259594 research outputs found
Sort by
AI algorithm for lung adenocarcinoma pattern quantification (PATQUANT): international validation and advanced risk stratification superior to conventional grading
No full textThe morphological patterns of lung adenocarcinoma (LUAD) are recognized for their prognostic significance, with ongoing debate regarding the optimal grading strategy. This study aimed to develop a clinical-grade, fully quantitative, and automated tool for pattern classification/quantification (PATQUANT), to evaluate existing grading strategies, and determine the optimal grading system. PATQUANT was trained on a high-quality dataset, manually annotated by expert pathologists. Several independent test datasets and 13 expert pathologists were involved in validation. Five large, multinational cohorts of resectable LUAD (patient n = 1120) were analyzed concerning prognostic value. PATQUANT demonstrated excellent pattern segmentation/classification accuracy and outperformed 8 out of 13 pathologists. The prognostic study revealed a distinct prognostic profile for the complex glandular pattern. While all contemporary grading systems had prognostic value, the predominant pattern-based and simplified IASLC systems were superior. We propose and validate two new, fully explainable grading principles, providing fine-grained, statistically independent patient risk stratification. We developed a fully automated, robust AI tool for pattern analysis/quantification that surpasses the performance of experienced pathologists. Additionally, we demonstrate the excellent prognostic capabilities of two new grading approaches that outperform traditional grading methods. We make our extensive agreement dataset publicly available to advance the developments in the field
Die Rolle des purinergen Rezeptors P2X4 in der Atherosklerose
No full textDie Atherosklerose ist ein lipidgetriebener chronisch entzündlicher Prozess der Arterienwand, dessen Folgeerkrankungen wie Herzinfarkt und Schlaganfall enorme gesundheitliche wie gesellschaftliche Auswirkungen haben. Bei der Entwicklung anti-inflammatorischer Therapien gilt es, zielgerichtet pro-atherogene Signalwege zu beeinflussen, ohne dabei die Immunabwehr zu beeinträchtigen. Die Bindung von extrazellulärem ATP als Gefahrensignal an den purinergen Rezeptor P2X4 wurde in verschiedenen Pathologien als pro-inflammatorischer Signalweg identifiziert, sodass die Hypothese einer pro-atherogenen Rolle von P2X4 hier geprüft werden sollte. Zunächst erfolgte bei LDL-R-/- Mäusen unter 16-wöchiger hochdosierter Cholesterindiät (HCD) der Nachweis einer erhöhten P2X4-Expression in atherosklerotischen Plaques im Aortenbogen. Anschließend wurden P2X4-/- Mäuse mit LDL-R-/- Mäusen gekreuzt. Diese P2X4-/- LDL-R-/- Mäuse zeigten nach 16 Wochen HCD geringere atherosklerotische Läsionen in der Aortenwurzel, im Aortenbogen und in der Aorta descendens mit unveränderter histologischer Komposition im Vergleich zur P2X4+/+ LDL-R-/- Kontrollgruppe. Die atherosklerotischen Plaques P2X4-defizienter Mäuse wiesen eine signifikant niedrigere RNA-Expression der pro-inflammatorischen Zytokine CCL-2, CXCL-1, CXCL-2, IL-6 und TNF-α sowie ein vermindertes Priming des NLRP3-Inflammasoms auf. Die Inflammasom-Aktivierung innerhalb der Plaque zeigte sich jedoch anhand der aktiven Caspase-1 unbeeinträchtigt. In-vitro führte die Stimulation von Knochenmarksmakrophagen (BMDM) mit LPS und ATP zu einer verminderten Sekretion von CCL-2, CCL-5, IL-1β und IL-6, was auf einen Makrophagen-vermittelten Effekt von P2X4 in der atherosklerotischen Plaque schließen lässt. Im peripheren Blut P2X4-defizienter Mäuse wurde zudem ein niedrigerer Anteil pro-inflammatorischer Ly6Chigh-Monozyten und entsprechend höherer Anteil anti-inflammatorischer Ly6Clow-Monozyten beobachtet. Intravitalmikroskopisch zeigte sich nach intraperitonealer ATP-Stimulation reduziertes Leukozyten-Rollen an der Gefäßwand P2X4-defizienter Mäuse. Hierzu passend zeigten die Plaques von P2X4-/- LDL-R-/- Mäusen eine verminderte endotheliale Expression des Adhäsionsproteins VCAM-1, sodass auch eine pro-atherogene Funktion von P2X4 auf Endothelzellen möglich erscheint. Weiterhin fanden sich Unterschiede des T-Zell-Phänotyps im peripheren Blut von P2X4-/- Mäusen, die sich histologisch und in der RNA-Expression innerhalb der Plaque nicht widerspiegelten. Schließlich wurde auch in humanen atherosklerotischen Plaques eine erhöhte Expression des P2X4-Rezeptors und Kolokalisation mit Endothelzellen gefunden. Zusammenfassend konnte eine Reduktion der Atherosklerose durch globale Defizienz des P2X4-Rezeptors gezeigt werden, die sich durch reduzierte Expression pro-inflammatorischer Zytokine und Adhäsionsmoleküle erklären lässt. Somit ist P2X4 eine vielversprechende Zielstruktur für die Entwicklung zielgerichteter anti-inflammatorischer Therapien für die Atherosklerose
Machine learning for fast inline solar cell characterization and semantic compression of solar cell measurement images
No full textIn solar cell production, end-of-line characterization is crucial for quality assessment, typically including current-voltage (IV) measurements and imaging techniques. IV evaluation yields parameters like efficiency or fill factor, allowing for global performance assessment and cell sorting. Alongside IV, imaging techniques, such as electroluminescence, reveal spatial defects and inhomogeneities not captured by global measurements.This thesis addresses two challenges in both IV measurements and imaging techniques. First, the increasing solar cell demand poses practical challenges for conventional IV measurements. The time and mechanical stress involved in contacting the cell constrain throughput, thereby increasing costs per measurement, and risk cell damage. This intensifies with shingle cells, where measuring each individually is inefficient, leading to longer measurement time relative to energy produced. Second, the full potential of measurement images is currently unused. Unlike IV parameters, images are difficult to compare, complicating their analysis. They contain a wealth of information on cell properties, like material traits, defects, and process variations, which overlap in one image, making it hard to extract these nuances. Current methods use only part of this information.The first throughput-related challenge is addressed using convolutional neural networks (CNNs) to predict IV measurements, bypassing physical contact and individual shingle assessments. The IV estimations are based on two approaches: purely contactless measurements for overall cell assessment and full host cell images for shingle-specific analysis. For contactless IV, a CNN processes photoluminescence images at different excitations and with partial shading. The predicted contactless IV parameters and curve align with contacted reference measurements, mostly within measurement uncertainty. For shingle IV determination, a CNN assesses full host cell images, focusing on individual shingle regions. The shingle IV predictions closely match the reference measurements and outperform the current industry practice, which assigns global host cell measurements to all shingles, overlooking local variations. Both contactless and shingle IV predictions enable quality sorting, assessable via module output power simulations. Compared to industry standards, contactlessly sorted cells show no additional module mismatch loss versus contact-based sorting, and shingle sorting based on the model yields lower mismatch loss across all investigated quality classes compared to global host sorting.To address the second challenge, CNNs are developed to harness the rich information within images through semantic compression, introducing two approaches. First, multiple measurements are summarized within an empirical digital twin, a vector representation capturing essential cell properties condensed from high-dimensional inputs like images. For this, a model correlates these inputs with cell quality parameters, such as efficiency or fill factor, encapsulating related features. Cells represented by digital twins naturally cluster into groups with similar characteristics. Expert analysis can assign quality properties such as defects, overall quality, and manufacturing variations to identified clusters, allowing new cells to be categorized when their digital twin falls into one of these clusters. The second approach involves stepwise decoding of the semantic compression for one parameter to extract spatially resolved IV parameter maps, e.g. efficiency maps, of up to 256 x 256 pixels resolution. Compared to sophisticated yet slow reference measurements, the retrieved maps provide accurate but slightly less detailed results. However, by relying solely on inline measurements, they offer an alternative that balances accuracy and practicality, enabling fast local loss analysis concerning common IV parameters.Overall, this work underscores the potential of machine learning as an interdisciplinary tool in solar cell characterization. In line with the challenges identified, the potential is shown in two areas: On one hand, it can enhance current practices by integrating measurement images to improve established processes, such as IV characterization, achieving similar or improved outcomes more rapidly and without physical contact. On the other, it broadens analytical capabilities by systematically extracting quality information, such as empirical digital twins, from complex data not directly accessible with human capabilities
Redox biology at the intersection of physical activity and air pollution: mechanisms, consequences, and complexity
No full textModel complexity reduction in Bayesian sensor calibration and its relation to principal component analysis
No full textCalibration is a costly, necessary part of sensor manufacturing. Bayesian sensor calibration leverages prior knowledge in the form of a sample of sensor response functions from a sensor ensemble such as a fabrication lot. Based on a response model with M parameters, it allows to infer the measurand of interest and its predictive uncertainty from output signals of specimens of the ensemble after their lean calibration with a small number N of measurements, possibly with N<M . This article answers the question of whether it is possible to reduce model complexity by choosing models with parameter number M˜<M , while guaranteeing a required predictive accuracy. Reduced models are obtained by projections of a high-dimensional model, termed full model, onto a subspace of its parameter space. An I-optimality-based loss function implementing the Bayesian calibration approach is derived. For given N and M˜ , it allows to find an optimal reduced model and, at the same time, the optimal experimental design of the calibration. The approach is applied to 48 specimens of a CMOS Hall sensor system cross-sensitive to temperature and mechanical stress. Starting from an 11-parameter full model, reduced models with M˜=2,4 , and 5 are identified for calibration routines with N=2,4 , and 6 measurements, respectively. Despite the significantly smaller parameter numbers, the resulting root mean square (rms) predictive uncertainties of 102, 77.0, and 64.8 μ T, respectively, are increased by less than 1% from the full-model values. A general conclusion is that model order reduction (MOR) in the present Bayesian framework invariably entails an uncertainty increase, similar to principal component analysis (PCA)
DNVF Memorandum Partizipative Versorgungsforschung (Teil 1)
No full textPatients, as central actors in healthcare, should be enabled to actively participate in health services research processes. In addition, other stakeholders, such as professionals from healthcare practice, are also essential for a comprehensive participatory approach. This DNVF memorandum focuses on participatory approaches in the context of health services research. It begins by outlining the key characteristics of participatory health services research and describing its current development and institutionalization in Germany. The DNVF memorandum also highlights the potential and benefits of participatory research. Finally, it addresses two cross-cutting topics that are particularly relevant for further development in this field: the theoretical and conceptual foundations, and the investigation of effects and effectiveness of participatory approaches
