742875 research outputs found
Sort by
Graph Representation Learning in Smart Environments:Sensor Embeddings with Graph Neural Networks
No full textThis research focuses on learning effective representations of graph-structured data to improve inference in smart environments. While state-of-the-art deep learning models are highly successful on grids or sequences, graphs pose unique challenges due to their arbitrary size, irregular structure, and lack of an explicit notion of locality. These characteristics make it nontrivial to directly apply conventional architectures, such as convolutional neural networks, to graph-based data. The thesis is organized into two parts, each focusing on a different domain, graph topology, and prediction task. Part I investigates Human Activity Recognition (HAR) using wearable and mobile sensors, where the graph topology is not explicitly given and must be inferred from data. In this setting, HAR is framed as a graph classification problem. This part highlights how conventional HAR methodologies can lead to misleading and impractical results, motivating the use of unbiased evaluation strategies. It then proposes a contrastive learning approach to capture global and local dependencies based on multiple graph constructions, leading to significant improvements in classification accuracy. Part II focuses on Water Distribution Networks (WDNs), where the graph topology is explicitly defined by the network layout. The task is formulated as a node-level regression problem aimed at reconstructing pressure signals across the network from sparsely located sensors. A graph-based model is introduced, along with robust training and evaluation strategies, resulting in a new state-of-the-art approach. Finally, it is discussed how these advances can support the development of Graph Foundation Models tailored to WDNs, outlining key challenges and strategies for their realization
Compressed Sparse Regression for Anchored Design of Experiments and Sensor Placement in Structure Health Monitoring
No full textThis study investigates sensor placement for condition monitoring in complex systems, focusing on capturing dominant dynamic responses that indicate abnormal conditions. Traditional sensor placement methods often rely on costly distributed sensors and heuristic strategies, which are not efficient in capturing the most informative response characteristics. To address these challenges, a data-driven Design of Experiment (DoE) approach is proposed, leveraging system science principles to optimize sensor allocation systematically. The implementation of this framework is formulated as a sparse regression problem, enabling an efficient selection of sensor locations that maximize information gain while minimizing redundancy. To solve this problem, a newly developed Compressed Orthogonalized Least Squares (Comp-OLS) algorithm is introduced. In order to validate the proposed approach, a case study on the DoE of a Duffing system is conducted. Compared with the commonly used Pivoting QR Factorization (PQRF) method, the results demonstrate that the Comp-OLS-based framework significantly enhances sensor placement efficiency, ensuring comprehensive coverage of system dynamics while anchoring the locations of required sensors. This study demonstrates the potential of data-driven DoE for improving condition monitoring in various engineering applications, offering a scalable and effective solution for sensor placement challenges.</p
Identifying Non-linear Output Frequency Response Functions Using Generalized Associated Linear Equations with Recursive and Coupled Computational Methods
No full textThe non-linear output frequency response functions (NOFRF), as an extension of the linear frequency response function (FRF) in the non-linear case, has been applied to weakly non-linear system study and engineering structural health monitoring (SHM). The computation of NOFRFs requires first solving a series of linear ordinary difference equations, i.e., generalized associated linear equations (GALEs), and then obtaining the system's results of each order according to the definition of NOFRFs in the frequency domain. However, in practical applications, the solution of GALEs often requires the aid of numerical integration. Therefore, accurate numerical computation of GALE is the first task in system analysis using NOFRFs. In our study, two different numerical methods are proposed for solving the system of linear differential equations of GALEs. The first computational method involves solving the GALEs of each order using a Recursive Computational Method (RCM). The second approach transforms the problem of solving GALEs into state-space equations, which are then solved using the integral solver of numerical computation software (e.g., MATLAB). This method is referred to as the coupled computational method (CCM). Finally, we compare the results of the two methods for computing NOFRFs using a non-linear differential equation (NDE) model with a fourth-order nonlinear term as an example. The final results show that the two methods give consistent results for low order NOFRFs. However, for higher order NOFRFs, CCM produces more accurate results than RCM. This provides ideas for calculating NOFRFs by GALE in nonlinear systems and also provides an important theoretical basis for calculating NOFRFs in multiple-input multiple-output (MIMO) systems.</p
Patterns of lymphatic spread in hypopharyngeal squamous cell carcinoma – Findings from a multicenter study
No full textIntroduction: Aiming for personalization of the elective nodal irradiation (ENI) in hypopharyngeal squamous cell carcinoma (SCC) patients, we describe the regional lymphatic spread patterns and risk of lymph node metastases, considering not only T-stage, location and lateralization of the primary tumor, but also involvement of adjacent lymph node levels (LNLs). Materials and methods: Patients with newly diagnosed hypopharyngeal SCC diagnosed at University Hospital Zurich between 2013–2021, UMCG Groningen between 2006–2023 and University Medical Center Freiburg between 2011–2019 were analyzed. Lymphatic involvement per level was assessed based on imaging and, if available, pathology. The dataset is made publicly available and can be visualized on https://lyprox.org/. Results: 390 patients with hypopharyngeal SCC were included, 81 % had one or more cervical lymph node metastases. Overall prevalence of involvement in LNLs II, III, IV, V was consistent with literature: ipsilateral 65 %, 54 %, 23 %, 11 %; contralateral 25 %, 16 %, 6 %, 3 %. For lateralized tumors not affecting the midline (N = 143), contralateral involvement was 11 %, 4 %, 1 % 1 %. When contralateral LNL II was negative (N = 291), involvement of downstream LNLs III, IV, V was 5 %, 3 %, 1 %. Ipsilateral LNL IV involvement was reduced to 7 % in patients with negative LNL II and III. Ipsilateral level I and VII involvement was 6 % and 13 % in T4-tumors, but only 2 % and 3 % in T1–T3 tumors. Conclusion: We provide detailed information about lymphatic spread patterns of hypopharyngeal SCC, where subgroups of patients may be identified in whom the ENI may be reduced. For lateralized tumors, contralateral irradiation may be limited to LNL II in patients without contralateral involvement.</p
Integrated geriatric assessment and intervention in the head and neck oncology care pathway reduces adverse events and does not affect survival
No full textObjective: The number of older/frail patients with head and neck cancer (HNC) is increasing. They are more frail compared to patients with other malignancies. Therefore, geriatric care is increasingly integrated into the HNC care pathway. The aim of this study was to investigate how integrated geriatric care affects treatment outcomes in HNC patients irrespective of treatment intention.Methods: This retrospective study compared treatment-related adverse outcomes (grade ≥ 2 Clavien-Dindo surgical complications and grade ≥ 2 CTCAE (chemo)radiotoxicity), and one-year mortality in two patient cohorts. In the first cohort (2014–2016), geriatric screening was only observational i.e. without intervention. In the second cohort (2019–2020), a complete geriatric pathway was integrated into the oncological care pathway, including an onco-geriatric MDT (multidisciplinary team meeting), referral to the geriatrician with intervention, if needed. Multivariable logistic regression analysis was performed to identify factors associated with adverse events and one-year mortality, including the cohort period.Results: This study included 640 patients; 369 from the first cohort and 271 from the second cohort. The second cohort showed significantly fewer adverse events (34.6 %) compared to the first (65.4 %) (OR 0.41; 95 % CI 0.27–0.63: p < 0.001). Reductions were seen in surgical complications (OR 0.57; 95 % CI 0.32–1.01) as well as (chemo)radiotoxicity (OR 0.39; 95 % CI 0.20–0.76). No significant differences were observed in one-year mortality (OR 0.88; CI 0.59–1.48). Adverse events were significantly linked to malnutrition, advanced tumor stage and concomitant radiotherapy treatment.Conclusion: Integration of geriatric care in the HNC pathway reduces treatment-related adverse events, without altering one-year mortality.</p
The actual performance of ML/AI models in predicting radiation-induced toxicity in head and neck cancer:a systematic review and meta-analysis
No full textAn increasing number of Artificial intelligence (AI) and machine learning (ML) models are being developed to predict radiation-induced toxicities (RITs) in patients with head and neck cancer (HNC). But their performance and reliability remain uncertain. This systematic review and meta-analysis evaluated the predictive accuracy and methodological quality of these models. We comprehensively searched PubMed, EMBASE, Web of Science, and the Cochrane Library to identify studies reporting on ML/AI models for predicting RITs in HNC patients. Eligible studies were assessed for bias risk using the PROBAST tool, and key performance metrics, including the area under the receiver operating curve (AUROC), were extracted. A hierarchical multilevel meta-analysis was performed to estimate pooled AUROC values, and subgroup analyses explored the influence of study characteristics on model performance. A total of 67 studies with a total of 568 models were included, showing moderate discriminatory power of ML/AI models, with a pooled AUROC = 0.76; 95 % CI: 0.73–0.78. Nonetheless, substantial heterogeneity was observed across studies. Incorporating imaging biomarkers significantly improved model performance. Prospective and internal validation showed comparable performance; external validation shows true generalizability. The predominance of retrospective designs and variability in predictor selection may have introduced bias, affecting model reliability and generalisability. ML/AI models hold promise for predicting RITs in HNC patients, but methodological constraints limit their applicability. Standardised and transparent reporting of model development and validation processes is vital for improving comparability among studies. Future research should explore hybrid modelling methods and the integration of clinical, dosimetric, radiomic, and genomic data to boost predictive accuracy.</p
ΔNp73 isoform defines a TP53-mutant-like poor-risk subgroup of acute myeloid leukemia
No full textAmong acute myeloid leukemia (AML) patients, a subgroup remains notoriously refractory to current treatment options, with underlying mechanisms poorly understood. Here, using a multi-omics approach, we reveal that this resistant patient subgroup is characterized by high expression of the oncogenic TP73 isoform ΔNp73, exhibiting similarly poor outcomes as TP53-mutant AML. ΔNp73, which lacks a transcriptional activation domain but retains chromatin-binding properties, competes with TP53 for specific gene targets, thereby downregulating TP53 signaling. We demonstrate that the transcription factor CEBPA controls ΔNp73 expression in AML cells by binding to an intragenic enhancer region. Genetic or pharmacological inhibition of the transcriptional activity of CEBPA with guanfacine reduces ΔNp73 levels and restores drug sensitivity involving ferroptosis-mediated cell death, acting synergistically with venetoclax. Our study sheds light on a previously undercharacterized poor-risk subgroup of AML, which may support patient stratification and inform treatment considerations.</p
A validated CT-based scoring system for lateral compression type one pelvic ring injuries provides insight into the spectrum of injury severity and guides treatment decisions; a prospective study
No full textPURPOSE: To gain insight into the spectrum of injury severity in lateral compression type 1 (LC1) pelvic ring injuries using a validated CT-based scoring system and to determine how injury severity relates to treatment and clinical outcomes.METHODS: A prospective study was performed in 203 patients presenting with LC1 injuries at a level one trauma center. CT-scans were assessed using a CT-based scoring system that quantifies injury severity on a scale of 5–14. Patients were categorized into three injury severity groups: low (scores 5–6), intermediate (scores 7–9), and high (scores 10–14) subgroups based on level of sacral displacement, Denis classification, sacral column involvement, inferior ramus displacement, and superior ramus fracture location. Clinical outcomes included delayed intervention due to mal- of non-union, Dutch Short Musculoskeletal Function Assessment and EuroQol-5D 5L at one-year follow-up. Normative data was used to determine recovery.RESULTS: All patients with low scores (n = 36), 94% of intermediate scores (n = 99), and 71% of high scores (n = 44) were treated conservatively. No conservatively treated patients required delayed intervention. In all subgroups, most recovered to the level of the normative data, with no significant differences in outcomes between operatively and conservatively treated patients.CONCLUSIONS: The LC1 CT-based scoring system provides insight into the spectrum of injury severity and helps guide treatment decisions for LC1 injuries. Patients with low and intermediate injury severity, determined by degree of sacral and pubic rami involvement, can be treated nonoperatively. Those with high injury severity can be treated either conservatively or operatively.SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1007/s00590-025-04619-4.</p
REBELS-IFU:Dust build-up in massive galaxies at redshift 7
No full textIn recent years, observations with the James Webb Space Telescope (JWST) have started to map out the rapid metal enrichment of the early Universe, while (sub)millimetre observations have simultaneously begun to reveal the ubiquity of dust beyond z ≳ 6. However, the pathways that led to the assembly of early dust reservoirs remain poorly quantified, and require pushing our understanding of key scaling relations between dust, gas, and metals into the early Universe. We investigate the dust build-up in twelve 6.5 ≲ z ≲ 7.7 galaxies drawn from the Reionization Era Bright Emission Line Survey (REBELS) that benefit from (i) JWST/NIRSpec strong-line metallicity measurements, (ii) Atacama Large Millimetre/submillimetre Array (ALMA) [Cii]-based redshifts and gas masses, and (iii) dust masses from single- or multi-band ALMA continuum observations. Combining these measurements, we investigate the dust-to-gas (DtG), dust-to-metal (DtM), and dust-to-stellar mass (DtS) ratios of our sample as a function of metallicity. While our analysis is limited by systematic uncertainties related to the [Cii]-to-H2 conversion factor and dust temperature, we explore a wide range of possible values, and carefully assess their impact on our results. Under a fiducial set of assumptions, we find an average log(DtG) = -3.02 ± 0.23, only slightly below that of local metal-rich galaxies. On the other hand, at fixed metallicity our average log(DtS) = -2.15 ± 0.42 is significantly larger than that of low-redshift galaxies. Finally, through a comparison to various theoretical models of high-redshift dust production, we find that assembling the dust reservoirs in massive galaxies at z ≈ 7 likely requires the combination of rapid supernova enrichment and efficient interstellar medium dust growth.</p
Pre-adolescent learners' foreign language classroom anxiety profiles and correlates:Insights from Chinese primary school students of English
No full textThis paper reports on a study investigating the level and correlates of foreign language classroom anxiety among pre-adolescent students. The participants were 385 L1 Chinese primary school students of L2 English, aged between 8 and 13 (with a mean age of 10.73), who completed a validated English Classroom Anxiety Scale and a questionnaire tapping: (1) three learner-centered predictor variables (i.e., gender; attitudes towards English, and perceived relative standing among peers in English proficiency; The participants’ age was provided by their parents or caregivers) and (2) six teacher-centered predictors (i.e., attitudes towards the English teacher; teacher strictness, friendliness, joking, and predictability; and the frequency of the teacher’s English usage in class). Data analysis showed that the participants generally experienced a moderately low level of English classroom anxiety. English classroom anxiety showed no significant difference among Years 3 to 5 participants but significantly decreased in Year 6. Girls and boys did not differ significantly in their English classroom anxiety levels. Attitudes towards English, attitudes towards the English teacher, perceived relative standing among peers in English proficiency, and age significantly negatively predicted English classroom anxiety, in descending order of magnitude. Teacher friendliness and the teacher’s frequency of English usage in class significantly and negatively predicted English classroom anxiety but only marginally so. Three variables under consideration, teacher joking, strictness, and predictability, were not significant predictors of English classroom anxiety. The results and their (pedagogical) implications are discussed and the limitations of this study are put forward