Scientific publications of the Saarland University
Not a member yet
14321 research outputs found
Sort by
Deep Learning-Based Diagnosis of Corneal Condition by Using Raw Optical Coherence Tomography Data
No full textBackground/Objectives: Keratoconus (KC) is the most common corneal ectasia. This
condition affects quality of vision, especially when it is progressive, and a timely and
stage-related treatment is mandatory. Therefore, early diagnosis is crucial to preserve
visual acuity. Medical data may be used either in their raw state or in a preprocessed
form. Software modifications introduced through updates may potentially affect
outcomes. Unlike preprocessed data, raw data preserve their original format across
software versions and provide a more consistent basis for clinical analysis. The objec tive of this study was to distinguish between healthy and KC corneas from raw optical
coherence tomography data by using a convolutional neural network. Methods: In to tal, 2737 eye examinations acquired with the Casia2 anterior-segment optical coherence
tomography (Tomey, Nagoya, Japan) were decided by three experienced ophthalmol ogists to belong to one of three classes: ‘normal’, ‘ectasia’, or ‘other disease’. Each
eye examination consisted of sixteen meridional slice images. The dataset included
744 examinations. DenseNet121, EfficientNet-B0, MobileNetV3-Large and ResNet18
were modified for use as convolutional neural networks for prediction. All reported
metric values were rounded to four decimal places. Results: The overall accuracy for
the modified DenseNet121, modified EfficientNet-B0, modified MobileNetV3-Large
and modified ResNet18 is 91.27%, 91.27%, 92.86% and 89.68%, respectively. The
macro-averaged sensitivity, macro-averaged specificity, macro-averaged Positive Pre dictive Value and macro-averaged F1 score for the modified DenseNet121, modified
EfficientNet-B0, modified MobileNetV3-Large and modified ResNet18 are reported
as 91.27%, 91.27%, 92.86% and 89.68%; 95.63%, 95.63%, 96.43% and 94.84%; 91.58%
91.65%, 92.91% and 90.24%; and 91.35%, 91.29%, 92.85% and 89.81%, respectively.
Conclusions: The successful use of a convolutional neural network with raw optical
coherence tomography data demonstrates the potential of raw data to be used instead
of preprocessed data for diagnosing KC in ophthalmology
Practical Test-Time Domain Adaptation for Industrial Condition Monitoring by Leveraging Normal-Class Data
No full textMachine learning has driven significant advancements across diverse domains. However,
models often experience performance degradation when applied to data distributions that
differ from those encountered during training, a challenge known as domain shift. This
issue is particularly relevant in industrial condition monitoring, where data originate from
heterogeneous sensors operating under varying conditions, hardware configurations, or
environments. Domain adaptation is a well-known method to address this problem; how ever, the proposed methods are not directly applicable in real-world condition monitoring
scenarios. This study addresses such challenges by introducing a Normal-Class Test-Time
Domain Adaptation (NC-TTDA) framework tailored for condition monitoring applications.
The proposed framework detects distributional shifts in sensor data and adapts pretrained
models to new operating conditions by exploiting readily available normal-class samples,
without requiring labeled target data. Furthermore, it integrates seamlessly with automated
machine learning (AutoML) workflows to support hyperparameter optimization, model
selection, and test-time adaptation within an end-to-end pipeline. Experiments conducted
on six publicly available condition monitoring datasets demonstrate that the proposed ap proach achieves robust generalization under domain shift, yielding average AUROC scores
above 99% and low false positive rates across all target domains. This work emphasizes
the need for practical solutions to address domain adaptation in condition monitoring and
highlights the effectiveness of NC-TTDA for real-world industrial monitoring applications
National bias in international sports judging: a scoping review
No full textThe subjective nature of performance evaluation in aesthetic sports competitions makes
it vulnerable to bias. Among these, national bias—where judges’ scores are influenced
by their own and the athlete’s nationality—poses a significant threat to the integrity of
competition outcomes. Existing literature is fragmented across sports, types of bias,
theoretical explanations, and methodological approaches, and lacks an overarching
synthesis. To address this gap, the present scoping review—conducted in accordance
with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)
2000 guidelines—synthesizes existing literature across four dimensions: (1) forms of
national bias, (2) underlying mechanisms, (3) degree of intentionality, and (4) proposed
mitigation strategies. Contributions addressing at least one of these dimensions were
included regardless of type or quality.The review identifies national bias as a multifaceted
phenomenon that manifests in several forms, including favoritism toward compatriots,
penalization of their competitors, vote trading among judges, and reactive scoring
based on perceived or expected national bias from colleagues. Vote trading reflects
intentional manipulation, whereas other forms may arise unintentionally through
cognitive or social mechanisms, complicating their detection and regulation. Existing
countermeasures demonstrate limited effectiveness and often entail trade-offs,
underscoring the need for more context-sensitive and robust interventions
Business process representation learning
No full textCompanies generate vast amounts of digital trace data, stored in event logs, when performing business processes. Such data is rich in information, describing what actions have been performed, by whom, at what time, and in what context. Process mining (PM), a data-driven discipline in the field of business process management (BPM), deals with detailed analysis of event data on a process instance basis, aiming to obtain operational insights into how companies’ processes have actually been performed. It has evolved as a major field of research, offering approaches for automatic event log analysis to enhance efficiency or ensure compliance, with growing industrial adoption. As PM applications increasingly require machine learning (ML) techniques, particularly for forward-facing support of business processes, the question becomes how to apply such techniques to event log data, since the characteristics of event logs and the processes they describe differ from common data modalities such as images and text. Given the complexity of event log data, applying ML techniques poses significant challenges, requiring adaptations of existing or the development of customized ML techniques to account for the characteristics of event logs. Representation learning, a research field in ML, deals with learning representations from data that make solving downstream tasks more effective and efficient by overcoming the laborious and expensive task of manual feature engineering. This thesis introduces the problem of business process representation learning, i.e., learning representations from event logs for solving BPM tasks like process prediction, anomaly detection, or task abstraction. By developing, analyzing, and applying various process representation models (RPMs), this thesis contributes to three research areas. First, on how to design PRMs in terms of architecture and training procedures to learn representations that follow the general priors of representation learning. Second, on the assessment of how well the models have learned characteristics of event log data and the underlying processes. Finally, on the application side, for which BPM tasks PRMs can be used, and how well they perform. The results demonstrate that PRMs are capable of learning accurate, context-specific representations from different concepts within event logs. These representations adhere to the principles of representation learning and can be utilized to solve real-world BPM tasks efficiently. Thereby, the results advance the fields of PM and ML, and especially their interplay
Reporting on Deliverable D1.4 – Delivering the teaching script to the online methods workshop on collected data and all applied methods to measure the concept of political trust and legitimacy in Europe
No full textHorizon Europ
The Use of Podcasts as a Learning Activity During a Year 5 Competency-Based Blended Learning Curriculum at Saarland University
No full text(1) Podcasts are increasingly used in undergraduate medical education. They differ from
traditional learning activities and may influence exam performance. Podcasts also offer
insights into learning behaviour and perceptions of family medicine (FM). Despite their
frequent use in medical education, it remains unclear how they can best be integrated into
competency-based curricula and motivate students to study for FM. This study examines
the impact of a medical podcast on learning behaviour and academic performance at Saar land University (UdS). (2) This exploratory mixed-methods study analyzed podcast-related
learning behaviour and exam relevance among year-five medical students at UdS in the
winter semester 2024/25. Demographic, quantitative, and qualitative data were collected
via an online questionnaire (Google Forms®) in January 2025. Data were descriptively and
analytically evaluated and linked to exam results. Qualitative data were analyzed using
Kuckartz’s content analysis. (3) Of 123 eligible students, 92 participated. Most listened to
episodes in full. Podcasts were seen as low-threshold means to access study content, but
they were often not perceived as a separate learning activity. Listening to podcasts did not
directly influence exam performance but helped connecting theory with clinical relevance
and increased motivation for FM. (4) Podcasts are popular for exploring clinical practice
and complex topics. Their didactic value lies in contextual learning and career orientation,
rather than improving exam performance
Vorhersagbarkeit der postoperativen Schmerzhaftigkeit durch präoperative digitale Fingerdruckmessungen an einem Patientenkollektiv zu proktochirurgischen Eingriffen
No full textPostoperative Schmerzen bleiben ein bedeutendes Problem im Bereich der medizinischen Versorgung. Die Identifizierung von Risikofaktoren sowie die Entwicklung geeigneter instrumenteller Ansätze zur Vorhersage postoperativer Schmerzen könnten das postoperative Schmerzmanagement erheblich verbessern. Solche Fortschritte würden nicht nur Komplikationen reduzieren, sondern auch die Lebensqualität und Zufriedenheit der Patienten steigern und gleichzeitig die Behandlungskosten senken.
Laut der Literatur kann die generalisierte Schmerzempfindlichkeit durch quantitative sensorische Tests (QST) beurteilt werden. Diese nicht-invasive Methode misst die Reaktionen der Teilnehmer auf kontrollierte externe Reize, wie Vibration, Druck oder Temperatur, und dokumentiert systematisch Veränderungen im Nervensystem, insbesondere im nozizeptiven System. Dadurch lassen sich individuelle Schmerzempfindlichkeit präzise bestimmen. Ein Zusammenhang zwischen QST und postoperative Schmerzen bestehen darin, dass Patienten mit erhöhter Schmerzempfindlichkeit möglicherweise ein höheres Risiko für postoperative Schmerzen aufweisen.
Die Anwendung quantitativer sensorisches Tests zur Vorhersage des postoperativen Schmerzempfindens in der anorektalen Chirurgie wurde in der bestehenden Literatur bislang wenig beobachtet und erforscht.
Im Rahmen dieser Untersuchung wurde ein elektronisches Testverfahren (J-Tech Commander Algometer Baseline 1200-304 (Push-Pull Force Gauge®) zur Messung der perioperativen Fingerdruckschmerzsensibilität bei 150 proktochirurgischen Patienten etabliert. Das Spektrum der Eingriffe umfasste Hämorrhoiden, Analfisteln, Analtumoren, Anal- und Mukosaprolapse, Analabszesse und Analfissuren. Die Schmerzsensibilität der Beeren des linken und rechten Kleinfingers wurde getestet, wobei jeder Finger dreimal untersucht wurde. Bei allen Patienten wurde der Schmerzlevel mittels VAS präoperativ sowie an den Tagen 1 und 3 nach der Operation, sowie während der Wiedereinbestellung in Woche 4 erfragt und eine Algometertestung vorgenommen.
In dieser Studie wurde ein signifikanter Zusammenhang zwischen der präoperativen Fingerdruckmessung und der frühen postoperativen Schmerzsensibilität festgestellt. Die Ergebnisse zeigen, dass eine höhere Druckschmerztoleranz vor der Operation mit geringeren Schmerzen der Patienten am ersten und dritten postoperativen Tag verbunden ist. Diese Erkenntnisse sind besonders bemerkenswert, da sie darauf hinweisen, dass präoperative Bewertungen wertvolle Informationen über den postoperativen Schmerzverlauf liefern können. Der statisch signifikante Zusammenhang (p <0,0001) verdeutlicht die Relevanz dieser Messungen für die Schmerztherapie. Interessanterweise verschwand dieser Effekt nach vier Wochen, was darauf hinweist, dass andere Faktoren die Schmerzwahrnehmung im späteren Verlauf beeinflussen könnten. Insgesamt zeigt die Studie, dass die präoperative Fingerdruckmessung ein nützliches Instrument zur Prognose der postoperative Schmerzbewältigung darstellt und zur gezielten Optimierung der Schmerztherapie beitragen kann.Postoperative pain remains a significant issue in the field of medical care. Identifying risk factors and developing appropriate instrumental approaches to predict postoperative pain could greatly enhance pain management after surgery. Such advancements would not only reduce complications but also improve patients' quality of life and satisfaction while simultaneously lowering treatment costs.
According to the literature, generalized pain sensitivity can be assessed through quantitative sensory testing (QST). This non-invasive method measures participants' responses to controlled external stimuli such as vibration, pressure, or temperature, and systematically documents changes in the nervous system, particularly in the nociceptive system. This allows for precise determination of individual pain sensitivity. A connection between increased pain sensitivity and postoperative pain suggests that patients with heightened sensitivity may have a higher risk of experiencing postoperative pain. However, the application of quantitative sensory testing to predict postoperative pain in anorectal surgery has been minimally observed and researched in existing literature.
As part of this study, an electronic testing procedure using the Tech Commander Algometer Baseline 1200-304 (Push-Pull Force Gauge) was established to measure perioperative finger pressure pain sensitivity in 150 patients undergoing proctological surgery. The range of procedures included hemorrhoids, anal fistulas, anal tumors, anal and mucosal prolapse, anal abscesses, and anal fissures. The pain sensitivity of the distal pads of the left and right little fingers was tested, with each finger being examined three times. The pain level was assessed preoperatively as well as on days 1 and 3 postoperatively, and during follow-up in week 4, using the Visual Analog Scale (VAS), in addition to conducting algometer testing.
In this study, a significant correlation was found between preoperative finger pressure measurement and early postoperative pain sensitivity. The results indicate that higher pressure pain tolerance before surgery is associated with lower pain levels in patients on the first and third postoperative days. These findings are particularly noteworthy as they suggest that preoperative assessments can provide valuable information about the postoperative pain trajectory. The statistically significant correlation (p < 0.0001) highlights the relevance of these measurements for pain management. Interestingly, this effect disappeared after four weeks, suggesting that other factors may influence pain perception in the later stages. Overall, the study demonstrates that preoperative finger pressure measurement is a useful tool for predicting postoperative pain management and can contribute to the targeted optimization of pain therapy
Morphological characterization of 3D cell cultures generated by liquid overlay technique
No full textCultivating cells in 3D is considered a significant advancement in cell culture models, as it better
reflects natural cellular environments compared to 2D cultures. However, analytical methods like
standard light microscopy are less effective for 3D cultures. In this study, 3D cell cultures were
generated using the liquid overlay technique with 10,000, 50,000, 100,000 and 200,000 Normal
Human Dermal Fibroblasts, analyzed on days 1, 2, and 3 post-seeding. We quantified the influence
of fixation with paraformaldehyde or glutardialdehyde/dehydration on their morphology
compared to living 3D cell cultures. They were analyzed by light microscopy, scanning electron
microscopy as well as by digital light microscopy (height profile measurement). Over time, the
cultures decreased in size, likely due to cell shrinkage and structural reorganization. The size
reduction could be mathematically described by an exponential decay function. The proportion of
round spheroids versus indented aggregates depended on cell number, culture age, and fixation
method. On day 1, cultures seeded with 10,000 cells formed nearly 100% round spheroids,
regardless of fixation. Higher cell numbers led to fewer round spheroids, and fixation further
reduced their number. This suggests that large cell quantities sediment in layers due to steric
hindrance, forming indentations. Since aldehydes are responsible for cross-linking proteins, we
hypothesize that this chemical reaction, combined with low stability of the 3D cell cultures, leads
to the increased formation of the indented 3D cell aggregates. This is consistent with an overall
increase in the number of round spheroids and a decrease of the negative influence of fixation
over time. In summary, it is important to consider the number of seeded cells, the incubation time,
as well as the possible fixation effects when generating stable spheroids using the liquid overlay
technique for down-stream experiments
What patients with proximal humerus fractures really want and what commonly used outcome scores measure
No full textBackground: This study aims (1) to identify patient-reported questionnaire items, independent of age or gender, that reflect healthy
shoulder function and treatment satisfaction in patients with proximal humerus fracture (PHF), and (2) to compare these items and
their weighted importance with items measured by the most frequently used outcome measures.
Methods: Patients who sustained a PHF from June 2016 to September 2021 were surveyed with a 29-item questionnaire on their per ceptions of item importance as a measure of shoulder function and treatment outcome. Items were generated from the following
outcome measures: American Shoulder and Elbow Surgeons score, Constant Score, Neer Score, Oxford shoulder Score (OSS), Quick DASH, and the University of California, Los Angeles shoulder rating score. A mean difference of at least 10% between gender and age
groups (<60 vs. 60 years) was defined as clinically significant. Items that were rated as at least 90% important without a clinically and
statistically significant mean difference between the groups were defined as essential items.
Results: One hundred forty-six patients with mean age 60.8 years (range 20-92 years) completed the questionnaires. Only 6 out of 29
items were identified as essential items. These include: being pain-free, being able to sleep/having no pain in bed at night, using a knife
and a fork at the same time, putting on a coat/to dress, managing toileting, washing under both arms. None of the scoring systems
covered all these items with appropriate weighting of scoring points. The OSS most closely covered patient interest with the most
appropriate weighting of points.
Conclusion: We identified 6 items from daily life that are of essential importance for patient-reported healthy shoulder function and
treatment satisfaction regardless of age and gender. Until a reliable and valid scoring system for PHF is developed that includes
these items, we recommend using the OSS, as it most closely reflects patient-reported interests