Informatik/Technik
Dauerhafte URI für die Sektionhttps://epub.uni-luebeck.de/handle/zhb_hl/4
Listen
Neueste Veröffentlichungen
Item Security and confidentiality on shared computational resources(2026-06) Bruhns, Ida DorotheeThe distinction between local and remote computing is increasingly blurred as modern computation relies extensively on the use of shared resources. Pervasive sharing of computational resources is evident in many use cases such as cloud computing, where computational tasks are outsourced to remote servers. Addition- ally, rented servers, Virtual Private Networks (VPNs), and even web browsers often rely on shared hardware infrastructure. While the benefits of shared computing resources, such as scalability and cost- effectiveness, are well-documented, this trend also introduces novel security risks. The reliance on shared hardware infrastructure creates opportunities for unautho- rized access, data breaches, and other malicious activities. One very prominent example of sharing both hardware and data are machine learning applications. The use of machine learning applications is rapidly increasing in almost every part of our lives, which includes granting them access to highly sensitive information like health or credit data. At the same time, the models that are used grow larger and larger, necessitating substantial computational resources. This surge in resource consumption has led to a rise in outsourcing both training and inference processes, resulting in the processing of sensitive data on untrusted machines. In this thesis, we examine how to protect data in distributed machine learning systems. In particular, we look at outsourced computations on a machine with a Trusted Execution Environment (TEE) and a fast processing unit, such as a Graphics Processing Unit (GPU). I examined the SLALOM protocol, a seminal work in privacy-preserving inference. In this theses I present a new method, CARNIVAL, to significantly speed up the preprocessing phase. CARNIVAL leverages the pseudo- randomness of the Subset sum problem to enable efficient outsourcing during the preprocessing phase. The findings from the performance benchmarks demonstrate that CARNIVAL is a promising candidate for real-world implementations. A second possibility to continue working with the SLALOM framework, DASH, is introduced briefly. It builds on arithmetic Garbled Circuits (GCs) in combination with a TEE.Item Interaction-aware model predictive control for autonomous highway driving(2026) Zhang, XiaorongAutomated driving technology has developed rapidly over the past decades, profoundly transforming traditional modes of transportation. It offers significant conveniences by replacing human drivers in simple scenarios and improves safety through driver assistance systems. However, despite these advances, achieving full automation still faces substantial challenges. One critical challenge is accurately predicting the motion of surrounding vehicles in real time and incorporating these predictions into the control of autonomous vehicles to ensure safety. The diversity in individual driving behaviors, coupled with the complexity of modeling interactions among traffic participants under uncertainties, makes this problem particularly difficult. To provide insights into these challenges, this thesis focuses on interaction-aware traffic prediction and the safe control of autonomous vehicles on highways. Highways are chosen as the study environment due to their structured lanes and lower traffic density. The aim is to establish a research foundation in this setting that can later be extended to more complex traffic environments, such as urban roads. Specifically, this thesis investigates two primary research directions to address these challenges. The first direction separates the tasks of interaction-aware traffic prediction and safe control of autonomous vehicles. The second direction aims to integrate these tasks into a unified control architecture for greater simplicity and efficiency. In the first research direction, diverse intention-based models along with the Interacting Multiple Model Kalman Filter (IMM-KF) algorithm are employed to estimate and predict vehicle motion states while considering interactions. Possible vehicle motion maneuvers are represented by a finite number of normal scenarios. Additionally, a “worst-case” scenario is considered along with normal scenarios in a Scenario-based Model Predictive Control (SCMPC) architecture to ensure the safety of the autonomous vehicle. Moreover, to reduce the conservatism of the control strategy associated with considering the “worst-case” scenario, a new Contingency Model Predictive Control (CMPC) scheme is explored with time-varying prediction horizons and an enlarged terminal set. In the second research direction, the Minimizing Overall Braking Induced by Lane Change (MOBIL) model is employed to find out the possible lanes that vehicles may occupy based on vehicle interactions. With this lateral traffic prediction, the longitudinal states of surrounding vehicles are modeled simultaneously with those of the autonomous vehicle within a Model Predictive Control (MPC) structure. They are determined by minimizing the collective control costs of all vehicles through dynamic interaction-aware mechanisms. All proposed control architectures are validated in a high-fidelity IPG CarMaker simulation environment. Simulation results demonstrate the effectiveness of these approaches in safely controlling autonomous vehicles while considering vehicles’ interactions on highways. The findings of this thesis provide new perspectives on addressing challenges in high way autonomous driving and establish the foundation for extending safe control strate gies to more complex traffic scenarios involving interactive behaviors. This research advances the understanding of interaction modeling in multi-agent traffic environments, contributing structured approaches to represent and leverage vehicle interactions for safer autonomous control. The modular design of the proposed frameworks facilitates their integration with broader automated driving systems, such as perception and planning modules. Moreover, validation in a high-fidelity simulation environment offers a practical reference for a potential real-world implementation.Item Hierarchies of action control(2026) Heinrich, Nils WendelHumans act in environments that are uncertain, dynamically changing, and only partially controllable. Successful behavior in such settings requires more than reactive feedback correction or rule-based strategies. Agents must continuously regulate how precisely action goals are specified, how strongly predictions guide action, and how they adapt when control deteriorates. Despite extensive work on motor control, decision-making, and learning, existing accounts lack a unified explanation of how behavioral strategies, subjective control beliefs, and computational learning mechanisms jointly support adaptive action control in these dynamic environments that are characterized by uncertainty and action–effect contingencies that change over time. This dissertation develops and empirically grounds a multi-level framework of action control that integrates hierarchical accounts of intention, graded degrees of control, and belief-based regulation of agency. Action control emerges from the dynamic interaction between behavioral markers of ongoing regulation, cognitive forward models supporting both reactive and proactive action selection, and computational mechanisms driving parameter adjustments of these internal forward models. Across four studies that share the same continuous task environment, action control is investigated at behavioral, cognitive, and computational levels. Studies 1 and 2 show that gaze behavior implements a hierarchical organization of action goals over different temporal horizons. Close fixations support immediate control and state-dependent regulation, whereas distant fixations anchor attention to future task-relevant locations, supporting proactive planning. These complementary fixation types adapt flexibly to changes in environmental dynamics and action–effect contingencies, revealing how perceptual control implements both reactive and anticipatory action. Study 3 treats the Sense of Control, the subjective feeling of being in control, as a latent belief that is updated via Bayesian integration. Participants provided control ratings after each trial, serving as observable indicators of their evolving beliefs. Analysis showed that participants derived their ratings by integrating prior expectations with observed performance outcomes. Control ratings decreased when environmental outcomes violated participants’ expectations, and individuals differed in how strongly they weighted performance evidence, indicating variability in belief updating. These results suggest that the Sense of Control reflects the predictive accuracy of internal forward models. Study 4 translated these insights into a cognitive architecture combining forward predictions with error-driven learning. The Sense of Control from Study 3, modeled using the Bayesian framework, was reinterpreted as uncertainty associated with an internal forward model of the task’s environmental dynamics. This uncertainty modulated whether forward predictions were used during action selection. Simulations showed that when high uncertainty led the architecture to suspend forward prediction, learning was severely impaired. Adaptation was slow, performance remained below that of human participants, and internal representations failed to improve because no informative prediction errors were generated. When the architecture relied on predictions despite high uncertainty, corrective error signals enabled efficient parameter updating, resulting in learning trajectories and performance comparable to human participants. This demonstrates that engaging predictive mechanisms early, even under uncertainty, is essential for effective adaptation to dynamic environments. Together, these studies establish a process-based framework in which adaptive action control is neither purely reactive nor rigidly predictive. Instead, it emerges from coordinated mechanisms operating across multiple levels and timescales, with uncertainty guiding when internal models are updated incrementally and when they must be fundamentally restructured. By linking gaze behavior, subjective control beliefs, and predictive learning, this dissertation provides a foundation for investigating structural learning, proactive control, and failures of agency in both human and artificial systems.Item An intelligent X-ray assistant(2026) Mairhöfer, DominikRadiographs are the most commonly used modality in diagnostic imaging. Although they are of fundamental importance, there are constantly images acquired that cannot be used for diagnosis. Since the development of digital radiography, such unusable images have mainly been caused by incorrect patient positioning or improper collimation. Importantly, these errors can directly impact patients' health. Repeating the procedure increases radiation exposure, delays treatment, and extends hospital stays. Poor-quality images can even lead to misdiagnosis and incorrect treatment if unnoticed. Ultimately, images that cannot be used for diagnostic purposes also harm the hospital by increasing costs, staff workload, and room occupancy. To improve radiograph image quality, this thesis develops deep learning based methods for assistance systems in the radiography process. First, a developed framework is used to learn how to automatically assess the quality of radiographs. While radiologists can usually assess quality immediately, radiographers often come to different judgments. The automatic assessment system can provide support here and spare patients unnecessary repeat examinations. The results show that deep learning models are capable of assessing quality at the level of a radiologist across different parts of the body. Based on these results, instead of assessing radiographs, the quality of radiographs is predicted using depth images. While automatic assessment allows low-quality images to be detected immediately, it cannot directly prevent them from being produced. To achieve this, depth cameras were used to record the patient's pose, and neural networks were then used to predict the quality of the resulting radiograph. This type of positioning assessment makes it possible to acquire a radiograph only when a poor-quality image is unlikely to result. Although quality can only be determined on the radiographs themselves, the results show that a similarly accurate prediction is possible on depth images. In addition to positioning, the irradiation area is another key factor affecting radiograph quality. Therefore, a system is developed to automatically determine the required irradiation area. Depth cameras are used to capture images of the patients, while simultaneously acquired radiographs are used to label the minimally necessary irradiation area. The system learns to predict the required area automatically and demonstrates performance comparable to that of trained radiographers. Overall, this thesis presents approaches to solving the most common causes of low-quality radiographs today. It presents the first studies to systematically evaluate radiographs or patient poses in terms of their suitability for diagnosis.Item Die richterliche Bereitschaft zur Nutzung von KI-Systemen(2026) Dhungel, Anna-KatharinaItem Von der Leitlinie zur Anwendung(2026) Mersmann, StefanDer zunehmende gesellschaftliche und politische Druck auf unser Gesundheitssystem, stetig mehr Behandlungsqualität bei gleichzeitig sinkenden Behandlungskosten zu liefern, ist in den letzten Jahrzehnten rasant gestiegen. Dies betrifft auch und im Besonderen die geräte- und pflegeintensiven, medizinischen Disziplinen der Anästhesiologie. Die den täglichen Alltag bestimmenden klinischen Prozesse bieten ein enormes Optimierungspotenzial, wenn sie denn geeignet durch entsprechende Software-Anwendungen unterstützt werden. Mit dieser Arbeit wird ein generatives, flexibles Software-Framework vorgestellt und bewertet, das die Entwicklung, den Betrieb, die Wartung sowie die Wieder- und Weiterverwendung von Software-Anwendungen für eben solche klinischen Prozesse unterstützt. Im Fokus stehen diagnostische wie auch therapeutische Prozesse als medizinische Leitlinien im Anwendungsbereich der Anästhesiologie. Das Software-Framework selbst ist in Java implementiert und basiert auf einer hybriden, mehrschichtigen Systemarchitektur, die den Einsatz einer virtuellen Maschine unterstützt und damit hohe Flexibilität hinsichtlich Portierbarkeit, Erweiterbarkeit und Wiederverwendung gewährleistet. Es wird gezeigt, wie klinische Software-Anwendungen zur Entscheidungsunterstützung, halb- und vollautomatisierten Therapie spezifiziert, entworfen, konstruiert, realisiert, verifiziert und validiert und schließlich im hochregulatorischen Umfeld in Verkehr - also in den Markt und damit an den Patienten - gebracht werden können. Entlang eines hybriden Vorgehensmodells werden die zentralen Themen Methodik, Software-Entwicklung und Knowledge Engineering, Technologie, Produktqualifizierung, klinische Evaluierung sowie Zulassung, Inverkehrbringung, Betrieb und Ökonomie eingehend betrachtet und untersucht. Darüber hinaus ermöglicht die Reflektion auf neuere Ansätze und Technologien aus diesem Bereich aktuelle Ideen und praktische Empfehlungen zu Transfer und Verwertung in gegenwärtigen aber vor allem auch zukünftigen, industriellen Anwendungsszenarien wie auch wissenschaftlichen Forschungsinteressen. Die Untersuchung klassischer Themen, Methoden und Ansätze aus dem modernen Geschäftsprozessmanagement klärt, inwieweit Technologien aus diesem Bereich das Software-Framework zukunftssicherer gestalten können.Item Optimizing synthetic and real training data distributions for deep learning in image recognition(2026) Niemeijer, JoshuaThe recent advances in deep learning have enabled a large variety of applications. Among these are, for example, the environment perception of robots, including self-driving cars, and medical image analysis, which helps identify medical conditions or planning treatment. To build deep learning systems that generalize well, large quantities of relevant human-labeled data must be available for training. This requirement introduces several challenges. Annotations are costly due to the large amounts of data that need to be labeled and the complex nature of the annotation process. This is made more complex by the fact that relevant data needs to be recorded before data can be labeled. Depending on the field of application, this can be challenging. The challenge arises because relevant data is seldom available, which introduces the need to capture large quantities of data to find rare but critical cases. The work investigates a more efficient use of manual annotation through intelligent data selection for labeling, utilizing active learning (AL). In this context, semi-supervised learning (SSL), which aims to replace manual annotation, is utilized. The thesis investigates the use of synthetic data to replace the acquisition of data itself. The work presents strategies to guide the generation process towards creating rare but critical data. Finally, it is shown how to utilize these insights to create models that generalize well toward unseen distributions with minimal human intervention. For each of these methodologies, the thesis contributes novel approaches and analyses. It is shown that the choice of active learning approaches is highly dependent on the type of distribution the selection is performed on and the annotation budget. Next, the work shows how AL and semi-supervised learning are effectively integrated. This insight shows how to develop best practices for the application of AL and SSL. For the use of SSL in adapting networks to novel data domains, this work provides an extensive review of this dynamic field and derives novel low-complexity methods from it. These methods prove useful in their application to the environment perception of autonomous vehicles and the medical domain, as well as for adapting from synthetic to real data. The work provides novel methods for the targeted creation of synthetic data. Building on the creation of synthetic data and the research on SSL, the thesis presents an approach for generalizing to unseen domains. Overall, this thesis provides solutions for minimizing the cost and human effort involved in annotating and acquiring relevant data. The solutions provide efficient adaptation and generalization to new domains and distributions.Item Analysis of 3D/2D image registration and a new registration approach based on residual neural networks(2026) Schulz, Pia FranziskaImage registration is an essential component of image processing with a wide range of applications, particularly in the medical field. The goal of image registration is to determine optimal deformations that align two or more images. Our contributions to this field are twofold. First, we analyze 3D/2D registration problems. A particular challenge with such registration problems is that a 3D deformation is sought based on 2D data. Since typically regularization is utilized to ensure the existence of reasonable solutions, the question arises of which regularization is appropriate for 3D/2D registration. We address this question and prove the suitability of a class of second-order regularizers. We additionally prove that first-order regularizers are generally not appropriate. Furthermore, we show that our working assumptions apply for common settings of 3D/2D registration and we also extend our results to registration problems of other dimensions. Our analysis contributes to a more comprehensive understanding of image registration problems. Second, we present the new residual neural network-based registration approach RNR. This method enables the registration of multiple consecutive images. Moreover, the method ensures diffeomorphic deformations under certain conditions. This is a desirable property in many applications, as diffeomorphisms are invertible and sustain image features. We show that RNR is theoretically sound and provide a comprehensive validation. In particular, we demonstrate that the method allows for larger deformations than comparative approaches and has competitive speed. Finally, we apply the method to build a breathing model, which in turn forms the basis for respiratory surface electromyography modeling. The latter is a highly relevant field that is used, for example, to improve mechanical ventilation for patients.Item A second-order divergence-curl regularization in variational image registration(2026) Neuber, SaskiaItem Diskurse mit anderen Augen sehen – AR-gestützte Beteiligungssysteme in kommunalen politischen Diskursen(2025) König, FlorianDigitale Beteiligungsformate spielen eine immer wichtigere Rolle in der kommunalen Bürgerbeteiligung, indem sie Einwohner und Interessengruppen einbinden und lokales Wissen nutzen. Zentrale Herausforderungen bestehen darin, komplexe Themen verständlich aufzubereiten und Beteiligungsprozesse zugänglich zu gestalten. Visualisierungen – insbesondere Augmented Reality (AR) – bieten Potenzial, um räumliche und abstrakte Informationen greifbarer zu machen. Diese Arbeit untersucht, wie eine mobile AR-Anwendung gestaltet werden kann, um Bürger aktiv in Beteiligungsprozesse der Stadtplanung einzubeziehen. Dazu wurde eine mobile AR-App systematisch konzipiert, iterativ in einem menschenzentrierten Prozess entwickelt und hinsichtlich ihrer Gebrauchstauglichkeit evaluiert. In einer randomisierten kontrollierten Laborstudie wurde sie mit einem analogen Beteiligungswerkzeug verglichen, um zu analysieren, inwiefern sie das Verständnis räumlicher Informationen sowie das Einbringen eigener Vorschläge unterstützt und welche Auswirkung der AR-Einsatz auf den Diskurs, die wahrgenommene Arbeitsbelastung sowie die Beteiligungsbereitschaft hat. Abschließend wurden basierend auf den gewonnenen Erkenntnissen allgemeine Empfehlungen für die nutzerzentrierte Gestaltung und den nachhaltigen Einsatz von AR in Beteiligungsprojekten abgeleitet. Die Ergebnisse zeigen, dass die entwickelte AR-App Nutzende dabei unterstützt, sich mit Beteiligungsprojekten auseinanderzusetzen, räumliche Informationen zu erfassen und eigene Vorschläge einzubringen. Augmented Reality zeigt Potenzial, die Beteiligungsbereitschaft zu steigern und vielfältigere Vorschläge zu fördern. In der vorliegenden Untersuchung ließen sich jedoch keine signifikanten Unterschiede gegenüber einem analogen Werkzeug belegen. Dennoch liefert die Arbeit wertvolle Erkenntnisse zur Gestaltung gebrauchstauglicher AR-Anwendungen für die Bürgerbeteiligung. Sie verdeutlicht, dass AR das Potenzial hat, Beteiligungsprozesse zu bereichern, wenn technologische, nutzerzentrierte und organisatorische Faktoren optimal aufeinander abgestimmt werden. Damit trägt die Arbeit zur Forschung im Bereich Mensch-Computer-Interaktion (HCI) bei, indem sie nutzerzentrierte Gestaltungsprinzipien für AR-basierte Beteiligungswerkzeuge entwickelt, umsetzt und im Rahmen eines Mixed-Methods-Ansatzes empirisch evaluiert. Der Beitrag liegt insbesondere in der systematischen Erprobung, wie AR-basierte Beteiligungswerkzeuge gestaltet sein sollten, um Beteiligung verständlich, zugänglich und wirksam zu unterstützen. Die Ergebnisse bieten darüber hinaus Anknüpfungspunkte für zukünftige Studien zur digitalen Beteiligung.Item Analyzing the progression of pathologies in medical images(2025) Andresen, JuliaMedical image analysis is a key component of modern healthcare, required not only for diagnosis, but also for treatment planning and disease monitoring. The number of medical images acquired every day is constantly increasing and with it the need for automated tools to process, segment and interpret these images efficiently and reliably. Over the past decade, deep learning-based approaches, especially convolutional neural networks, have revolutionized the field providing unprecedented performances for almost all medical image analysis tasks, including semantic segmentation and image registration. However, the training of deep neural networks needs vast amounts of data, whereas most annotated medical datasets are small. The manual delineation of anatomical and pathological structures needs expert knowledge, and is both time-consuming and error-prone. These problems are even more severe in the analysis of disease progression, where not just one image but several have to be analyzed together. Furthermore, pathologies exhibit higher variability than anatomical structures and occupy comparatively small image areas, further increasing the data demands for training. This dissertation aims to develop deep learning-based algorithms for the automatic analysis of medical time series image data, focusing on pathological progression over time, such as retinal fluid in optical coherence tomography and brain lesions in magnetic resonance imaging. The main goal is to segment pathologies across all time points in order to monitor disease progression. Expert segmentations are typically unavailable for extensive time series data, requiring weakly supervised or fully unsupervised methods. Therefore, longitudinal registration of medical images is investigated as a tool for pathology tracking and unsupervised segmentation. To achieve the goals described, the present work follows three complementary research directions. First, unsupervised clustering is used to segment individual images. Second, registration-based approaches are developed for the joint analysis of longitudinal data with simultaneous segmentation of non-correspondences that reflect evolving or disappearing pathologies. Third, registration approaches inspired by metamorphosis models are used to model the formation of new pathologies. To improve the plausibility of the resulting deformations, these models are designed to separate displacements of anatomical structures from volumetric changes of the pathologies. The methods presented in this thesis enable the unsupervised segmentation of pathological structures, without relying on manually generated pathology segmentations. By leveraging weak supervision through anatomical labels and exploiting temporal information in longitudinal data, the proposed approaches can identify disease-related changes in an unsupervised manner. Overall, this work provides novel, annotation-efficient strategies for the automated analysis of medical image time series data, with the potential to support clinical workflows in the assessment of disease progression.Item Ambient Serious Games(2025) Brandl, Lea ChristineItem Force sensitive robotic ultrasound for diagnostics and therapy of peripheral arterial disease(2025) Osburg, JonasPeripheral arterial disease (PAD) is one of the most common cardiovascular diseases, affecting more than 230 million people worldwide. It is a form of peripheral atherosclerosis involving the narrowing or blockage of arteries, most commonly in the legs, due to plaque buildup. This reduces blood flow and can lead to tissue damage. The early diagnosis of PAD is crucial to prevent the progression of the disease, maintain the quality of life of patients, and avoid serious complications. The prevalence of PAD increases sharply with age, resulting in an increasing demand for pre- and post-examination of PAD, particularly in industrialized countries with an aging population. Ultrasound imaging (US) is most commonly used as a diagnostic tool, as it enables a fast, non-invasive, radiation-free, and painless examination of vessels. However, the time-consuming and complex nature of US, combined with the growing shortage of clinical staff, is expected to create a gap in its availability in the future. Thus, a force sensitive robotic US system was developed in this thesis, which will enable automatic vascular US examinations for diagnostic purposes, ensuring reliable patient care in the future. Several challenges in the development of such a system have been addressed. This included the investigation of preparatory aspects prior to the examination, such as optimal robot-patient positioning, as well as the safe guidance of the probe during the robotic scanning process. Firstly, the relationship between the geometry of the probe holder and the positioning of the robot’s base in relation to the patient was investigated. By determining adapted use-case specific probe holder geometries, optimal reachability of the anatomical region to be examined (e.g. the femoral artery) was ensured. The use of an appropriate probe holder geometry strongly increased the number of potential base placements with high reachability, especially in scenarios where obstacles restricted a part of the available space. Subsequently, a volunteer study was conducted to validate the feasibility of robotic US scans of leg arteries. An adapted interaction control scheme that also accounts for probe contact force was implemented to ensure patient safety. During the scan along the leg, the artery was automatically kept in the center of the US image for optimized image quality. Doppler ultrasound proved to be an effective solution for arterial tracking in this context, allowing for reliable differentiation between arteries and veins based on blood flow visualization. Moreover, additional specific features have been developed to further enhance the system. An approach to automatically adjust the orientation of the probe has been developed to enable automatic scanning of highly curved surfaces. Furthermore, the US gel application, which is essential for US examinations, was automated. Finally, the accuracy of the robot’s wrench estimation was investigated. Accurate wrench estimation of the robot is essential to ensure the safety of the system, especially in contact with the patient. Thus, a learning approach was proposed in which a neural network was trained on the dependencies between the joint torques in the robot’s axes and the forces and moments acting on the end effector. This strongly increased the accuracy and robustness of the wrench estimation model of the robot, in particular by reducing extreme outliers that occurred in the manufacturer’s model when the robot moved close to singularities. In summary, this thesis presents novel methods for the development of a robotic US system. The findings of this work represent a step towards future fully automated US examinations. Especially in the context of PAD diagnosis and monitoring, the system has the potential to considerably improve patient care through standardized and automated imaging.Item Generalizing deep Learning methods for volumetric medical image analysis(2025) Weihsbach, ChristianThe emergence of volumetric CT and MRI imaging technologies has dramatically improved clinical diagnostics and research, enabling visualization of body parts and organs in three dimensions. Deep learning, with its fundamental principles invented in the last century, has become a de facto standard for the automated processing of medical images, supporting clinicians in image interpretation and diagnosis. However, despite their widespread success, deep learning methods often achieve inferior results when applied in clinical practice compared to the training stage. This drop in performance is caused by the shifted properties of the images used during the deep learning models’ training and the images encountered later at the time of inference, combined with the models’ insufficient generalization capabilities. The shift in data properties to which the models fail to generalize may not be foreseen, and problematic image differences for the deep learning algorithms may be invisible to the human eye and not understandable by well-trained radiologists who can reliably diagnose patients’ conditions. In this thesis, four methods for volumetric medical imaging are presented that reliably generalize. It is researched in which areas and on which levels the generalization for volumetric medical images can be enabled and improved. The developed methods cover various fields of application, such as cardiac, abdominal, spinal, and brain volumetric medical imaging. Generalization was enabled by modeling acquisition processes for cardiac shape reconstruction, by effectively combining generalization and adaptation paradigms to overcome CT to MRI image intensity differences, by harnessing image registration in combination with loss-based modifications for generalizing segmentation of brain tumors across differently weighted MRI images, and by model parameter design modifications targeting the inner units of deep learning architecture to infer results from rotated or reflected input data reliably. All methods proved to work even for small-scale datasets with far less than one hundred samples, proving the efficiency of the methodological contributions as an alternative to following the trend of increasing dataset sizes and along with additional computational effort during training.Item Experimental investigation and validation of CFD simulations of steady flow in stenosis and pharynx using 2D PC-MRI and 4D flow MRI(2025) Gurumurthy, PragathiObstructive sleep apnea (OSA) is a sleep disorder of repetitive disrupted breathing caused by partial or complete closure of the upper airway, despite the effort to breathe. The sleep disorder not only causes social impact on the patient such as daytime sleepiness, fatigue but it has also been linked to several heart conditions. A combination of anatomical variations, impaired neuromuscular functions, ventilatory instability and premature awakening cause OSA. Due to the complex and heterogeneous nature of the disease, the etiology of OSA is not well understood. There are several invasive and non-invasive treatments available for the problem such as uvulopalatopharyngoplasty, maxillomandibular advancement, upper airway stimulation, use of continuous positive airway pressure and dental appliances. However, these have moderate to poor success rate. The identification of the factors contributing to OSA and development of cause driven treatments are not possible with the existing methods. Therefore, more recently numerical simulations or computational fluid dynamics (CFD)is being used to simulate physiological flow to observe the flow phenomena to help identify the problem causing OSA and derive an effective treatment plan. However, the results of the simulations are highly dependent on the mathematical model, boundary conditions, grid size and so on. Hence, a comparison of simulation results with experimental results is important to validate the accuracy of the simulation results. In-vitro phase-contrast magnetic resonance imaging (PC-MRI) based velocity measurements provides a powerful and non-invasive method to acquire spatially registered fluid velocity. This thesis proposes the use of 2D PC-MRI and 4D flow MRI as an investigative and validation tool for CFD of fluid flow in the upper airway during OSA In the current work, two models are chosen for investigation. One is an idealized rigid axisymmeteric stenosis model with 75% occlusion, which is a narrowing in the arteries resulting from plaque build up and also a simplified version of the occlusion occurring in the anatomically complex pharynx model. This model is primarily used to validate the MRI techniques using previously published laser doppler anemometry (LDA) data and also study the effects and progression of atherosclerosis. The second model is an anatomically accurate and OSA patient individual pharynx model to investigate the flow dynamics in the upper airway during OSA using the above validated 2D PC-MRI and 4D flow MRI. The results are used to understand the cause and effects of OSA. Both 2D PC-MRI and 4D flow MRI are used to measure the velocity in both the models at different boundary conditions. The stenosis model is investigated in laminar and turbulent flow condition. The pharynx model is studied at average inspirational and expiration flow rate. In a statistical framework the results of the velocity measurements in the stenosis and pharynx are compared with computational fluid dynamics (CFD) results to validate the numerical simulation results. Also, with the use of 4D flow MRI other pathophysiological parameters such as wall shear stress and recirculation patterns are quantitatively examined, validated with published data and compared with 2D PC-MRI and CFD data. The role of these parameters in atherosclerosis and OSA are also discussed.Item IT-Sicherheit in der Kritischen Infrastruktur BOS-Leitstelle(2025) Christiansen, JensItem Action regulation in energy-efficient driving(2025) Moll, Vivien EstherBattery electric vehicles (BEVs) offer substantial potential for reducing emissions but introduce cognitive and behavioural challenges for energy-efficient driving. In contrast to internal combustion engine vehicles (ICEVs), energy flow in BEVs is less tangible, and relevant consumption patterns are more complex to perceive, predict, and interpret. Current ecodriving research often lacks cognitive grounding, a focus on the specific challenges in BEVs, and a profound analysis beyond performance measures. This dissertation addresses the need for user-centred, cognitively aligned feedback by examining how different feedback approaches affect drivers’ perception, judgements, behaviour, knowledge, and perceived support of action regulation and the mental model of ecodriving. The theoretical foundation integrates adaptive control and action regulation models, cognitive information processing, and the role of mental models and perceived capability in goal-directed behaviour. It posits that energy-efficient driving with BEVs requires continuous situational adaptation and knowledge-based reasoning. Four empirical studies were conducted using experimental designs combined with qualitative and quantitative methods across diverse settings, including an online experiment, driving simulations, and real-world driving. Each study assessed both subjective and objective indicators of action regulation and knowledge. Study 1 (N = 55, online experiment) laid the conceptual foundation by exploring how drivers interpret typical consumption feedback derived from simplified acceleration dynamics. Rooted in bounded rationality, results revealed a systematic overestimation of energy use, particularly for high and brief maximum consumption values. There was no significant correlation between the correct energy efficiency ranking and the ranking derived from participants’ estimations. The study also identified interindividual differences in heuristic information processing, showing that both stimulus properties and cognitive predispositions shape perception. Study 2 (N = 63, driving simulator study) focused on knowledge gaps and their behavioural implications. It contrasted three feedback approaches: a baseline without support, a consumption trace display, and a recommendation system indicating optimal speed. Drivers frequently relied on incomplete or inaccurate conceptions of energy efficiency. While those using the recommendation system felt less uncertain, this confidence did not translate into better performance or more accurate knowledge. However, their tendency to verbalise more vehicle- and environment-related information suggests a more active reasoning process regarding energy-efficient driving. Study 3 (N = 50, field study) built on these findings and introduced a comprehension-based approach with pre-drive tip lists. When behavioural strategies were paired with technical reasoning, drivers reported higher perceived knowledge, stronger support for action regulation and the mental model, and better driving performance. This highlights the potential of explanation-based feedback to improve effectiveness, knowledge, and user experience. Study 4 (N = 112, driving simulator study) extended this approach into real-time driving by integrating elaborated auditory ecodriving tips into a recommendation system. This combined approach significantly improved driving performance and strengthened perceived mental model support, although cognitive load, information acquisition, and subjective information processing awareness were negatively influenced. The dissertation offers novel instruments and methods to evaluate ecodriving feedback. Key contributions include a new experimental paradigm for assessing dynamic magnitude perception, and two new constructs: perceived support of action regulation and perceived support of the mental model, enabling a finer-grained evaluation of action regulation quality beyond conventional usability or satisfaction metrics. Furthermore, existing items for measuring perceived ecodriving knowledge were revised based on theoretical considerations. Finally, an AI-assisted method was employed to systematically analyse verbalised driving strategies and their technical explanations, demonstrating scalable content analysis. Theoretically, the dissertation integrates psychological frameworks with an emphasis on mental models and information processing, provides a systematic literature review, and links various feedback approaches to cognitive processing and behavioural regulation. Moreover, it extends established cognitive biases by identifying a novel bias specific to dynamic data visualisation. Empirically, it demonstrates that comprehension-oriented feedback can improve energy-efficient behaviour, deepen understanding, and enhance perceived support, especially when it explains behavioural strategies and clarifies causal relationships. The practical implications are synthesised into design guidelines for future feedback systems in BEVs and beyond. The innovations in this dissertation extend beyond the context of BEVs. Action regulation in complex and dynamic systems—such as aviation, industrial control, or AI-assisted decision-making, especially in light of the growing role of generative, speech-based AI—can benefit from these findings. When users must form accurate mental models or interpret raw data in real-time, feedback should explain mechanisms and facilitate information analysis rather than merely presenting outcomes. This dissertation lays the groundwork for future research on cognitively aligned feedback systems that foster effective action regulation, adequate mental models, and user experience.Item Ein Hyperthermie-Einsatz zur Integration in ein präklinisches MPI-System mit Erweiterungsraum für Zusatzausstattung(2025-10-13) Behrends, AndréItem Altersgerechte Technikentwicklung(2025) Volkmann, TorbenDie allgegenwärtige und für jeden verständliche Interaktion mit digitalen Technologien stellt eine der zentralen Herausforderungen für die Mensch-Computer-Interaktion-Forschung dar. Dabei wird die Teilhabe aller Gesellschaftsgruppen zunehmend wichtiger, um von den Vorteilen der Digitalisierung profitieren zu können. Insbesondere ältere Erwachsene stehen oft vor Barrieren im Umgang mit digitalen Technologien, weshalb der Erwerb digitaler Kompetenzen und die Förderung lebenslangen Lernens entscheidend für ihre Teilhabe sind. Dies erfordert nicht nur technische Lösungen, sondern auch einen gesellschaftlichen Wandel, um die digitale Spaltung zu verringern und allen Bevölkerungsgruppen, insbesondere in einer alternden Gesellschaft, den Zugang zu digitalen Technologien zu erleichtern. Diese Arbeit zielt darauf ab, sowohl die digitale Teilhabe älterer Erwachsener als auch demokratische Prinzipien wie Inklusion und Gleichberechtigung im Entwicklungsprozess zu stärken. Indem ältere Erwachsene aktiv in die Entwicklung digitaler Technologien einbezogen werden, sollen ihre spezifischen Bedürfnisse und Präferenzen in die Gestaltung der Lösungen einfließen. Der partizipative Ansatz verringert die digitale Spaltung und fördert eine inklusive Gesellschaft, in der alle Bevölkerungsgruppen – unabhängig von Alter oder digitaler Vorerfahrung – gleichermaßen von der fortschreitenden Digitalisierung profitieren können. Ein konkretes Beispiel für die Anwendung dieses partizipativen Ansatzes ist das Historytelling-System, das es älteren Erwachsenen ermöglicht, ihre Lebensgeschichten digital festzuhalten und zu teilen. Um die Forschungsfrage zu beantworten, werden im Folgenden vier zentrale Ergebnisse präsentiert, die den Entwicklungsprozess und die Gestaltungsprinzipien des Systems darlegen. Erstens wird ein erweitertes Modell zur Technologieakzeptanz speziell für ältere Erwachsene präsentiert. Zweitens wird die Entwicklung von Gestaltungsrichtlinien vorgestellt, die altersbedingte Veränderungen berücksichtigen und im Historytelling-System Anwendung finden. Drittens wird ein agiler, partizipativer Technikentwicklungsprozess beschrieben, der die Entwicklung des Historytelling-Systems unterstützt. Viertens wird ein Reflexionsframework entwickelt, das die Akteure, Methoden und Ziele partizipativer Technikentwicklungsprozesse systematisch einordnet. Darauf aufbauend wurde ein Reflexionswerkzeug erstellt, mit dem die Methodendurchführungen der Historytelling-Systementwicklung eingeordnet wurden. Damit leistet diese Arbeit insgesamt einen wichtigen Beitrag zur Gestaltung inklusiver digitaler Technologien und bietet einen Ansatz, der die Teilhabe älterer Erwachsener fördert und gleichzeitig zur digitalen Inklusion in einer alternden Gesellschaft beiträgt.Item KI-gestützte Gewebeanalyse auf Basis der optischen Kohärenztomographie für die Tumorerkennung in der Neurochirurgie(2025) Strenge, PaulHirntumorerkrankungen stellen für Patient:innen und ihr Umfeld eine erhebliche Belastung dar. Der chirurgische Eingriff ist ein zentraler Bestandteil der Therapie, wobei das vollständige Entfernen von Tumorgewebe für das Überleben entscheidend ist. Gleichzeitig erschwert das diffuse Wachstum vieler Tumoren die intraoperative Abgrenzung von gesundem Gewebe, da etablierte Methoden wie MRT oder Fluoreszenzmikroskopie nur eingeschränkt zuverlässig sind. Die optische Kohärenztomographie (OCT) bietet eine kontaktfreie, nichtinvasive Bildgebung mit mikrometergenauer Auflösung und stellt eine vielversprechende Alternative dar. Diese Arbeit untersucht die OCT hinsichtlich ihrer Eignung zur Identifikation von Tumorgewebe und Infiltrationszonen. Grundlage ist ein weltweit einzigartiger Datensatz aus rund 700 pixelweise annotierten OCT-B-Scans, die im Rahmen einer klinischen Studie mit 21 Patient:innen ex-vivo während Resektionen aufgenommen wurden. Zwei OCT-Systeme mit unterschiedlichen Wellenlängen und Auflösungen kamen zum Einsatz. Histologische Schnittbilder wurden neuropathologisch annotiert und durch ein formbasiertes Verfahren auf korrespondierende OCT-B-Scans übertragen. Die Analyse begann mit einem Vergleich der Systeme anhand optischer Gewebeeigenschaften und einer binären Klassifikation zwischen gesundem und tumorösem Gewebe. Während keine signifikanten Unterschiede zwischen den Systemen erkennbar waren, konnte weiße Masse zuverlässig von stark infiltrierter weißer Masse (>60 %) unterschieden werden (Genauigkeit: 91 %). Graue Masse zeigte jedoch hohe Ähnlichkeiten mit Tumorgewebe, was die Genauigkeit bei Einbezug zusätzlicher Gewebetypen auf etwa 60 % reduzierte. Zur Verbesserung wurden strukturelle Eigenschaften einbezogen und sowohl klassische Methoden als auch maschinelles Lernen angewandt. Neuronale Netze ermöglichten eine Klassifikation in drei Klassen (weiße Masse, graue Masse, stark infiltrierte weiße Masse). Mit einem evidenzbasierten Lernansatz konnten Klassifikationsunsicherheiten quantifiziert werden. Für sichere Vorhersagen über alle Methoden hinweg ergaben sich eine Präzision und Sensitivität von jeweils 83 %. Die Ergebnisse belegen das Potenzial der OCT für die intraoperative Tumorerkennung und schaffen eine Grundlage für weitere klinische Forschung.