- Preprint Article
- 10.21203/rs.3.rs-8936009/v1
Fungal melanin induces host lipid defense by rewiring macrophage metabolism
- Mar 04, 2026
- Research Square
- Georgios Chamilos + 13 more +13
Publications from 2021 to 2026
Showing 10 of 183 papers
Fungal melanin induces host lipid defense by rewiring macrophage metabolism
Prospective Multicenter Validation of Machine Learning Models for Mortality Prediction in Adult Critically Ill Patients using Transfer Learning
<title>Abstract</title> Mortality prediction in critically ill patients remains challenging due to poor cross-institutional performance and limited generalizability of machine learning models. This study addresses this, by systematically benchmarking and prospectively validating transfer learning frameworks. We trained our models on MIMIC-IV and validated them on a multicenter prospective cohort of 539 patients from three hospitals. We compared tree-based methods and modern deep learning architectures for tabular data. Results demonstrated that both Domain Adaptation (DA) and Inductive Transfer Learning (ITL) significantly enhanced model performance under realistic conditions where target-domain data are limited. DA consistently improved discrimination across all evaluated models, with LightGBM showing the most significant gains in Area Under the Receiver Operating Characteristic Curve (AUC) (p = 0.0010), and XGBoost yielding the largest improvements in Area Under the Precision-Recall Curve (AUPRC) (p = 0.0419). Among all evaluated models, Random Forest (RF) achieved the highest discriminative performance, achieving 90.7% AUC with DA and 81.3% AUPRC with ITL. Notably, the domain-adapted models significantly outperformed APACHE II (p = 0.0044) and SOFA (p = 0.0077). These findings suggest that transfer learning provides a robust and data-efficient pathway for improving model generalizability across heterogeneous populations, offering a pragmatic solution to the challenge of model degradation in clinical deployment.
Read moreState of Health Estimation in Lithium-Ion Batteries: Methods, Comparisons, and Future Directions
<div class="section abstract"><div class="htmlview paragraph">As the world is moving towards electric vehicles, we are observing a wide use of Lithium-Ion batteries in modern transportation. Lithium-Ion Batteries offer several advantages over conventional battery systems, including higher energy density that is energy stored per unit mass, longer Cycle Life, faster Charging rates, low Self-Discharge, lighter weight, and ease of maintenance as the memory effect present in other batteries is absent. However, despite these advantages, the system faces significant technical challenges arising from inaccurate battery State of Health (SOH) estimation techniques. These inaccuracies can lead to unexpected vehicle failures and a degraded end-user experience, especially due to incorrect “distance to empty” predictions. In this paper, different SOH estimation techniques are reviewed and compared in detail. The SOH estimation approaches are broadly classified into three main categories: Model based estimation techniques, data driven estimation techniques, and fusion technology typically involving the combination of multiple estimation methods). This review highlights the strengths and limitations of each technique, offering a comparative analysis that enables researchers and engineers to select the most suitable approach based on system requirements and application constraints. Additionally, this paper emphasizes the importance of reliable SOH estimation in enhancing the safety, longevity, and overall performance of battery-powered systems, and discusses potential future directions for developing more accurate, adaptive, and real-time SOH estimation frameworks. A robust SOH framework can reduce warranty costs for manufacturers, prevent thermal runaways by timely identifying degradation patterns, and improve user trust in E-vehicle technology.</div></div>
Read moreTotem-Pole PFC and LLC Resonant Converter based 6.6kW On-Board Charger for LEV Applications
<div class="section abstract"><div class="htmlview paragraph">As light electric vehicles (LEVs) gain popularity, the development of efficient and compact on-board chargers (OBCs) has become a critical area of focus in power electronics. Conventional AC-DC topologies often face challenges, including high inrush currents during startup, which can stress components and affect system reliability. Furthermore, DC-DC converters often have a limited soft-switching range under light load conditions, leading to increased switching losses and reduced efficiency. This paper proposes a novel 6.6 kW on-board charger architecture comprising a bridgeless totem-pole power factor correction (PFC) stage and an isolated LLC resonant DC-DC converter. The main contribution lies in the specific focus on enhancing startup behavior and switching performance. In PFC converters, limiting inrush current during startup is crucial, especially with fast-switching wide-bandgap devices like SiC or GaN. Conventional soft-start techniques fall short in of ensuring smooth voltage transitions. Moreover, maintaining stable operation across a universal input voltage range and achieving a high-power factor under light load conditions remain persistent challenges. Although resonant converters are widely used for their natural soft-switching ability, achieving zero voltage switching (ZVS) over a wide range of loads, especially at light load conditions, is still a technical challenge. Existing solutions rely on complex control strategies or hardware modifications, which increase cost and design complexity. The proposed architecture was modeled and simulated using MATLAB/Simscape to assess dynamic and steady-state behavior under a range of operating conditions. Results demonstrated high input power factor, line/load regulation, and switch-node waveforms to confirm ZVS operation. Additionally, the proposed charger exhibits low harmonic distortion, ensuring compliance with IEC 61000-3-2 power quality standards. These findings confirm the topology’s effectiveness for high-performance LEV charging and set a strong foundation for future experimental validation and hardware development.</div></div>
Read moreAn Emotion-Aware Conversational Agent for Symptom and Quality-of-Life Monitoring in Hematologic Malignancies
Effectiveness of osteopathic treatment in women with primary dysmenorrhea: A randomised controlled trial.
Effectiveness of chain-mail mat for improving actuators efficiency in the pin-and-bed adaptive moulding
Meta Domain Adaptation Approach for Multi-Domain Ranking
In a real industry recommendation system, the distribution of recommended domains is very redundant. Different domains may address the same problem, such as the Click-Through Rate (CTR) prediction, and may share the same features. Multi-domain learning aims to use one model to complete the same task of all domains, in order to improve the model effect and achieve savings in computing resources and manpower. However, different domains have different data distributions and insufficient training data. In constructing a multi-domain model encompassing several domains with small sample sizes, how to make the best use of the existing data of other domains and ensure the generalization performance of the model in the training process has become the key problem of the research. To meet the challenge, in this paper, we propose a Meta Domain Adaptation model for Multi-Domain ranking (MDAMD), specifically designed for CTR prediction across multiple domains. It employs the unsupervised domain adaptation method to ensure the global unity of training for model optimization among all domains. Maximum Mean Discrepancy (MMD) loss calculation is added after the gate unit to ensure adaptive training of the model in feature extraction. Meanwhile, MDAMD applies a meta-learning method to avoid overfitting in the domain adaptation stage. Results from the industrial and public datasets demonstrate the effectiveness of MDAMD. Moreover, our model has been deployed online and obtained the CTR and Gross Merchandise Volume (GMV) improvements.
Read moreLightweight Privacy Architecture for Homomorphic Operations in e-Health Systems
In recent years, traditional hospital facilities are adopting new technologies that provide novel services. Nevertheless, these services may be performed by third parties, that pose a threat to data privacy. As health data is sensitive, it must be constantly encrypted from the moment it is collected. Thus, an appropriate privacy scheme that is suitable for the resource constrained e-Health environment is required. In this paper, a lightweight privacy architecture, applied for homomorphic operations in IoT systems for e-Health is proposed. The focus is its easy implementation in devices with resource limitations, while also enhancing privacy. Two security versions of the same architecture are presented, with different parameters. Each proposed version achieves a minimum of 70.27% and 49.2% resource reduction compared to published related works. The proposed compact version achieves a minimum of 19.7% area-delay reduction compared to published designs. The main reasons for these results are the employment of synthesizable operations, the fusion of functions and the pipeline architecture of operations.
Read moreLight Exposure on Alertness after Wake-Up in Healthy Men: Comparing Dim, Bright, Red, and Blue Light
Introduction: Light is a key factor in moderating human alertness, both subjective and objective. However, the methodology applies in research on the effects of exposure to light of different wavelengths and intensities on objective and subjective alertness varies greatly and evidence on objective alertness in particular is still inconclusive. Thus, the present, highly standardized within-subject laboratory study on N = 44 healthy males explored how LED light of different intensities (dim vs. bright light) and wavelengths (red vs. blue) affected objective (reaction time/RT) as well as subjective (sleepiness) alertness in the morning after wake-up. Methods: Participants spent two separate nights in the laboratory and were exposed to either one of the two light intensities or colors for 60 min after wake-up. Additionally, they indicated their sleepiness on the Karolinska Sleepiness Scale and participated in an auditory RT task before and after light intervention. It was hypothesized that both bright and blue light would lead to greater subjective and objective alertness when compared to dim and red light, respectively. Results: Results indicated that average RTs were longer for participants in the bright light condition (p = 0.004, f2 = 0.07) and that RTs decreased post-light exposure irrespective of light being dim or bright (p = 0.026, f2 = 0.07). However, dim versus bright light and RT did not interact (p = 0.758, f2 = 0.07). Chronotype was a significant covariate in the interaction of dim versus bright light and subjective sleepiness (p = 0.008, f2 = 0.22). There was no difference in RTs when comparing exposure to red or blue light (p = 0.488, f2 = 0.01). Findings on subjective sleepiness and light of different wavelengths revealed that sleepiness was reduced after light exposure (p = 0.007, f2 = 0.06), although the wavelength of light did not appear to play a role in this effect (p = 0.817, f2 = 0.06). Conclusion: Hence, neither of the hypotheses could be confirmed. However, they indicated that evening types might benefit from exposure to bright light regarding sleepiness, but not morning types.
Read more