17628 research outputs found
Sort by
DFL-RUL: Decentralised Federated Learning for Battery Remaining Useful Life Estimation on Heterogeneous Edge-to-cloud
No full textAccurate remaining useful life (RUL) prediction of lithium-ion batteries is essential for reliable and cost-effective electric vehicle operation, yet existing approaches largely rely on centralised training or overlook deployment constraints and data heterogeneity. This paper introduces DFL-RUL, a decentralised federated learning framework specifically designed to address feature-space inconsistency, temporal generalisation, and edge-level feasibility in real-world battery prognostics. Unlike prior federated RUL methods that assume aligned feature representations across clients, DFL-RUL integrates unsupervised, client-side PCA to automatically align heterogeneous sensor features before model aggregation. Local battery degradation is modelled using lightweight LSTM networks, while global knowledge is learned through FedAvg-based aggregation without sharing raw data. To reflect practical forecasting conditions, the framework is evaluated under a forward-in-time validation protocol, where only early-life cycles are available during training. Extensive experiments demonstrate that DFL-RUL achieves accuracy comparable to or exceeding local and centralised baselines, while significantly reducing communication cost and training latency. Moreover, runtime profiling on EV-class edge hardware confirms low inference latency and low energy consumption, validating the framework’s suitability for on-device deployment. These results show that reliable battery RUL estimation can be achieved in a privacy-preserving, hardware-aware, and temporally robust federated setting
Adoption of artificial intelligence in primary health care: systematic synthesis of stakeholder perspectives
Full text linkIntroduction: Primary care, the cornerstone of healthcare systems, faces increasing pressures from aging populations, chronic diseases, and resource constraints. Artificial intelligence (AI) offers transformative potential to enhance diagnostic accuracy, streamline workflows, and improve patient outcomes. However, its integration into primary care is challenged by technical, ethical, and organizational barriers. This systematic review examines AI’s role in primary care, focusing on stakeholder perspectives and implementation dynamics. Methods: A systematic synthesis of qualitative studies was conducted following Noblit and Hare’s framework and Braun and Clarke’s thematic analysis. Searches spanned PubMed, Scopus, Web of Science, CINAHL, and grey literature (2015–2025), identifying qualitative studies on AI in primary care. Studies were screened using predefined criteria, with quality assessed via the Critical Appraisal Skills Programme (CASP) checklist. Data were extracted systematically and synthesized, with initial search for 1416 studies. Results: Finally, 23 studies from diverse regions (e.g., UK, USA, Australia, Cameroon) and involving stakeholders like physicians, patients, and policymakers were included. Six themes emerged: Barriers (technical, organizational, policy, knowledge, cultural), Facilitators (benefits, trust, support systems, evidence), Impact on Healthcare Delivery (workflow, decision-making, roles, engagement), Ethical/Legal/Social Implications (privacy, accountability, equity, public perception), Stakeholder Perspectives, and Future Directions. AI improved efficiency and diagnostics but faced challenges like data quality, trust deficits, and ethical concerns. Conclusion: AI holds significant promise for transforming primary care by enhancing efficiency and patient care, but its adoption is hindered by multifaceted barriers from stakeholder perspectives. Transparent AI systems, robust training, and ethical frameworks are crucial to build trust and ensure equity. Future research should focus on longitudinal impacts and inclusive strategies to align AI with primary care’s patient-centered ethos
Cross-modal invariant learning with latent diffusion for reliable medical diagnosis under dynamic shifts
Full text linkRobust and reliable medical diagnosis using artificial intelligence is crucial, yet real-world clinical environments present significant challenges due to dynamic covariate shifts affecting multi-modal data (images, text, tabular). Existing methods, including the single-modal robust classifier LaDiNE, often fail under these complex, multi-modal shifts, lacking mechanisms for cross-modal invariance, dynamic modality fusion, and fine-grained uncertainty attribution. To address this gap, we propose DyMoLaDiNE (Dynamic Multi-Modal Latent Diffusion Nested-Ensembles), a framework designed for reliable medical diagnosis under dynamic multi-modal covariate shifts. DyMoLaDiNE introduces four key innovations: (1) a Cross-Modal Invariant Feature Extractor leveraging multi-modal Vision Transformers and contrastive learning to derive robust latent representations, (2) a Dynamic Modality Weighting Mechanism that adaptively adjusts modality contributions based on instance-specific reliability scores, (3) a Robust Multi-Modal Diffusion Ensemble utilizing conditional diffusion models conditioned on multi-modal inputs and reliability scores for flexible, calibrated density estimation, and (4) Modality-Attributed Uncertainty Quantification to decompose predictive uncertainty by input source. Extensive evaluations on diverse datasets (MedMD&RadMD, MultiCaRe, PadChest, TCIA RE-MIND, BRaTS, Camelyon16, PANDA) demonstrate that DyMoLaDiNE significantly outperforms (p0.005) state-of-the-art methods (LDM, CMCL, CGMCL, CIIM, DTTL, FFL, ALDM, LaDiNE) in terms of classification accuracy, robustness under dynamic perturbations, confidence calibration (ECE), and precise uncertainty quantification (CPIW, CNPV), while providing superior modality attribution fidelity. Ablation studies confirm the necessity of each component. DyMoLaDiNE represents a significant advancement in trustworthy, robust multi-modal medical AI. Code supporting this study DyMoLaDiNE https://github.com/SaeedIqbal/DyMoLaDiN
Non-Centralized Quantum Neural Networks for Cell-Free MIMO Systems
Full text linkThis paper propose a two-stage quantum neural network (QNN) framework for cell-free multiple-input and multiple-output (MIMO) wireless communication systems. Cellfree MIMO, which has been regarded as a key technology for enhancing the performance of the next-generation wireless communication systems, leverages the collective capability of multiple distributed access points (APs), allowing collaboration between them. However, optimizing cell-free MIMO can pose challenges for centralized optimization schemes. In particular, complexities associated with the joint optimizations of usertransmission assignment and transmission precoding, two factors which are of much importance for determining the quality-ofservice, grow with the number of APs and served users. To this end, a unified scheme employing distributed QNNs is used to optimize downlink transmitter-user assignment and transmit precoding with the goal of maximizing the achieved sum rate. Firstly, the cloud processing unit, which holds holistic information about the particular wireless communication network, employs QNN to assign each AP to its designated mobile terminal. Secondly, the edge processing units, which are computed in proximity relative to the AP in order to reduce latency, estimate transmission precoding for their corresponding APs. Moreover, numerical results are presented to showcase the performance of the proposed protocol
Evaluation under uncertainty: Autism, dyslexia and the epistemic architecture of fragile healthcare systems
No full textThis paper develops a general mathematical theory of evaluation in healthcare grounded in operational fragility and regime-based management. Using national autism and dyslexia assessment systems in England, it formalises evaluation as a stochastic and information-theoretic process rather than a neutral record of clinical reality. Evaluation outcomes are shown to emerge through mutable methodological regimes, probabilistic authority functions, and shifting forms of classification governance. The framework defines explicit fragility parameters (, 2 regime/2 measurement), derives six formal theorems describing discontinuity, information decay, variance bounds, and authority non-commutativity, and validates them empirically with NHS data (2022–2025). Rather than extending qualitative critiques of measurement and classification, the study constructs a falsifiable mathematical architecture in which inference, authority, and system observability replace objectivity as the central analytic categories. The resulting regime-aware evaluation model generalises across healthcare and education systems and provides operational metrics for epistemic stability and system fragility
Match and training injuries sustained by professional male rugby union players in Scotland
No full textRugby union (“rugby”) is a full contact sport, with previous studies across the globe reporting a high incidence of injury. However, no injury surveillance study of professional male players in Scotland exists in contemporary literature. The current study therefore aimed to describe the incidence, severity, burden and nature of match and training injuries sustained by male professional club rugby players in Scotland. A prospective cohort study of injuries sustained during matches and training across the 2017/18 and 2018/19 seasons was undertaken, with injury and exposure definitions in line with the international consensus statements. Injury incidence was 136.2/1000 player match hours and 4.1/1000 player training hours, median injury severity was 7.0 days (match) and 7.5 days (training), and injury burden was 2,887.0/1000 player match hours and 102.3/1000 player training hours. Concussion (match) and posterior thigh muscle injuries (training) were the most common specific diagnoses. Injury incidence in this population was higher than reported elsewhere in previous studies. However, high incidences of tackle injuries and concussion injuries agree with previous literature, reinforcing the need for mitigation strategies targeting these areas
Systems Thinking for Sustainability: Shifting to a Higher Level of Systems Consciousness
Full text linkThe grand challenges encapsulated in the seventeen UN Sustainable Development Goals to be achieved by 2030, are complex, messy and interconnected. Fulfilling these goals necessitates a shift in mindset from ego-to-ecosystems awareness and an imperative for stakeholder collaboration. Systems thinking is crucial to address sustainability challenges and an agenda for sustainable development. While some management approaches, like Doughnut Economics and Circular Economy, have roots in systems thinking, there is limited research into system thinking for sustainability. Nevertheless, the authors suggest we can learn from many systems-based contributions in the environmental science/studies literature that address ecological/Earth issues (e.g., Gaia, autopoiesis) and the Operational Research/Systems literature rich in a tradition of engaging communities in analysis and taking action. We ask, “How can systems thinking help businesses to meaningfully engage their stakeholders in a shared sense of purpose, value and impact?” The “systemic sustainability” framework (SSF) is proposed to address this, extending Laszlo’s concept and incorporating traditional systems thinking principles. The SSF emphasises that organisations and their stakeholders engage at four levels of systems awareness, reflecting on organisational purpose, and balancing organisational viability with planetary pressures. Interdependence, legitimacy and thrivability are highlighted as critical concepts in systems thinking for sustainability
How well are national policies addressing transport poverty in Scotland?
Full text linkTransport poverty can be caused by a lack of transport options that are available, reliable, affordable, accessible or safe. This study aimed to assess whether selected national transport policies were likely to achieve a population level impact on transport poverty in Scotland. We identified a long list of relevant policies from sources including the national transport strategy annual delivery plan. Transport Scotland officials prioritised 12 of these policies for review. Eight policies addressed affordability, three safety, and two accessibility, one of which addressed both accessibility and safety. We used available evidence, mainly from policy documentation and evaluations, to score whether these were: systematically applied; scaled up appropriately; resourced in the long term; and evidence based, to generate an overall assessment of likely population level impact. We scored eight policies as high population level impact, three medium and one low. The policies were all legislative or universally available for defined populations, with few barriers to uptake. We identified bus concessionary schemes as particularly important to improve affordability, but some low-income populations who could most benefit are not eligible. We assessed three legislative policies as likely to have a population impact on accessibility and/or safety. We conclude that addressing each dimension in isolation is not sufficient to reduce transport poverty. A broad transport poverty strategy addressing all dimensions of transport poverty should be developed to ensure everyone can access transport options to meet their needs
Death on the table: how do operating room staff experience intraoperative deaths? A narrative synthesis of qualitative evidence
Full text linkBackground: Intraoperative deaths, though statistically rare, may evoke varied emotions among operating room (OR) staff that remain underrecognized and inadequately addressed.Aim: To synthesise the qualitative evidence regarding experiences of OR staff following patient death in the OR. A secondary aim is to unpack strategies to support OR staff following an intraoperative death experience.Design: Narrative review of qualitative studies.Data sources: Peer-reviewed databases (PubMed, EMBASE, CINAHL, Web of Science, Scopus and Cochrane Review Library) and grey literature sources (such as thesis databases) were extensively searched for peer-reviewed primary studies and non-peer-reviewed literature respectively reporting on intraoperative deaths or deaths occurring in the OR.Results: Six studies were retained. The synthesis revealed that unexpected OR deaths or those deaths perceived as sudden or preventable evoked more severe and enduring psychological repercussions, marked by guilt, hypervigilance, emotional and moral distress.In contrast, anticipated fatalities, particularly in patients with advanced illness, evoked less intense emotions but did not eliminate emotional tolls. The findings revealed divergent coping mechanisms among OR professionals: surgeons often engaged in meaning-making or employed emotion-focused and problem-focused strategies to process loss. In contrast, anaesthetists described emotional desensitisation over time. Nurses, meanwhile, navigated a pervasive culture of silence.Conclusion: The emotional toll captured underscores urgent needs for interventions, such as team-based debriefing support, alongside systemic reforms to normalise vulnerability and integrate emotional stewardship into institutional policies.Addressing this is not only ethically imperative but critical to sustaining a resilient workforceand ensuring patient safety in an era of escalating surgical demand
AI-powered automated building façade segmentation and BIPV system potential prediction using CycleGAN and PVGIS
No full textBuilding-integrated photovoltaics (BIPV) represents a promising pathway for advancing urban sustainability. Accurately identifying suitable façade surfaces is essential for maximising opportunities for BIPV integration. This study develops a novel AI-powered framework based on Cycle-Consistent Generative Adversarial Network (CycleGAN) to segment building façade regions suitable for BIPV installation. The developed unsupervised learning approach enables unpaired image-to-image translation between real-world facades and their corresponding segmentation masks, thus eliminating the need for pixel-level annotations, reducing reliance on manually labelled datasets, and minimising system sizing time. The resulting façade segmentation mask was post-processed and used as input to PVGIS, accessed via its application programming interface (API) through Python for energy yield prediction, while PVsyst simulation was employed for PVGIS validation. A case study conducted in Edinburgh (Lat/Lon 55.933, −3.213) for a south-facing façade demonstrated the model’s ability to identify 81 m2 of usable BIPV areas, representing 41.55% of the total surface. The CycleGAN model achieved an intersection over union (IoU) of 0.78 and a Dice coefficient of 0.88, confirming stable adversarial learning. The end-to-end processing time per façade image ranged from 2 to 5 s. The results showed close agreement between both tools, with annual energy generation values of 11.8 MWh (PVGIS) and 11.5 MWh (PVsyst), corresponding to a relative deviation of approximately 2.5%. The findings highlight the practicality, scalability, and cost-effectiveness of integrating AI-façade segmentation with energy simulation tools for early-stage BIPV assessment. This integrated workflow provides a foundation for urban BIPV planning and pre-feasibility studies, supporting innovative renewable integration within city infrastructure
Data sources:
Access Repository Dashboard
Do you manage Repository@Napier? Access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard!