OOPS - Oldenburger Online-Publikations-Server (Univ. Oldenburg)
Not a member yet
4628 research outputs found
Sort by
Abnormality prediction and forecasting of laboratory values from electrocardiogram signals using multimodal deep learning
No full textData-Driven State-of-Health Quantification for Industrial Proton Exchange Membrane (PEM) Water Electrolyzer Fleets
Full text linkProducing green hydrogen through water electrolysis using renewable electricity is key to decarbonizing industries. Proton exchange membrane (PEM) water electrolysis is a mature technology with advantages such as allowing for fast dynamic operations. However, its adoption is limited by the high cost and short lifetime. To address this, operators must maximize utilization by optimizing operations and maintenance, which requires monitoring the state of health (SOH). Existing methods rely on laboratory measurements under controlled conditions, unsuitable for industrial use. This research focuses on voltage as an easily measurable degradation indicator and uses data-driven methods to correct the measured voltage to predefined reference conditions, denoted as Urc – the SOH indicator. Two methods are proposed: one based on transfer linear regression, adjusting voltage models with new data; and another using Bayesian inference, with fleet knowledge as prior. These methods are validated with both industrial and synthetic data. They enable automatic SOH monitoring, enhance safety and profitability, and provide insights into degradation behavior
Gesundheitsbezogene Lebensqualität von Kindern, Jugendlichen und Erwachsenen nach der Diagnosestellung eines Meningeoms im Kindes- und Jugendalter (MEyLIFE)
Full text linkMeningeome sind im Kindes- und Jugendalter selten. Es existieren kaum Daten zur gesundheitsbezogenen Lebensqualität (Health-Related Quality of Life, HRQoL) von pädiatrischen Meningeompatient*innen. In einer Querschnittsstudie wurde nun die HRQoL von 35 dieser Patient*innen nach beendeter Therapie mittels Fragebögen erhoben. Dies fand im Median 10 Jahre nach Diagnose statt. Minderjährige Teilnehmende erreichten nach dem PedsQL Fragebogen eine mediane HRQoL von 86 und adulte Teilnehmende nach dem EORTC QLQ-C30 Fragebogen von 75. Erwachsene mit NF2 gaben eine niedrigere HRQoL an. Ein Rezidiv/Progress ging mit einer höheren HRQoL einher. Das Alter bei Diagnose, der WHO-Grad, Tumorreste, Strahlentherapie, die Tumorlokalisation, die Tumorentität, der Grad der Behinderung und die Zeit seit Diagnose zeigten keinen Einfluss. Diese Studie bietet Hinweise, dass pädiatrische Meningeompatient*innen nach abgeschlossener Therapie eine mit der Allgemeinbevölkerung vergleichbare HRQoL aufweisen
Comparison of robustness, resilience and intrinsic capacity including prediction of long-term adverse health outcomes: The KORA-Age study
No full textBackground: Frailty, resilience and intrinsic capacity (IC) are concepts to evaluate older person`s health status, but no comparison of their associations with adverse health outcomes exists. We therefore aimed to assess which concept is most useful for determining long-term health of older adults.
Methods:Analyses were based on the KORA (Cooperative Health Research in the Region of Augsburg)-Age study (n = 940, 65–93 years). Frailty was evaluated using the physical frailty-phenotype by Fried et al. For comparability to resilience and IC, we chose the protective concept of robustness instead of frailty in the present analysis. Resilience was measured by the 11-item resilience-scale. The IC-score was based on 4 domains (locomotion, cognition, vitality and psychiatric capacities). Associations with falls, disability, and hospitalization at 3-year and 7-year follow-up and with mortality were evaluated by multivariable adjusted logistic and Cox regression. Concept overlaps were illustrated by a Venn-diagram.
Results: In the fully adjusted models, robustness showed significant inverse associations with most outcomes (3-year follow-up: OR (95%CI): disability 0.448 (0.300−0.668), 7-year follow-up: falls 0.477 (0.298−0.764), hospitalization 0.547 (0.349−0.856), and all-cause mortality 0.649 (0.460−0.915)) while resilience and IC showed significant inverse associations with disability only (e.g., 7-year-follow-up: resilience: 0.467 (0.304−0.716), IC: 0.510 (0.329−0.793)). 23% of the participants met the criteria for both robustness and IC while 22% met those for robustness and resilience.
Conclusion: Robustness was the most useful concept, showing the strongest protective associations for most adverse health outcomes. IC and resilience showed their main strengths in capturing protective associations for disabilities. Robustness overlapped with resilience and IC, supporting the concept of mind-body-interaction
Salience in Musical Scene Analysis: Psychoacoustic Experiments and Models
Full text linkThis dissertation explores the psychoacoustic origins of auditory salience in musical scenes—the ability of sounds to attract attention. The first study shows that some instruments draw more attention than others, with prior knowledge about the instrument’s sound mitigating this effect. While bass instruments lack salience, vocals stand out as particularly salient, coining the term ‘vocal salience.’ The second study identifies frequency micro-modulations in vocals as crucial for salience. The third study reveals that vocal sounds don’t possess salience automatically, but require unique acoustic factors, such as spectral distinctiveness. The final study reveals a perceptual bias toward the spectral edges of an auditory scene, contrasting with the lack of salience observed in bass instruments. These studies highlight the interplay of acoustic properties and cognitive factors that enable auditory salience in musical scenes
Die Bedeutung umfassender Molekulardiagnostik mittels Hybrid Capture Next Generation Sequencing beim nicht-squamösen NSCLC
Full text linkDer Fortschritt der molekularpathologischen Charakterisierung von Lungenkarzinomen hat zu einer deutlichen Verbesserung der Behandlungsoptionen geführt. Vor allem Adenokarzinome der Lunge sind häufig durch somatische Treibermutationen gekennzeichnet und können somit gezielt mit Tyrosinkinaseinhibitoren (TKI) behandelt werden. Auch neue Biomarker wie die Tumormutationslast (TMB) können durch die technologische Weiterentwicklung analysiert werden und wegweisend für die Behandlung sein. Um eine vollumfängliche Diagnostik und somit ein optimales Therapiemanagement zu gewährleisten, bietet sich das Next Generation Sequencing (NGS), welches sowohl für Gewebe- als auch Flüssigbiopsien etabliert ist, an. Ziel dieser Arbeit ist, die NGS-basierte Mutationsanalytik im Routinealltag zu evaluieren. Dabei werden sowohl praktische Aspekte, wie die Dauer vom Materialeingang bis zum Befund, als auch der Informationszugewinn auf molekularer Ebene und dessen Auswirkung auf die Therapie erläutert. Das TMB wird mit anderen molekularen Markern korreliert und das Resistenzspektrum von Tumoren in Abhängigkeit der TKI-Therapiesequenz untersucht
S4Sleep: elucidating the design space of deep-learning-based sleep stage classification models
No full textOptimal hedging strategies in robust market models under capital constraint
Full text linkThis thesis explores optimal hedging strategies under capital constraints within a robust market modelling framework, where uncertainty is addressed by considering multiple model measures. It introduces an indifference curve of optimal strategies, capturing trade-offs between different market assumptions. All strategies along this curve are derived using worst-case martingale measures, whose continuity properties are proven to ease computation. The thesis also tackles practical interpretation challenges, especially with non-equivalent measures, by proposing a modelling approach using pushforward measures, and analyzes when this approach aligns with the original results
Decomposition of Stochastic Surplus Processes in Life Insurance
Full text linkSeveral surplus decomposition formulas have been presented in the actuarial literature. However, all contributions use heuristic arguments. A comprehensive decomposition principle that allows existing decomposition formulas to be compared and modern risks to be added is still missing. The thesis closes that gap by introducing a so-called infinitesimal sequential updating (ISU) decomposition principle. The ISU decomposition principle improves the sequential updating (SU) decomposition principle by eliminating its order effects while retaining the desired additivity. The plausibility of the ISU decomposition principle is demonstrated by replicating the surplus decompositions known from the actuarial literature. In addition, the application of the ISU decomposition principle to martingales reveals its great potential in risk management. In particular, conditions are presented under which the ISU decomposition coincides with the martingale representation theorem (MRT) decomposition. Finally, evidence for the numerical feasibility of the ISU decomposition principle is provided using multilevel Monte Carlo methods
Neural Bug Detection
Full text linkSoftware bugs cost developers and companies significant time and money. To help developers find bugs early in the development process, neural bug detectors have been proposed. Neural bug detectors learn from millions of examples to find novel bugs in code. Although effective for simple bugs, they still have many limitations that make them difficult to use in practice. This thesis addresses two key challenges. First, prior work often relied on artificial mutants for training, which do not reflect real bugs. We propose a contextual mutator and mine public repositories for real bug fixes to create more realistic training data. Our evaluation shows that training with this data improves detection of real bugs. Second, current detectors often lack sufficient context, leading to false alarms. We propose an LLM-based validator that leverages extra context to reduce false alarms. Together, these contributions result in a neural bug detector that is significantly more accurate and practical for real-world use