Password-based authentication systems remain the most widely used method for user verification despite being highly susceptible to offline dictionary attacks. To mitigate such attacks, server-aided password-based authentication schemes utilize an independent server, which helps to harden the credentials to be stored on the website database. Existing server-aided password-based authentication schemes rely on number-theoretic assumptions that are vulnerable to quantum-enabled adversaries and incorporate complex computations such as bilinear pairings, exponentiation, and Zero-Knowledge Proofs. In this work, we introduce a novel post-quantum secure server-aided password-based authentication scheme based on the Module Learning With Errors (M-LWE) problem. A defining feature of our protocol is its complete operational transparency as it integrates with existing web interfaces without requiring users to modify their login behaviour or perform additional computation. To ensure long-term resilience, our scheme includes a transparent key rotation mechanism that allows service providers to update the entire credential database with a fresh secret key without user intervention. We provide a formal security analysis in the Real-or-Random (RoR) framework. This analysis demonstrates that our protocol's resistance to offline dictionary attacks reduces to the underlying hardness of the M-LWE problem, and the system achieves forward secrecy through a key rotation mechanism. Through an optimized Number Theoretic Transformation (NTT)-based implementation for faster polynomial multiplications, our empirical analysis demonstrates high computational efficiency, with average registration and authentication latencies of 0.88 ms and 0.96 ms, respectively.
Arginylation is a well-known, evolutionarily conserved, and recognizable post-translational modification (PTM). In this modification, an arginine residue is enzymatically attached to the target residue, such as aspartic acid (D), glutamic acid (E), and oxidized cysteine (Cys-SO₂H, Cys-SO₃H) by arginyltransferase 1 (ATE1) utilizing arginyl-tRNA. This phenomenon occurs both at the N-terminal and mid-chain of a protein. Arginylation is required for various biological activities like overall cellular development, cytoskeletal organization, and cellular stress adaptation. For predicting arginylation likelihood, only laboratory-based experimental procedures are currently available. To solve this, we developed the web server ARGpredict to predict arginylation events in protein sequences. The designed web server analyzes the arginylation target sites and assigns specific scores. To evaluate its effectiveness, this web server was tested against various known and experimentally validated datasets. ARGpredict yielded promising prediction results, which were also verified by statistical methods. The web server is available at https://iitg.ac.in/sachinku/ARGpredict/.
Electronic health records offer vast clinical data for health care research, but interoperability challenges often hinder comprehensive analysis. The Health Level Seven Fast Healthcare Interoperability Resources (FHIR) standard addresses these challenges, although its nested and interconnected resource format can be complex for analytics. Several tools have emerged to facilitate analytical access, either by querying FHIR servers via representational state transfer (REST) APIs or encoding resources in relational formats. However, the performance implications of these methods remain largely unexplored. This study aimed to benchmark the performance characteristics of different FHIR-based analytical approaches comparing REST API queries against SQL- and Spark-based big data frameworks operating on FHIR-encoded data. We benchmarked the FHIR-PYrate library, which interfaces with a FHIR server's REST API, against Pathling, a library built for analytics based on Apache Spark, and Trino, a general-purpose SQL query engine. We defined and implemented multiple queries in each engine using 3 common analytics scenarios-data aggregation, counting, and extraction. Execution times were measured across Synthea-generated datasets of increasing size. On the largest dataset, containing 71,285,064 FHIR resources, Trino completed the aggregate query more than 12,000 times faster, and Pathling did so approximately 500 times faster than FHIR-PYrate. On average across all queries, Trino outperformed FHIR-PYrate, executing extraction queries 33 times faster and count queries 1.8 times faster. Pathling achieved a 2.6-time speedup for extraction queries, but FHIR-PYrate was approximately 13 times faster for count queries. While the REST-based FHIR search API is useful for standard queries and retrieving specific patient records and can outperform alternatives for some count queries, it generally lacks the performance and expressiveness needed for complex analytics. In contrast, alternative engines such as Trino and Pathling demonstrated substantial performance advantages for these scenarios.
Cloud service providers face the challenge of determining optimal server replenishment policies that minimize inventory costs while ensuring the expected demand satisfaction rate. This article addresses a long-term, single-echelon inventory optimization problem tailored to the unique characteristics of cloud, incorporating varying given demand satisfaction rates for different cloud products. Traditional replenishment methods, based on statistical approaches or conventional reinforcement learning (RL) algorithms, are ill-suited due to the uncertainties in user demand retention and the diversity in demand satisfaction requirements. To overcome these limitations, we propose an enhanced replenishment policy under given demand satisfaction rates using RL, combined with our policy imitation. Specifically, we introduce a data-driven safety stock (DSS) model as an expert policy, utilizing Kullback-Leibler (KL) divergence and advantage values for effective action imitation. Based on the results of DSS, we design an efficient reward function and propose an approach for hyperparameter estimation. Together, they enable our method to learn better policies that align with the given demand satisfaction rate. We further design an end-to-end network that integrates demand retention prediction with RL, incorporating a learnable Gaussian kernel to model the lead time effects and local features, and a mean self-attention module for capturing global features. To accelerate training, we also introduce a parallel algorithm. Extensive experiments on real-world data from multiple internet data centers (IDCs), across different time scales and given demand satisfaction rates, demonstrate the superior performance of our method. Additionally, ablation studies and parameter sensitivity analyses verify the efficacy of our policy imitation method and network architecture.
As a distributed approach to Artificial Intelligence (AI) model construction over wireless networks, federated learning (FL) based on multi-device collaborative training can protect data privacy, as well as increase the computing load of local model updates. In contrast, split learning (SL) with proper model splitting can adapt to the computation and transmission capabilities among devices. In this paper, while taking advantage of FL and SL, we concentrate on a semi-decentralized hybrid federated split learning (SD-HFSL) framework, in which we surpass the limitations of a single central server and allow the shared split models to be aggregated among multiple edge servers. To verify the importance of latency optimization for training efficiency, we analyze the convergence performance of SD-HFSL while jointly considering the limited computation and communication resources. Then, aiming at maximizing the long-term training efficiency, we propose an online optimization problem that includes local model splitting and device association. Considering that the training latency is unknown to the system a priori, a context-aware online training algorithm with sublinear regret is proposed based on the framework of contextual multi-armed bandit (CMAB), where the edge servers can observe the context information of device sites for latency estimation, followed by the iterative optimization based on the evaluated information in different contexts. Experiments on several neural network models show that the proposed algorithm reduces training latency and improves test accuracy compared with the selected benchmarks.
Spectral library matching is the most common method for compound identification in mass spectrometry (MS), but it is limited by the coverage of experimental libraries. In-silico libraries can expand the searchable chemical space, yet their utility is hindered by the experiment-to-prediction gap arising from the discrepancies between experimental and predicted spectra. Here, we propose CSU-EP, a Contrastive spectral unification framework that bridges this gap with a two-stage training strategy. First, it is self-supervisedly pretrained on 1,883,697 predicted electron ionization (EI) mass spectra using masked peak prediction. Second, it is fine-tuned via contrastive learning to unify representations of paired experimental and predicted spectra. Using this fine-tuned encoder, we assemble 2.24 million NEIMS-predicted spectra into a spectral embedding database (CSU-EP-DB). CSU-EP achieves a Recall@1 of 47.10% on the NIST replib benchmark when searching against CSU-EP-DB, outperforming LLM4MS by 7.28% and FastEI by 12.47%. It also demonstrates excellent ability to identify compounds absent from experimental libraries. In a plasma metabolomics application, incorporating a molecular weight filter boosts Recall@5 to 92.86%. To ensure broad accessibility, a user-friendly web server is deployed. By effectively bridging the experiment-to-prediction gap, CSU-EP establishes a new paradigm for reliable, scalable compound identification using in-silico libraries. The source code, models, and CSU-EP web server are accessible at https://github.com/tingxiecsu/CSU-EP.
Federated learning enables collaborative model training between central servers and distributed clients without collecting users' raw sensitive data, which effectively promotes the large-scale deployment of intelligent collaborative services. Considering the high sensitivity of local training data and model gradient parameters in federated learning, protecting identity privacy and interaction security has become extremely critical. Therefore, mutual identity authentication is indispensable to restrict illegal client access and prevent malicious parameter transmission and data tampering. In this paper, we propose a lightweight anonymous authentication scheme for federated learning (FedLAS), which realizes secure mutual authentication between servers and clients and establishes a shared session key for subsequent encrypted interaction. In particular, the proposed scheme eliminates the reliance on high-cost cryptographic operations such as bilinear pairing, thus minimizing computational and communication overhead. Furthermore, informal security analysis demonstrates that our FedLAS scheme can resist multiple common attacks and meet predefined security requirements. Extensive comparative experiments show that the FedLAS scheme achieves excellent performance in computational and communication cost. It is well suitable for resource-constrained federated learning scenarios.
Cryogenic electron microscopy (cryo-EM) has become a central technology for determining macromolecular structures in structural biology. However, interpreting density maps by structure modeling from intermediate (∼5-10Å) to low resolution (over 10Å) cryo-EM maps often remains challenging. To meet the demand for accurate modeling tools in cryo-EM, we developed the EMSuite server (https://em.kiharalab.org/), a freely accessible platform that provides 14 computational tools for structure modeling and validation. In this chapter, we focus on three methods developed for intermediate resolution maps, DiffModeler, DMcloud, and VESPER, presenting their usages through the web server. DiffModeler builds full protein complexes by enhancing backbone positions with a diffusion-based generative model and using AlphaFold to predict structural models. Similarly, DMcloud performs local structure fitting instead of global structure fitting to eliminate incompatible regions in structure models to the map. Finally, VESPER is a general-use structure fitting method, which performs global structure-map and map-map fitting.
Personalized Federated Continual Learning (PFCL) requires that the server not only effectively integrate temporal knowledge accumulated across past tasks and spatial knowledge distributed among heterogeneous clients, but also ensure strong personalized performance of the global model for each client. Existing methods, whether in Personalized Federated Learning (PFL) or Federated Continual Learning (FCL), have overlooked the multi-granularity representation of knowledge, which can be utilized to overcome Spatial-Temporal Catastrophic Forgetting (STCF) and enable coarse-to-fine personalization. Furthermore, most approaches rely on local client-side personalization, increasing computational load and failing to address the risks posed by malicious or low-quality clients. To this end, we propose FedMGP+, which utilizes multi-granularity prompts to address these challenges, namely coarse-grained global prompt and fine-grained local prompt. The former focuses on efficiently transferring shared global knowledge without spatial forgetting, and the latter emphasizes specific learning of personalized local knowledge to overcome temporal forgetting. Visualization results and theoretical analyses further reveal that coarse-grained prompts primarily guide regional attention, whereas fine-grained prompts enrich object-level representations within those regions. Building upon this, Personalized Selective Prompt Fusion is designed to exclusively fuse coarse-grained knowledge on the server to generate client-specific global prompts, reducing local overhead and resisting poisoning attacks from malicious clients. Extensive experiments demonstrate that the proposed FedMGP+ effectively mitigates forgetting, significantly enhances personalized performance, and provides robust defense against poisoning and gradient leakage attacks.
The increasing complexity, scale, and diversity of cyber threats pose significant challenges to modern network security systems, particularly in heterogeneous network environments. Existing intrusion detection methods often rely on dataset-specific models that struggle to maintain consistent performance across different data distributions and typically lack interpretability, limiting their applicability in critical systems. In this paper, we present an explainable deep learning framework evaluated across multiple heterogeneous cyber attack datasets, including Kitsune (IoT traffic), Server-Based network data (enterprise logs), and Malware Traffic datasets. The proposed pipeline employs a consistent preprocessing strategy involving feature normalisation, encoding, and label alignment, followed by a multi-layer perceptron (MLP) classifier with dropout regularisation. Model training incorporates cosine learning-rate scheduling and early stopping, while explainability is achieved via SHAP-based global feature attribution and LIME-based instance-level explanation. Performance is evaluated using accuracy, ROC-AUC, and F1-score metrics. The model achieves 96.0% accuracy on the Kitsune dataset, 99.99% on the Server-Based dataset, and 99.58% on the Malware dataset, with ROC-AUC values approaching 1.00 across most classes. The results demonstrate the effectiveness of the proposed framework in handling heterogeneous network traffic while providing interpretable insights into model predictions.
Identifying meaningful associations between microbial communities and measured physiological or environmental variables becomes increasingly complex and computationally demanding given the continuous growth of microbiome datasets. The Coracle machine learning (ML) framework was recently developed to address this by integrating multiple data transformations, feature selection techniques, and ML models to yield condensed lists of features that align to target variables of interest. Further, we recently developed the UniCorP feature aggregation algorithm to identify uniquely correlated features (UNICORNs) based on the UniCor metric that iteratively enrich each taxonomic level in an automated bottom-up approach. Here we present UniCoracle, a fully automated analytical framework that integrates UniCorP's bottom-up propagation approach with a subsequent and newly developed top-down skimming (TDS) strategy, implemented with the Coracle ML framework. This combined approach leverages the inherent taxonomic structure of microbiome community data (e.g., ASVs derived from 16S rRNA gene amplicon sequencing) to maintain predictive stability, reduce computational runtime, and identify biologically meaningful taxonomic associations. We compare the original, non-hierarchical Coracle with the TDS Coracle method and the UniCoracle approach. Evaluations across the tested datasets show that UniCoracle achieves competitive or improved predictive performance relative to both Coracle's multi-step and the TDS-based Coracle implementations. Our results demonstrate UniCoracle's improvements in predictive accuracy over both Coracle's multi-step and the TDS Coracle methods. UniCoracle provides full control over feature set size and runtime, offering a streamlined and user-friendly framework for biological hypothesis generation. It identifies features (e.g., bacterial taxa) at the lowest (most specific) hierarchical level (e.g., ASV or species within a taxonomic hierarchy) that are associated with continuous target variables. Availability: UniCoracle is freely accessible via a dedicated web server at micportal.org. The source code is open-source and available on GitHub at github.com/SebastianStaab/UniCoracle.git and Zenodo at https://doi.org/10.5281/zenodo.19050205. Supplementary information: Example datasets and a tutorial are provided on the web server: micportal.org.
We present a collaborative, interactive, web-based visualization and processing system for Cryo-Electron Tomography (cryo-ET) data that supports the complete workflow, from raw tilt-series projections to final 3D segmentations. The system integrates tilt-series alignment, motion correction, tomographic reconstruction, manual and pseudo-annotation, deep model training, inference, and visualization, all accessible through a unified web-based interface. The backend manages computationally intensive tasks, including advanced tomographic reconstruction, pseudo-annotation generation, deep learning model training, and large-scale segmentation of new datasets. It features separate queues for CPU and GPU tasks, allowing users to prepare and manage large workloads in advance. The frontend supports lightweight tomographic reconstruction, manual data annotation, segmentation with pretrained models, interactive exploration of pipeline stages, and real-time monitoring of queued tasks. Key processing steps are available on both the client and server sides, providing flexibility for diverse computational environments, enabling visualization of externally hosted datasets, and supporting data privacy when required. The system is designed for collaborative workflows, enabling multiple users to share input data, annotations, and trained models within joint projects. Its modular architecture allows easy extension with additional stages on either the client or server side, making it adaptable to specialized pipelines within individual institutions. We demonstrate the system's capabilities on representative cryo-ET datasets, highlighting its utility for both individual researchers and distributed teams working on structural analysis of complex cellular environments. The initial release of the system is available at https://cocryovis.lgm.fri.uni-lj.si/ (Project repository: https://github.com/nanovis/cocryovis, mirror deploy: https://tomography.kaust.edu.sa/), and the entire system will become open source so that any visualization or domain researcher can further enhance its functionality.
Microneme protein 3 (MIC3) is a key adhesion molecule in Toxoplasma gondii that is expressed during multiple stages of infection. We aimed to computationally characterize the immunological and structural features of the T. gondii MIC3 protein to assess its potential suitability as a vaccine candidate. A comprehensive set of bioinformatics tools and web servers was employed to predict the physicochemical properties, allergenicity, antigenicity, solubility, post-translational modification sites, subcellular localization, transmembrane domains, signal peptides, secondary and tertiary structures, potential B- and T-lymphocyte epitopes, and simulated immune responses of the TgMIC3 protein. A total of 75 post-translational modification sites were predicted in TgMIC3. Furthermore, secondary structure analysis using GOR IV, SOPMA, and NetSurfP-3.0 indicated that random coils and extended strands were the predominant structural elements. In addition, several high-affinity B- and T-cell epitopes were identified across the protein sequence. Subsequent structural validation revealed that 82.91% and 98.60% of residues were located in favored regions in the initial and refined 3D models, respectively. The findings of the allergenicity and antigenicity assessments indicated that the MIC3 antigen seemed to be a non-allergen with an immunogenic nature. Moreover, immune simulation using the C-ImmSim server demonstrated that TgMIC3 could induce robust humoral and cell-mediated immune responses following three simulated antigen administrations. Conclusion: This study provides foundational computational evidence supporting the potential of TgMIC3 as a vaccine antigen and offers a useful framework for future experimental investigations targeting vaccine development against acute and latent toxoplasmosis.
Women's health research is limited. The Canadian Longitudinal Study on Aging (CLSA) offers rich data relevant to women's health. However, no review has examined how CLSA data have been used to study women's health. This scoping review: 1) identifies key research areas and trends; 2) synthesizes findings; 3) examines sex and gender integration; and 4) highlights gaps and opportunities. We searched CINAHL, EMBASE, Medline, PsycINFO, and preprint servers. Studies were included if they: a) included CLSA female participants aged 45+; b) focused on women's health in their aims/objectives; and c) were primary analyses from peer-reviewed publications or preprints. Data were extracted on study aims, CLSA data use, population, methods, sex and gender information, and study results. Twenty-two studies met all inclusion criteria (out of 801 screened). Studies focused on postmenopausal females (n = 12), all reproductive stages (i.e., all females, n = 9), and pre/perimenopausal females (n = 1). Exposures of interest centered on the reproductive system (e.g., menopausal transition, gynecological cancers) and menopause hormone therapy (MHT), whereas outcomes spanned broader health domains, including musculoskeletal, cognitive, metabolic, and cardiovascular. Postmenopausal status and earlier age at menopause were associated with adverse health outcomes. MHT use was linked to both reduced and increased health risks. Women's health research using the CLSA data is growing but remains narrowly focused on reproductive health and MHT. Future research should expand beyond the reproductive system and include understudied populations and the menopausal transition.
This paper considers minimax and adaptive transfer learning for nonparametric classification under the posterior drift model with distributed differential privacy constraints. Our study is conducted within a heterogeneous framework, encompassing diverse sample sizes, varying privacy parameters, and data heterogeneity across different servers. We first establish the minimax misclassification rate, precisely characterizing the effects of privacy constraints, source samples, and target samples on classification accuracy. The results reveal interesting phase transition phenomena and highlight the intricate trade-offs between preserving privacy and achieving classification accuracy. We then develop a data-driven adaptive classifier that achieves the optimal rate within a logarithmic factor across a large collection of parameter spaces while satisfying the same set of differential privacy constraints. Simulation studies and real-world data applications further elucidate the theoretical analysis with numerical results.
Over the past decades, the threat of Staphylococcus aureus and multidrug-resistant S. aureus has escalated, creating an urgent need for novel therapeutic strategies that possess both bactericidal and antivirulence properties. We focused on gramicidin D, a conventional pore-forming antibiotic derived from Bacillus brevis, as a multi-target agent. The minimum inhibitory concentration (MIC) and half-maximal inhibitory concentration (IC50) against several Gram-positive pathogens were determined by broth microdilution assay. Furthermore, the combinatorial effect with other conventional antibiotics was also assayed via the checkerboard method. The effects on biofilm formation and mature biofilm were evaluated using the crystal violet staining method, supported by colony-forming unit (CFU) counts. Additionally, inhibition of staphyloxanthin biosynthesis was quantified by methanol extraction, followed by measurement of optical density via a microplate reader. Finally, in silico protein-peptide docking was performed to support and analyze the properties of the peptide. Gramicidin D demonstrated significant antibacterial efficacy against S. aureus and methicillin-resistant S. aureus (MRSA), with corresponding MIC values of 8 and 0.25 µg/mL, respectively. Moreover, gramicidin exhibited synergistic effects with oxacillin and gentamicin against S. aureus; however, it showed synergy with oxacillin only against MRSA. It inhibited S. aureus biofilm formation at sub-MICs, especially at 0.125 µg/mL, where 80.0% of biofilm was inhibited. Furthermore, it effectively reduced established biofilm at concentrations ranging from 128 to 4 µg/mL. Additional SEM images showed that membrane pore formation by gramicidin D was supported. It also inhibited staphyloxanthin synthesis at all concentrations ≥ 1 µg/mL. HDOCK server-based protein-peptide docking provided structural support for the in vitro experiments. These findings altogether illustrated the potential of gramicidin D as a multi-target agent against S. aureus.
An evidence synthesis informed the 2023 WHO infection prevention and control (IPC) guideline for Ebola disease (EBOD) and Marburg disease (MARD). A series of rapid reviews addressed 14 key questions (KQs) related to IPC measures for EBOD and MARD across three themes: (1) transmission/exposure, (2) personal protective equipment (PPE) and (3) decontamination/disinfection. Databases (Embase, Cochrane Database of Systematic Reviews, CENTRAL, Global Index Medicus) were searched from inception to May 2024, in addition to searches in MEDLINE and bio/medRxiv servers through an artificial intelligence tool (continuous active learning, CAL). Eligible populations included health workers (HWs), staff and patients in healthcare and/or community settings. Interventions included PPE evaluation, hand hygiene/glove decontamination, spraying of HWs, handling of heavily soiled linen, management of the deceased, environmental decontamination, work exclusion and the IPC ring approach. Infection with EBOD or MARD was a primary outcome of interest and secondary outcomes of interest varied by key question. Screening, data extraction, quality assessment (Cochrane RoB-2, Risk Of Bias In Non-Randomized Studies - of Interventions (ROBINS-I)) and evaluation of the certainty of evidence (Grading of Recommendations Assessment, Development and Evaluation (GRADE)) were conducted by an independent reviewer with verification. Data and relevant contextual information were narratively synthesised. Following screening of 5540 studies in CAL and 1062 full texts, nine studies were included that addressed four KQs: one related to transmission/exposure (IPC ring approach), two related to PPE (head/neck skin and mucous membranes coverage; order of eye protection and head covering) and one related to disinfection/decontamination (spraying of HWs during PPE doffing). No studies were included for the remaining KQs, but relevant contextual information in reviewed studies was summarised for all KQs. Evidence was judged to be of low or very low certainty using the GRADE framework due to risk of bias, indirectness and imprecision, and substantive conclusions could not be drawn on the effectiveness of the evaluated IPC interventions. While providing the evidence base for the WHO IPC guideline for EBOD and MARD, this series of rapid reviews identified several critical research gaps. CRD#42022331537.
With the rapid growth of data center (DC) energy consumption and the large-scale integration of renewable energy, DCs increasingly face time-varying power upper-bound constraints jointly shaped by grid power supply capability, renewable energy fluctuations, and demand response mechanisms. Meanwhile, DC power consumption exhibits a typical information-load-driven characteristic. The computing tasks hosted by virtual machines affect server-side IT power consumption through resource utilization states such as CPU, memory, disk I/O, and network I/O, and are further coupled with non-IT auxiliary power consumption from cooling, power distribution, and networking equipment. In such cyber-physical operation scenarios, physical-layer sensing data and hypervisor-level virtualization monitoring data jointly provide the state basis for power estimation, power warning, and migration decisions. To address the mismatch between dynamic power upper bounds and time-varying information loads, this paper investigates the information load scheduling problem under constrained power loads and proposes a two-stage virtual machine (VM) migration optimization framework. In the VM selection stage, a Multi-Factor Balanced (MFB) algorithm is designed. By introducing a warning-line trend model based on the arctangent function, MFB comprehensively considers resource utilization, power load variation trends, and service level agreement (SLA) violation levels to dynamically identify candidate VMs for migration. In the VM placement stage, a Multi-Factor Equilibrium Ant Colony Optimization (MFEACO) algorithm incorporating a Random Roulette Wheel (RRW) selection mechanism is proposed. By constructing normalized multi-dimensional equilibrium factors, MFEACO coordinates the trade-off among energy consumption, load balancing, and SLA violations. Simulation experiments are conducted on an improved CloudSim platform using real-world cluster trace data from Google and Alibaba. The results show that, while satisfying dynamic power constraints, the proposed MFB-MFEACO framework achieves a favorable comprehensive trade-off among energy consumption control, SLA violation suppression, and migration reduction. Compared with traditional heuristic methods and a power-constrained genetic algorithm baseline, the proposed framework demonstrates better dynamic adaptability and scheduling stability.
Background: Chest X-ray imaging is a widely used diagnostic modality for identifying various lung diseases. Accurate multiclass classification of lung diseases enables timely treatment and improves patient survival. However, disease detection using chest X-ray images remains challenging due to heterogeneous data, overlapping radiographic features, and data privacy concerns. Furthermore, distinguishing among different lung diseases is difficult because of their similar clinical manifestations and imaging characteristics. Method: To address these challenges, a novel chaotic evolution optimization-enabled deep channel-attention broad convolutional neural network (CCEO-DCABNet) is proposed for multiclass lung disease classification within a federated learning (FL) framework. The proposed model ensures enhanced data privacy by allowing multiple client nodes and a central server to collaboratively train the model without sharing raw data. Prior to classification, image preprocessing is performed using Gaussian filter-based denoising followed by multiscale unsharp masking-based image sharpening. Subsequently, multiclass disease classification is carried out using DCABNet, whose parameters are optimized through the proposed CCEO algorithm. In addition, the federated learning process employs an averaging strategy for local model updates and global aggregation. Results: The proposed CCEO-DCABNet achieves an accuracy, true positive rate (TPR), and true negative rate (TNR) of 96.98%, 96.41%, and 97.45%. Conclusions: Experimental results demonstrate that the proposed CCEO-DCABNet framework effectively classifies multiple lung diseases from chest X-ray images while preserving data privacy through federated learning. The model achieves superior classification performance and can support reliable computer-aided diagnosis in clinical settings.
The COVID-19 pandemic has placed a significant burden on health professionals (HPs). They face higher infection risks due to the nature of their work environment and patient care responsibilities. Their ability to access and apply reliable COVID-19 information affects their own preventive behavior and that of those around them. In this context, health literacy (HL) has become increasingly important. Despite extensive research, information to foster COVID-19-related HL in HPs remains limited. This systematic review aimed to identify, appraise, and synthesize intervention studies on the effectiveness of COVID-19-related HL interventions in HPs. Five electronic databases (eg, PubMed (MEDLINE), Embase), six clinical trials registries (eg, ISRCTN registry), one preprint server (MEDRXIV), published conference proceedings, and five gray literature databases (eg, opengrey.eu, ProQuest) were searched in May 2022 and updated in August 2025. Reference lists of included studies were screened manually. Two reviewers independently screened titles, abstracts, and full-texts according to eligibility criteria and extracted data; disagreements were resolved by discussion or consultation with a third reviewer. We included randomized controlled trials (RCTs), nonrandomized studies of interventions, and uncontrolled before-and-after studies evaluating the effectiveness of any COVID-19-related HL intervention. Primary outcomes include COVID-19-related HL, its four facets (access, understand, appraise, and apply COVID-19 information), and indicators (eg, COVID-19-related knowledge), assessed at postintervention and follow-up. When studies were sufficiently similar, random-effects meta-analyses were performed; otherwise, a narrative synthesis was provided. Risk of bias was assessed using validated tools based on study design, and the overall certainty of the evidence was evaluated by the GRADE (Grading of Recommendations, Assessment, Development, and Evaluation) approach. We included 15 RCTs (2034 participants), 4 nonrandomized studies of interventions (291 participants), 74 uncontrolled before-and-after studies (327,298 participants), 5 ongoing studies, and 1 study with awaiting classification. Interventions targeted a broad range of health occupational groups. Intervention type, delivery mode, methods, settings, and comparator varied widely. No outcome measure explicitly referred to an HL model. Most studies aimed to enhance COVID-19-related knowledge and skills, and had a high risk of bias. COVID-19-related interventions may increase knowledge of vaccines (standardized mean difference 1.00; 95% CI 0.33 to 1.67, I2=24%), and the infection prevention control skills, such as donning and doffing of personal protective equipment (standardized mean difference 1.95; 95% CI 1.82 to 3.09, I2=46%), but the evidence remains very uncertain. COVID-19-related HL interventions may promote HP's short-term competencies in infection control. However, the evidence remains uncertain, primarily due to the low quality of studies, characterized by a high risk of bias. Interventions specifically designed to enhance the full COVID-19 HL operationalized by its four facets are lacking. High-quality RCTs with sufficient statistical power, grounded in HL theoretical principles, are needed to achieve precise understanding.