelling can complement each other synergistically and fuel tomorrow's artificial intelligence applications to further our understanding of physiology and disease mechanisms, to generate new hypotheses and assess their plausibility, and thus to contribute to the evolution of preventive, diagnostic, and therapeutic approaches.

Abstract:

PurposeIdentifying and quantifying coronary artery calcification (CAC) is crucial for preoperative planning, as it helps to estimate both the complexity of the 2D coronary angiography (2DCA) procedure and the risk of developing intraoperative complications. Despite the relevance, the actual practice relies upon visual inspection of the 2DCA image frames by clinicians. This procedure is prone to inaccuracies due to the poor contrast and small size of the CAC; moreover, it is dependent on the physician’s experience. To address this issue, we developed a workflow to assist clinicians in identifying CAC within 2DCA using data from 44 image acquisitions across 14 patients.MethodsOur workflow consists of three stages. In the first stage, a classification backbone based on ResNet-18 is applied to guide the CAC identification by extracting relevant features from 2DCA frames. In the second stage, a U-Net decoder architecture, mirroring the encoding structure of the ResNet-18, is employed to identify the regions of interest (ROI) of the CAC. Eventually, a post-processing step refines the results to obtain the final ROI. The workflow was evaluated using a leave-out cross-validation.ResultsThe proposed method outperformed the comparative methods by achieving an F1-score for the classification step of 0.87 (0.77-0.94) (median ± quartiles), while for the CAC identification step the intersection over minimum (IoM) was 0.64 (0.46-0.86) (median ± quartiles).ConclusionThis is the first attempt to propose a clinical decision support system to assist the identification of CAC within 2DCA. The proposed workflow holds the potential to improve both the accuracy and efficiency of CAC quantification, with promising clinical applications. As future work, the concurrent use of multiple auxiliary tasks could be explored to further improve the segmentation performance.

Abstract:

BACKGROUND AND OBJECTIVE: Radiomics extracts quantitative features from magnetic resonance images (MRI) and is especially useful in the presence of subtle pathological changes within human soft tissues. This scenario, however, may not cover the effects of intrinsic, e.g., aging-related, (physiological) neurodegeneration of normal brain tissue. The aim of the work was to study the repeatability of radiomic features extracted from an apparently normal area in longitudinally acquired T1-weighted MR brain images using three different intensity normalization approaches typically used in radiomics: Z-score, WhiteStripe and Nyul. METHODS: Fifty-nine images of hearing impaired, yet cognitively intact, patients were repeatedly acquired in two different time points within six months. Ninety-one radiomic features were obtained from an area within the pons region, considered to be a healthy brain tissue according to previous analyses and reports. The Intraclass Correlation Coefficient (ICC) and the Concordance Correlation Coefficient (CCC) in the repeatability study were used as metrics. RESULTS: Features extracted from the MRI normalized with Z-score showed results comparable in both ICC (0.90 (0.82-0.98)) and CCC (0.82 (0.69-0.95)) distribution values, in terms of median and quartiles, with those extracted from the images normalized with WhiteStripe (0.89 (0.84-0.92)) and (0.80 (0.73-0.84)), respectively. CONCLUSION: Our findings underline the importance of, providing useful guidelines for, the intensity normalization of brain MRI prior to a longitudinal radiomic analysis.

Abstract:

Radiotherapy (RT) is a cancer treatment technique that involves exposing cells to ionizing radiation, including X-rays, electrons, or protons. RT offers promise to treat cancer, however, some inherent limitations can hamper its performance. Radio-resistance, whether innate or acquired, refers to the ability of tumor cells to withstand treatment, making it a key factor in RT failure. This perspective hypothesizes that nanoscale surface topography can impact on the topology of cancer cells network under radiation, and that this understanding can possibly advance the assessment of cell radio-resistance in RT applications. An experimental plan is proposed to test this hypothesis, using cancer cells exposed to various RT forms. By examining the influence of 2D surface and 3D scaffold nanoscale architecture on cancer cells, this approach diverges from traditional methodologies, such as clonogenic assays, offering a novel viewpoint that integrates fields such as tissue engineering, artificial intelligence, and nanotechnology. The hypotheses at the base of this perspective not only may advance cancer treatment but also offers insights into the broader field of structural biology. Nanotechnology and label-free Raman phenotyping of biological samples are lenses through which scientists can possibly better elucidate the structure-function relationship in biological systems.

Abstract:

Regular activation of the heart originates from cyclic spontaneous depolarisations of sinoatrial node cells (SANCs). Variations in electrolyte levels, commonly observed in haemodialysis (HD) patients, and the autonomic nervous system (ANS) profoundly affect the SANC function. Thus we investigated the effects of hypocalcaemia and sympathetic stimulation on the SANC beating rate (BR). The β-adrenergic receptor (β-AR) signalling cascade, as described by Behar et al., was incorporated into the SANC models of Severi et al. (rabbit) and Fabbri et al. (human). Simulations were conducted across various extracellular calcium ([Ca]) (0.6-1.8 mM) and isoprenaline concentrations [ISO] (0-1000 nM) for a sufficient period of time to allow transient oscillations to equilibrate and reach a limit cycle. The β-AR cell response of the extended models was validated against new Langendorff-perfused rabbit heart experiments and literature data. The extended models revealed that decreased [Ca] necessitated an exponential-like increase in [ISO] to restore the basal BR. Specifically at 1.2 mM [Ca], the Severi and Fabbri models required 28.0 and 9.6 nM [ISO], respectively, to restore the initial BR. Further reduction in [Ca] to 0.6 mM required 170.0 and 43.6 nM [ISO] to compensate for hypocalcaemia. A sudden loss of sympathetic tone at low [Ca] resulted in a loss of automaticity within seconds. These findings suggest that hypocalcaemic bradycardia can be compensated for by an elevated sympathetic tone. The integration of the β-AR pathways led to a logarithmic BR increase and offers insights into potential pathomechanisms underlying sudden cardiac death (SCD) in HD patients. KEY POINTS: We extended the sinoatrial node cell (SANC) models of Severi et al. (rabbit) and Fabbri et al. (human) using the β-adrenergic receptor (β-AR) signalling cascade Behar et al. described. Simulations were conducted across various extracellular calcium ([Ca]) (0.6-1.8 mM) and isoprenaline concentrations [ISO] (0-1000 nM) to reflect conditions in haemodialysis (HD) patients. An exponential-like increase in [ISO] compensated for hypocalcaemia-induced bradycardia in both models, whereas interspecies differences increased the sensitivity of the extended Fabbri model towards hypocalcaemia and increased sympathetic tone. The extended models may help to further understand the pathomechanisms of several cardiovascular diseases affecting pacemaking, such as the high occurrence of sudden cardiac death (SCD) in chronic kidney disease (CKD) patients.

Abstract:

Cardiac fibrosis is a key factor in electrical conduction disturbances, yet its specific impact on conduction remains unclear, hindering predictive insight of cardiac electrophysiology and arrhythmogenesis. Among the different cardiac disorders, arrhythmogenic cardiomyopathy (ACM) is known to be associated with massive fibrotic remodelling of the myocardium, and it accounts for most cases of stress-related arrhythmic sudden death. To explore ACM further, we employed a Desmoglein-2-mutant mouse model and developed a correlative imaging approach to integrate macro-scale cardiac electrophysiology with 3D micro-scale reconstructions of the ventricles, to characterise the dynamics of conduction wavefronts and relate them to the underlying structural substrate. Our findings confirm that this ACM model shows localised replacement of cardiomyocytes with collagen and non-myocytes, contributing to electrical dysfunction. Moreover, we observed that conduction through fibrotic tissue areas shows a frequency-dependent behaviour, where conduction fails at high stimulation frequencies, promoting re-entrant arrhythmias, even in regions that were electrophysiologically inconspicuous at lower stimulation rates. Using a computational model, informed by high- resolution structural data, we found that frequency-dependent conduction through fibrotic tissue cannot be explained solely by collagen deposition or cardiomyocyte re-organisation. Indeed, fibrotic areas feature electrophysiological remodelling which acts as a low-pass filter for conduction, which can be quantitatively explained by electrotonic coupling of cardiomyocytes with non-myocytes. Collectively, our study provides a novel structure-function mapping pipeline and describes a previously unrecognised pro-arrhythmogenic mechanism in ACM, underscoring the need for dynamic assessment of functional conduction block in fibrotic myocardium using multiple diagnostic pacing protocols.

Abstract:

Background: Computer models for simulating cardiac electrophysiology are valuable tools for research and clinical applications. Traditional reaction-diffusion (RD) models used for these purposes are computationally expensive. While eikonal models offer a faster alternative, they are not well-suited to study cardiac arrhythmias driven by reentrant activity. The present work extends the diffusion-reaction eikonal alternant model (DREAM), incorporating conduction velocity (CV) restitution for simulating complex cardiac arrhythmias. Methods: The DREAM modifies the fast iterative method to model cyclical behavior, dynamic boundary conditions, and frequency-dependent anisotropic CV. Additionally, the model alternates with an approximated RD model, using a detailed ionic model for the reaction term and a triple-Gaussian to approximate the diffusion term. The DREAM and monodomain models were compared, simulating reentries in 2D manifolds with different resolutions. Results: The DREAM produced similar results across all resolutions, while experiments with the monodomain model failed at lower resolutions. CV restitution curves obtained using the DREAM closely approximated those produced by the monodomain simulations. Reentry in 2D slabs yielded similar results in vulnerable window and mean reentry duration for low CV in both models. In the left atrium, most inducing points identified by the DREAM were also present in the high-resolution monodomain model. DREAM's reentry simulations on meshes with an average edge length of 1600$\mu$m were 40x faster than monodomain simulations at 200$\mu$m. Conclusion: This work establishes the mathematical foundation for using the accelerated DREAM simulation method for cardiac electrophysiology. Cardiac research applications are enabled by a publicly available implementation in the openCARP simulator.

Abstract:

BACKGROUND: Longitudinal Displacement (LD) is the relative motion of the intima-media upon adventitia of the arterial wall during the cardiac cycle, probably linked to atherosclerosis. It has a direction, physiologically first backward in its main components with respect to the arterial flow. Here, LD was investigated in various disease and in presence of a unilateral carotid stent. METHODS: Carotid acquisitions were performed by ultrasound imaging on both body sides of 75 participants (150 Arteries). LD was measured in its percent quantity and direction. RESULTS: Obesity (p = 0.001) and carotid plaques (p = 0.01) were independently associated to quantity decrease of LD in the whole population. In a subgroup analysis, it was respectively 143% in healthy (n = 48 carotids), 129% (n = 34) in presence of cardiovascular risk factors, 121% (n = 20) in MACE patients, 119% (n = 24) in the carotid contralateral to a stent, 110% (n = 24) in carotids with stents. Regarding the direction of LD, in a subgroup analysis an inverted movement was identified in aged (p = 0.001) and diseased (p = 0.001) participants who also showed less quantity of LD (p = 0.001), but independently with age only (p = 0.002) in the whole population. CONCLUSIONS: This observational study suggests that LD within carotid wall layers is lower additively with ageing, cardiovascular risk factors, cardiovascular diseases, and stent. Even if stent is surely beneficial, these data might shed some light on stent restenosis, emphasising the need for interventional studies.

Abstract:

Cardiac arrhythmias are a leading cause of morbidity and mortality, with atrial fibrillation (AF) being the most prevalent sustained arrhythmia, significantly impacting global public health. Despite its importance, AF treatment strategies remain insufficient due to an incom- plete understanding of the underlying mechanisms. Computational models are valuable tools for advancing research in cardiac electrophysiology and enhancing patient diagnostics and treatment. While reaction diffusion (RD) models are widely used, they are computationally intensive. In contrast, the eikonal model is computationally less demanding, making it attractive for large-scale simulations and clinical scenarios where faster computation times are critical. However, several limitations hinder its application in realistic settings. This work aims to explore how the eikonal model can be better understood and expanded to improve research, diagnosis, and treatment of atrial arrhythmias.To investigate the standard eikonal model, it was first compared to RD models by simulating single beats under sinus rhythm, both with and without fibrosis. Then, the standard eikonal model was modified to provide additional outputs, including conduction velocity (CV) magnitude and wavefront propagation direction under anisotropic conditions, alongside activation times. These outputs served as ground truth to evaluate the radial basis function method, which estimates CV based on activation times while ignoring anisotropy.The eikonal model was then extended by combining it with the RD model, forming the diffusion-reaction eikonal alternant model (DREAM). A new cyclical fast iterative method (cycFIM) was introduced to solve the anisotropic eikonal equation while enabling reactivations, a complex challenge for iterative methods.To address the eikonal model’s limitations, particularly its inability to account for source- sink mismatch, regression models were developed. Bidomain simulations examined the effects of wall thickness and tissue curvature on CV. Moreover, monodomain simulations investigated pacing frequency and source-sink mismatch effects on diffusion current (DC) amplitude. These findings were integrated into regression models for the eikonal model. To facilitate the regression, a new method was introduced to quantify source-sink mismatch based solely on activation times and node coordinates.Comparison with RD models showed that the eikonal model reasonably simulates sinus rhythm in non-fibrotic atrial geometry but significantly declines in performance with fibrosis due to its inability to capture source-sink mismatches. Additionally, when assessing the radial basis function it was found that neglecting anisotropic propagation when estimating CV can lead to significant errors up to 700 mm/s. DREAM simulations maintained low computational costs while effectively simulating action potentials, node reactivations, and reentries. Investigating CV restitution revealed that the steepness of restitution curves can modulate the dynamics of the vulnerable window and the average duration of reentry. Regression models based on RD simulations successfully predicted key factors in cardiac electrophysiology, such as CV and DC amplitude, using data available during eikonal simulations, including activation times and geometric factors like wall thickness and tissue curvature.This research highlights the eikonal model’s potential in advancing the understanding and clinical management of cardiac arrhythmias. As a computationally efficient alternative to more complex models, it provides valuable insights into arrhythmia mechanisms, diagnosis, and treatment, making it ideal for large-scale simulations and clinical settings with limited computing resources. The DREAM was applied to explore the role of personalized CV restitution curves in reentry dynamics. This study paves the way for broader applications of eikonal-based models in cardiac electrophysiology, ultimately improving patient outcomes. Additionally, the new cycFIM, embedded in the DREAM, could be used to simulate cyclical wave propagation in other fields beyond medical applications.

Abstract:

Accurate pelvic alignment is essential for assessing femoroacetabular impingement (FAI) when using CT-based three-dimensional bone models, as the acquisition is performed in a supine rather than a standing position, which does not accurately reflect the patients kinematics. This study investigated whether reliable pelvic alignment could be achieved using a reduced, unilateral field of view (FOV) based solely on treatment-side anatomical landmarks,with the goal of reducing both radiation exposure and segmentation time.Correlation analyses of landmarks, found on the pelvis and femora, showed strong interdependence, with Pearson correlation coefficients reaching up to 97 %, enabling the selection of the femoral head center (FHC), anterior inferior iliac spine (AIIS), and pubic tubercle (PT) for alignment. Regression models were trained on these landmarks to accurately predict both treatment-side and contralateral anterior superior iliac spine (ASIS) positions, which are required for the standard alignment method, achieving mean absolute errors between 2.39 mm and 6.03 mm across all three cartesian axes. Error propagation analyses indicated that the alignment differences resulting from the predictions averaged by 2.8 →, which was expected to influence only the average anteversion measurement in less than 15 % of the cases. Isolated outliers were observed for Center Edge and Tönnigs Angle.Reducing the superior scan limit, from the ASIS to the AIIS, allowed for an average reduction of 21 % in scan length, translating into a proportional decrease in radiation exposure. Furthermore, the findings of this study support the hypothesis that unilateral landmark information is sufficient for reliable pelvic alignment and FAI evaluation, offering a practical approach to optimize segmentation time and reduce patient dose. Further validation on larger datasets is recommended to ensure robustness and generalizability.

Abstract:

Upper limb amputation can severely affect motor function and daily activities, requiring extensive rehabilitation. Conventional rehabilitation methods are limited by therapist availability, timing of prosthesis fitting and inconsistent effects on body schema. Virtual Reality (VR)-based methods offer a promising alternative, providing customizable visual feedback that supports early-stage rehabilitation and can help address perceptual distortions.In this thesis, a method to calibrate a virtual limb for upper limb amputees in rehabilitation is implemented and assessed regarding its alignment with the physical limb. The thesis evaluates the overall spatial accuracy, the influence of temporal drift and HMD movement and the effect of target location on fingertip alignment. The goal was to determine whether a virtual limb can achieve sufficiently good positional alignment for practical use as a rehabilitation approach.The visual movement of the virtual limb is based on a motion tracker on the residual forearmand the calibration uses the intact arm as a reference with the HMD’s built-in hand tracking.Validation was performed using a 3D camera to track the physical fingertip position using image processing. Registration and transformation between the camera and VR space enable the quantification of positional deviations under controlled experimental conditions.Experimental results showed a median fingertip deviation of 5.9 cm (mean: 8.6 cm), which is below the perceptual threshold reported in literature (7.8 cm), indicating that the approach can achieve sufficient accuracy. Temporal drift and HMD movement had no significant effect, whereas target location significantly influenced accuracy. Further examination of error sources showed that some errors stem from external tracking limitations, suggesting that practical accuracy may be even better than the one measured.Overall, this work demonstrates that automated VR limb calibration is feasible and accurate, providing a reliable basis for further work with amputees, technical improvements and applications in rehabilitation and training.

Abstract:

Background: Transfer learning has proven to be a powerful technique in medical image analysis, particularly when dealing with limited training data. This thesis investigates the application of three-dimensional (3D) convolutional neural networks (CNNs) with transfer learning for the classification of systemic sarcoidosis using [18F]-fluorodeoxyglucose ([18F]FDG)-PET images. Sarcoidosis is a multisystem inflammatory disease that can affect various organs, with systemic sarcoidosis being particularly challenging to diagnose. Early and accurate detection is crucial for patient management and treatment planning. The primary motivation for this research stems from the challenges in sarcoidosis diagnosis,where traditional methods often lack sensitivity and specificity. PET imaging has emerged as a valuable tool for detecting active inflammation, but manual interpretation is time-consuming and subject to inter-observer variability. This study aims to develop an automated classification system that can assist clinicians in identifying patients with systemic sarcoidosis from PET scans, and to compare its performance against 2D models using quantitative metrics.Methods: The methodology employs 3D residual networks (ResNets) architectures pre-trained on the 3DSeg-8 dataset, which contains diverse 3D medical imaging data. The pre-trained models are fine-tuned on a dataset of 260 samples from patients with confirmed sarcoidosis status. To address the limited dataset size and ensure robust evaluation, a 5-fold cross-validation strategy is implemented. The input PET volumes are preprocessed to a standardized size, with intensity normalization and data augmentation techniques including elastic deformation and random cropping are applied during training. The training pipeline incorporates several optimization strategies, such as adaptive learning rate scheduling, early stopping to prevent overfitting, and differential learning rates forpre-trained and newly added layers. Multiple optimizer configurations (Adam, AdamW, SGD) are evaluated to determine the optimal training setup. The classification performance is assessed using comprehensive metrics including accuracy, precision, recall, specificity, F1-score.Result: Based on the experiments, the proposed 3D models did not show better classification performance than those based on 2D data. Despite extensive experimentation with various state-of-the-art techniques, the developed 3D models consistently exhibited significant overfitting.

Abstract:

Respiratory rate is one of the most fundamental clinical vital signs and often serves as the first indication of an upcoming decrease in health status. The aim of this work is to develop an al- ternative concept for the continuous monitoring of respiratory rate. This will be implemented using bioimpedance measurement via a two-electrode-system as an alternative to the existing system. This allows the existing infrastructure to continue to be used without limitation, enabling direct integration into the existing measurement system. The measurement system developed is based on a multi-frequency approach in which an alternating current is genera- ted via a voltage controlled Howland current source with three carrier frequencies (43 kHz, 48 kHz and 53 kHz). The use of multiple carrier frequencies allows a flexible adaptation to different measurement conditions, such as changes in electrode position and increases the reliability of the system, as it is possible to switch to alternative frequencies without any problems in the case of interference. The measurement circuit is designed to be compatible with cables of different impedances. As part of this work, envelope demodulation and synchronous demodulation of the simulated respiratory signal were investigated. The analysis showed that synchronous demodulation delivers the most reliable results. When comparing the demodulated signal with a simu- lated reference signal, frequency deviations of up to 7.5% were observed, with particular uncertainties in the amplitude determination. It is implemented on the software side, which allows for easy adaptation to different system requirements and increases the flexibility of the overall system. Measurements also show that phase-based analysis between the useful signal and a re- ference signal makes it possible to detect the type of cable connected. This allows the output voltage or output current to be adaptively adjusted to the specific properties of the respective cable. In order to optimally utilise the dynamic range of the analog-to-digital converter (ADC), the input signal is specifically amplified or attenuated at the ADC. Measurements showed that, in addition to the amplification of the carrier signal, the amplitude of the demo- dulated signal was also specifically increased. The targeted amplification of the amplitude of the demodulated signal enables more efficient use of the dynamic range of the ADC, which improves robustness against interference and increases the resolution of the signal.

Abstract:

Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia, leading to ineffective atrial contraction. Its prevalence is expected to rise significantly in the coming decades, further increasing the associated clinical and economic burden. While clinical treatments are often ineffective, computational models have become valuable tools for investigating AF mechanisms and improving therapy. However, the high computational cost of detailed reaction-diffusion (RD) models limits their use in real-time clinical applications, whereas faster eikonal models cannot capture complex arrhythmic dynamics. The diffusion-reaction eikonal alternant model (DREAM) offers a promising compromise but relies on internal parameters whose potential and influence have not yet been systematically investigated.\\This thesis introduces a sequential three-step pipeline to characterize and optimize DREAM parameters in 2D reentry simulations. First, two global sensitivity analyses were conducted to identify the most influential parameters for different cardiac tissue properties. Based on these insights, a multi-objective optimization was designed to improve both simulation accuracy and computational efficiency.Finally, a fusion step was applied to derive a singlerobust parameter configuration, suitable across different tissue properties.In support of this pipeline, novel approaches were developed for quantifying simulation accuracy and detecting non-physiological propagation patterns.These approaches included distributional analyses of electrophysiological parameters and the application of interpretable machine learning techniques, such as the Gaussian Mixture Model and Decision Tree Classifier.\\The sensitivity analyses revealed that accuracy was more sensitive to parameter variation than computation time, both primarily driven by the propagation interval of the DREAM (\gls{tau_inc}). A significant portion of variation was linked to the prevention of propagation artifacts, which depended on specific model parameters and the tissue anisotropy ratio (AR). Similarly, the optimization results showed that the optimal parameter sets varied by performance objective and AR. Therefore, a distinct set of parameters was identified for each objective.The simulation error, measured relative to a monodomain model, decreased by up to 40\% and computation time by up to 7\% compared to the default parameter set. Although each objective required a distinct optimal set, a corresponding configuration was identified for each that remained robust across varying ARs.Overall, this thesis provides not only an optimized parameter configuration but also offers the insights needed to adapt it to new or more complex simulation scenarios. Furthermore, the developed approaches for quantifying accuracy and detecting non-physiological propagation patterns are applicable beyond DREAM in cardiac modeling research.

Abstract:

Magnetic Resonance Imaging (MRI) machines acquire data in the complex valued k-space. To perform an image reconstruction and convert this data to the image space, an inverse Fourier transformation is used. In most cases, clinical and preclinical datasets only contain the magnitude information, since the phase information was discarded. Training a neural network to reconstruct undersampled k-space data requires complex valued images.While Deep Learning (DL) techniques have been created to generate phase maps for a given magnitude image, they were constrained to low-resolution images, where the dataset was acquired with a single protocol for a specific anatomical region. These works did not assess the quality of the generated phase directly and only compared the reconstruction results for a network trained with a datasest containing this generated data. This work uses a conditional Generative Adversarial Network (cGAN) and a Bi-directional Conditional Adversarial Auto-Encoder (BiCAAE) to generate the phase maps for a given 2D magnitude image. By adding a complex valued neighbouring slice as a further conditionalinput, features in the third dimension can be considered by the network. This 3D network used 2D operations to reduce the computational strain. The preclinical images used to train the networks were acquired by machines with various field strengths, using different protocols to image a variety of volumes.To evaluate the quality of the generated phases for each network, the Structural Similarity Index Measure (SSIM) and Jensen-Shannon Divergence (JSD) were used.The 2D cGAN achieved the best JSD score, while the 3D cGAN had the greatest similarity according to the SSIM, despite both networks generating unrealistic phase maps. In contrast, the 2D and 3D BiCAAE did not score as well, since they mainly contained minimumand maximum phase values. They did however have the most realistic looking phases. The training of these networks was more stable than the cGAN and converged for most hyperparameters.The presented networks managed to generate high-resolution phase maps from magnitude images acquired with various protocols and MRI machines. Using the SSIM and JSD, the generated phase could be evaluated.

Abstract:

Atrial fibrillation (AF) is the most common sustained arrhythmia and remains challenging totreat due to its complex underlying mechanisms. While the role of electrical remodeling inpromoting and maintaining AF is well established, it is still unclear whether this remodeledstate is reversible and if diseased atrial cardiomyocytes can be functionally restored. Toaddress this question, experimental Action potentials (APs) from porcine atrial cells wererecorded under different conditions: Sinus Rhythm (SR), AF after 4 and 8 weeks, and reverseremodeling stages after cardioversion in SR for another 4 and 8 weeks. AP features wereautomatically extracted using the Cardiac Action Potential Parameter Analysis (CAPPA)algorithm. In parallel, single-cell simulations were performed using the Courtemanche andPig Peris-Yague models. A diverse set of 500 model variants per group was generated usingLatin Hypercube Sampling (LHS) of twelve key ionic conductances. The aim was to investi-gate whether different conductance combinations can produce similar AP morphologies, andwhether characteristic differences between groups could be linked to distinct ion channelprofiles.The final populations were iteratively optimized to match the mean and variance of theexperimental data, while assuring model parameters were in physiologically meaningfulranges.These results provide insights into the ionic basis of atrial remodeling and its potentialreversibility. Additionally, trends in the conductance distributions suggest a partial reversalof remodeling in the 4- and 8-week Reverse Remodeled Sinus Rhythm (RE) groups, movingtoward SR characteristics, although no definitive conclusions can be drawn.

Abstract:

The generation of synthetic computed tomography (sCT) has become essential in the clinical workflow, especially for radiotherapy treatment planning (RTP). By providing the electrondensity maps required for dose calculation, sCTs enable the integration of magnetic resonance imaging (MRI)-only radiotherapy.While federated learning (FL) approaches have already shown promising results, using deep learning (DL) methods for MRI-to-sCT translation and preserving patient privacy, they are limited to a single anatomical region.To address this, this study validates the robustness of a decentralised MRI-to-sCT framework on differential anatomical sites, allowing multiple centers with different anatomical regions to collaboratively train one DL model. Specifically, different model architectures –U-Net, generative adversarial network (GAN) and a GAN with regions of interest (ROI)-based discriminators– and different federation setups (Brain-Pelvis-Abdomen and Pelvis-Abdomen) were evaluated. The best performance was obtained by training a U-Net in the pelvis-abdomen federation, achieving a median mean absolute error (MAE) of 62.0 HU, structural similarity index measure (SSIM) of 0.90 and peak signal-to-noise ratio (PSNR) of 29.95 dB on pelvic test data and a median MAE of 105.80 HU, SSIM of 0.91 and PSNR of 27.92 dB on the abdominal test data.Although the quantitative and visual results remain unsatisfactory, this study represents a first exploration of FL feasibility across different anatomical sites.

Abstract:

Heart failure (HF) refers to a condition in which the heart’s capacity to pump blood is reduced, making it unable to meet the body’s demands. It is a syndrome affecting 1 to 2 % of the adult population in the developed world, with its prevalence rising due to the aging of the population. Cardiac resynchronization therapy (CRT) has been identified as an effective treatment for patients with HF and a left ventricular ejection fraction (LVEF) ≤ 35 %. However, approx. 30 % of patients receiving the current standard CRT treatment do not benefit from it in the long run. Alternative pacing strategies targeting the cardiac conduction system emerged in recent years and show the potential to outperform the present standard methods. Nevertheless, the main challenges of predicting the long-term outcome of CRT and choosing the optimal pacing strategy remain since the underlying mechanisms are not well understood.This work uses an electromechanical heart model with an integrated His-Purkinje system (HPS) to simulate different CRT strategies. The basis for the study setup consists of simula- tions for both a healthy heart and one with left bundle branch block (LBBB), varying this scenario in implementing a scar in the septum of the heart. The heart suffering from LBBB with and without scar tissue was then treated with six different pacing strategies. All simula- tions were analyzed and compared using appropriate metrics regarding electrophysiology, mechanics and hemodynamics.Comparing the results from the base simulations, mechanical imbalances appear to be the potential root cause for the pathological decrease in LVEF typical for HF patients. The electrical dyssynchrony induced by LBBB leads to mechanical dyssynchrony without causing a significant reduction in LVEF. In line with clinical data, this work concludes that major reductions in LVEF are a consequence of remodeling due to an unbalanced ventricular load. Analyzing the different pacing methods, selective left bundle branch area pacing (sLBBAP) proves to be the most effective in terms of mechanical resynchronization and reduction of load imbalances for the scenario without scar tissue, closely followed by non-selective left bundle branch area pacing (nLBBAP). Considering both scenarios with and without scar tissue, biventricular pacing (BVP) (with and without interventricular delay) shows a more reliable performance than LBBAP. One disadvantage of BVP is an unphysiological strain peak at the side of the pacing lead. Both right ventricular apex pacing (RVAP) and left ventricular septum pacing (LVSP) provoke only minor improvements compared to the untreated scenario and were thus considered unsuitable for our case. Our work on modeling LBBB and different CRT methods provides valuable insights into the underlying effects following electrical dyssynchrony and resynchronization. Our find- ings indicate that changes in electrical activation immediately affect mechanics more than hemodynamics. For our model, LBBAP showed the best performance in electrically and mechanically resynchronizing the ventricles.

Abstract:

Sudden cardiac death (SCD) presents a significantly elevated risk to patients with chronickidney disease (CKD), a risk compounded by the electrolyte fluctuations induced duringhaemodialysis (HD). These fluctuations directly impact cardiac pacemaking, particularlythe sinoatrial node (SAN). Furthermore, the autonomic nervous system (ANS) exerts sig-nificant control over cardiac rhythm, making ANS modulation an important component ofaccurate sinoatrial node cell (SANC) modeling. Finally, the continuous advancements inelectrophysiology necessitate a flexible and adaptable approach to update and refine existingSANC models, ensuring incorporation of new experimental data and current formulations.This thesis addressed these challenges by pursuing two interconnected objectives. Therefore,an enhanced human SANC model based on experimental data collected by Verkerk et al.,incorporating dynamic concentrations of intracellular potassium concentration ([K+]i) andintracellular sodium concentration ([Na+]i) as well as the effects of ANS modulation wasdeveloped. Moreover, a robust and adaptable optimization pipeline, aiming at the parametersof the underlying current equations, was implemented.Thus, the resulting model was based on a combination of components of the Fabbri et al. andLoewe et al. model, both of which already aligned well with the experimental data, while theformer lacked dynamic formulations of [Na+]i and [K+]i and the latter showed an impairmentin terms of ANS stimulation. The Nelder Mead method was leveraged to optimize 16 selectedparameters based on a cost function consisting of the action potential (AP) and calciumtransient markers (CaT) markers obtained from parallel openCARP simulations comparedagainst experimental data.The achieved AP markers of the adapted model using an optimized parameter set yieldedgood accordance with the experimental data and improved on the original Fabbri and Loewemodels with regard to the overshoot (OS). The optimized model incorporated dynamic[Na+]i and [K+]i and correct effects of isoprenaline (ISO) and acetylcholine (ACh) for ANSmodulation. During the model development process, an adaptable and flexible optimizationpipeline was successfully established, and provides a great tool for further research in thistopic.

Abstract:

Patients with chronic kidney disease undergoing haemodialysis treatment face an increased risk of sudden cardiac death, a complication often linked to electrolyte imbalances and alterations in the autonomic nervous system. The sinoatrial node (SAN) is directly influenced by these changes, making it critical for investigating cardiac arrhythmias. Electrophysiological models of the SAN offer valuable insights into the pathomechanisms underlying these conditions. The aim of this study was to address the incomplete implementation of the cholinergic cell reaction in the extended sinoatrial node cell (SANC) models of Severi (rabbit) and Fabbri (human). A literature research on the effects of acetylcholine (ACh) on single rabbit SANCs was conducted, forming the basis for optimising the response of the extended Severi and Fabbri models to ACh. An updated formulation of the acetylcholine- activated K+ current (IKACh) was implemented in both models, and the influence of ACh on the AC-cAMP-PKA signalling cascade was modified by adjusting the kach rate constant. Additionally, the formulation of the L-type Ca2+ current (ICaL) gate shifting in the extended Severi model was adjusted to achieve lower beating rates (BRs). A Nelder-Mead simplex optimisation process was used to tune the model’s response to ACh application. At concentrations of 10, 50, and 100 nM, the updated extended Severi model predicted BRs of 156, 77, and 48 bpm, corresponding to reductions of 14 %, 59 %, and 74 % relative to the basal BR, closely aligning with experimental data. The updated extended Fabbri model produced BRs of 61 and 31 bpm upon application of 10 and 50 nM [ACh], translating to reductions of 18.0 % and 59 %, respectively, covering the full physiological BR range in humans. Further, simulating the accentuated antagonism effects reproduced experimentally observed trends. In summary, this work successfully addresses a key limitation of the extended Severi and Fabbri SANC models by incorporating an optimised representation of the cholinergic response.