Stuttgart University of Applied Sciences

Abstract Breast cancer is one of the most common cancers in women worldwide. Determination of the tumor margins is essential for complete resection of the tumors while preserving healthy tissue. To improve the differentiation between healthy and malignant tissue, a hyperspectral measurement setup with a broad spectral range was developed to detect characteristic spectral differences by measuring ex vivo surgical specimens. The results described here, are obtained from an ex vivo study including 19 patients, in which for each patient one tissue sample from the tumor and one healthy site are examined. The obtained spectra show promising spectral differences, which allowed the tissue to be differentiated from a test group of four patients not involved in the training with a classification accuracy of approximately 85 % using only spectral information without taking into account the spatial structure of the tissue. Comparable results of approximately 80 % and 86 % were also achieved with an evaluation reduced to four and six spectral channels, respectively. It was thus demonstrated that tumor and normal tissue can be differentiated with few spectral channels in the visible to near-infrared spectral range, enabling the development of a real-time capable multispectral measurement system for intraoperative use.

This study examines eye-tracking sensor data from 121 participants interacting with six websites to evaluate the predictive value of different eye movement features for perceived usability and user experience (UX). In total, over 36 hours of eye-tracking data, recorded at 250 Hz and totaling 32 million gaze points, were transformed into second-order eye movements events and then into third- and fourth-order machine learning features. Usability and UX ratings, collected through short versions of the User Experience Questionnaire (UEQ) and AttrakDiff questionnaires, served as labels for training six machine learning models. Results show that spatio-temporal metrics, particularly saccade sequences and Area of Interest (AOI) transitions, are the most predictive. Models trained on fourth-order features consistently outperform those based only on aggregated third-order metrics, highlighting the importance of preserving scanpath structure and sequential gaze behavior for predictive performance in eye-tracking-based Human-Computer Interaction (HCI) research.

This study examines whether combining conventional textbook materials with the app-based ultrasound simulator Scanbooster enhances learning outcomes in fetal echocardiography, compared to using either method alone. It also evaluates the impact of the learning sequence.210 medical students were randomized into two groups: · PDF → App group (n=103): Textbook study followed by the Scanbooster ultrasound simulator app. · App → PDF group (n=107): Scanbooster ultrasound simulator app followed by textbook study. Knowledge was assessed before and after by identifying marked structures in ultrasound videos.There were no significant differences between the groups regarding age (p=0.564) and prior ultrasound experience (p=0.746). Both groups showed a significant improvement in identifying sonographic structures after the learning phase (p<0.001), with a mean learning effect of 104.96% and 102.16%, respectively, for fetal echocardiography-specific knowledge. The order of learning materials had no significant effect on learning success (p>0.05). In comparison with our previous study, we found that the combination of app-based simulation and textbooks led to the best learning outcomes in fetal echocardiography. This approach resulted in significantly higher performance compared to pure PDF learning (p<0.001) and was slightly, yet still significantly better than pure simulation (p=0.031).The combination of conventional textbook materials and app-based ultrasound simulation is the most effective approach for teaching fetal echocardiography. The learning sequence does not significantly impact the outcome.

Accurate segmentation of retinal vessel images is critical for diagnosing ophthalmic diseases. However, retinal vessel segmentation is challenging due to the scarcity of training samples, retinal vessel morphology polymorphism, and the low contrast with the background. A novel Adaptive Feature Modulation and Implicit Prompt Network (AFMIP-Net) based on the encoder-decoder structure is proposed to address these issues. To improve the segmentation accuracy with complex background and different scale of blood vessels, the modulating feature maps from the high-level semantic layer to the low-level semantic layer in the encoder are considered. To improve the capillary segmentation, the multi-scale implicit prompt is extracted and applied to each decoder layer by suppressing redundant features with fusion of the encoding and decoding information. Moreover, data enhancement with the Gaussian filter is adopted to solve the issue of training sample scarcity, which improves the focus on difficult-to-segment complex vessels with interference of lesions and further enhances generalization. Experimental results on DRIVE, CHASE, STARE datasets show that the F1 value of our model is optimal with all three datasets. Also, its sensitivity and accuracy performance stand out across most datasets. The visual comparison further validates the superiority of our method in identifying retinal vessel topological shapes and micro-vessels. It is concluded that the overall performance of AFMIP-Net is significantly improved compared to other state-of-the-art methods.

The Stratospheric Observatory For Infrared Astronomy (SOFIA) was a special-mission aircraft, which was built, commissioned, and operated between 1999 and 2022. Its heart consisted of a 2.5 metres-diameter mirror telescope having been integrated into a Boeing 747SP jet to make infrared measurements of a wide range of astronomical objects. The telescope specifications required an inertial pointing stability of 0.2 arcsec rms during observation flights, while the telescope was exposed to the atmosphere at an altitude of 45.000 ft in an open cavity of the aircraft. To achieve this demanding performance, three low-noise fiber-optic gyroscopes, the Photonetics FOG180, were the core component of the telescope pointing control system.The FOG180s were not only vital telescope components but represented also a milestone in the development of fiber-optic gyroscopes (FOGs). Employing technologies like closed-loop operation, all-digital processing, Erbium-Doped fiber sources, and fiber Bragg grating, the company Photonetics (later iXsea and iXblue, now part of Exail) was able to take a leap from tactical grade gyroscopes beyond the range of navigation grade sensors and opened reliable gyrocompassing to FOGs. The FOG180 embodied this achievement and was the key to a whole family of gyroscopes and inertial measurement units with applications in the marine and space areas up to the level of strategic grade.SOFIA was decommissioned in fall 2022. This opened the opportunity to remove the FOG180s for retesting with a legacy of more than 7800 flight hours in the environment of cosmic radiation, dry air, and the high magnetic field of the telescope attitude drive. The paper presents results of these tests (noise, in run bias, thermal sensitivity, orthogonality), compares them with the outcome of former tests, and illustrates the notable long-term performance of these technologically important gyroscopes.

The core goal of Industry 4.0 is to build smarter, more adaptive, and interoperable manufacturing systems. Therefore, the concept of Asset Administration Shell (AAS) was proposed as a standardized interface and carrier for the digital expression of all objects that have value for an organization. Through years of development, AAS has become the standard for advanced development and application in the manufacturing industry. Predictive maintenance (PdM), as an important strategy in Industry 4.0, can detect early signs of failure through real-time data, optimizing equipment availability, reducing downtime and maintenance cost. However, current PdM implementations often suffer from data fragmentation, isolated modeling, limited generalizability, etc. Although AAS has great potential to effectively address these challenges, research in this area remains limited. This paper proposes a predictive maintenance method that integrates AAS and anomaly detection techniques. The method leverages the idea of proxy task by training a Long Short-Term Memory (LSTM) network on time-series sensor data and constructing anomaly scores using cosine distance metrics to effectively assess equipment conditions and provide predictive alerts based on it. In addition, the AAS models and corresponding submodels were designed and implemented with AASX Package Explorer, integrated with the PdM method using FA<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>ST tool. Finally, an industrial testbed was used as a practical case to verify the effectiveness and practicality of the proposed model.

Accurately locating object instances and generating precise segmentation masks remains a major challenge in complex, real-world environments due to occlusions, cluttered scenes, and scale variations. In this work, we propose FRISNET, a fast and real-time instance segmentation network that effectively fuses frequency-domain and spatial-domain representations to enhance both semantic understanding and spatial precision. Building upon the YOLACT framework-the current speed benchmark in singlestage instance segmentation-we incorporate a 2D Fast Fourier Transform (FFT)-based module to extract and isolate low- and high-frequency components from feature maps. These frequency-domain features are then reintegrated with spatialdomain representations to capture both global context and finegrained details. Additionally, we enhance the standard Feature Pyramid Network (FPN) by introducing a novel bottom-up pathway and same-level skip connections, which reinforce multi-scale feature representation and preserve high-resolution spatial information. FRISNET is trained and evaluated on two widely used benchmarks: PASCAL Segmentation Boundary Detection (PASCAL SBD) and Microsoft COCO. Experimental results demonstrate that FRISNET achieves a mean Average Precision (mAP) of 34.5 at 31.8 FPS on MS COCO, outperforming YOLACT++ with a ResNet-50 backbone by $1.17 \% \mathrm{mAP}$. Its superior accuracy-speed trade-off makes FRISNET particularly suitable for time-critical applications, such as robotic motion planning and tactile-sensor-assisted object grasping in unstructured environments.

Long-duration spaceflight imposes significant physiological stress on astronauts, including profound alterations in immune function. This study investigated epigenetic changes in immune cells following prolonged orbital spaceflight by analysing histone modifications in CD4+ and CD8+ T-cells from astronauts before, immediately after, and during recovery from spaceflight. Using Cleavage Under Targets and Tagmentation (Cut&Tag) to assess H3K27ac modifications, we identified significant alterations in chromatin accessibility, predominantly involving immune response pathways, gene regulation, and cellular adaptation mechanisms. While some epigenetic changes were transient, others persisted beyond 50 days post-return, suggesting long-term effects. These findings enhance our understanding of immune adaptation to spaceflight and have implications for mitigating spaceflight-associated health risks. Furthermore, they provide valuable insights into immune system regulation under high-stress conditions, potentially informing research on immunodeficiency disorders, cancer epigenetics, and aging-related immune decline on Earth. This study underscores the critical role of epigenetics in long-term space missions and terrestrial health applications.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-17930-1.

HighlightsWhat are the main findings?Sub-wavelength-sized alumina ceramic spheres are characterized with good accuracy with respect to their size and relative permittivity from scattering measurements at millimeter-wave frequencies.The characterization procedure is explained in detail and systematic error sources are analyzed.What are the implications of the main findings?Sub-resonant scattering measurements allow the characterization of relatively smaller particles with respect to traditional resonance-based methods.Alternatively, for a given particle size, sub-resonant characterization allows use of a lower operating frequency and, thus, simpler equipment.When measuring the size or relative permittivity of a dielectric particle, usually one of the parameters needs to be known for determining the other one. Scattering measurement methods offer an alternative that allows for extracting both the size and the permittivity of the particle under test at the same time. In this paper, bi-static scattering measurements at millimeter-wave frequencies are applied to characterize sub-resonant dielectric spheres of sub-wavelength size. The size and relative permittivity are extracted simultaneously from measurements at Ka-band (26.5–40 GHz). The experimental setup and the data processing procedure are detailed in depth, and the sources of the systematic errors are discussed. Alumina ceramic spheres (with relative permittivity of approximately 10) of diameter as small as 0.8 mm (less than 1/10 of free-space wavelength) were investigated. The extracted diameters and permittivities agree well with the expected values.