Abstract: Dorsal metacarpal disease (DMD) is a common musculoskeletal injury in young racehorses, and effective non-invasive treatments remain of clinical interest. This study aimed to evaluate whether high-intensity laser therapy (HILT), applied as the sole intervention, reduces clinical signs of DMD compared with untreated horses withdrawn from training. During the 2023–2024 racing seasons, 15 Arabian racehorses diagnosed with DMD were enrolled; 9 received HILT and 6 served as controls without laser therapy. The treatment protocol consisted of five daily HILT sessions followed by five sessions administered every other day. Thermographic, orthopedic, and radiographic examinations of the third metacarpal bones were performed before and after the treatment period in both groups, with additional thermographic and clinical assessments conducted throughout therapy. The results showed that horses treated with HILT did not exhibit a significant reduction in pain and lameness accompanied by thermographic changes consistent with decreased inflammation. These findings indicate that HILT did not alleviate clinical signs associated with DMD in the affected horses; however, further controlled studies are required to determine its effect on tissue healing processes and to optimize treatment duration.

Abstract: The Formosan subterranean termite, Coptotermes formosanus, is a globally distributed species that feeds on lignocellulose and causes substantial economic losses annually. Current control strategy heavily relies on chemical pesticides, raising concerns about environmental impacts associated with their overuse. Allyl isothiocyanate (AITC), a plant-derived pesticide, has demonstrated significant insecticidal activity. However, its effects on key physiological and biochemical systems in C. formosanus remain poorly understood. In particular, its impact on antioxidant enzymes, including superoxide dismutase (SOD), catalase (CAT), and peroxidase (POD), detoxification enzymes such as carboxylesterase (CarE), glutathione S-transferase (GST), and neural enzymes (e.g., acetylcholinesterase, AChE) has not been systematically investigated. Transcriptome data were used to predict coding sequences (CDSs) of antioxidant, detoxification, and neural enzymes, followed by phylogenetic analysis. C. formosanus was treated with AITC at LC₅₀ for 24 h, and enzyme activities and gene expression levels were assessed. Molecular docking was performed to evaluate interactions between AITC and the five enzymes. AITC exposure significantly increased the activities of all six enzymes. Gene expression analysis revealed differential regulation across enzyme families, with notable upregulation of AChE and several CarE, SOD, POD, and GST genes. Docking analysis indicated favorable binding affinity to target enzymes (binding energy < -1.2 kcal/mol). These findings suggest that AITC induces coordinated enzymatic and transcriptional responses in C. formosanus, providing insight into its mode of action and supporting its potential as a botanical termiticide with low environmental impact.

Abstract: Oral lichen planus (OLP) is a chronic inflammatory oral disease associated with immune dysregulation and malignant transformation risk. Vitamin D and probiotics may modu-late immune and microbial pathways involved in OLP. In this study, we evaluated their effects on clinical outcomes and multi-omics profiles in 25 adult OLP patients. Vitamin D-deficient patients received 2,000 IU/day vitamin D3, and all participants received a probiotic blend (Limosilactobacillus reuteri LRE11, Lactica-seibacillus rhamnosus LR04, and Lacticaseibacillus casei LC04) for 16 weeks. Clinical assess-ments and analyses of saliva, serum, oral swabs, and stool samples were performed be-fore and after treatment. Following the intervention, 76% of participants achieved clinical remission. Significant metabolomic changes were observed mainly in saliva and feces. Serum cytokines, me-tabolites, and lipoproteins showed no significant differences. Microbiome profiling demonstrated treatment-related compositional shifts in oral and fecal samples, including increased Lacticaseibacillus abundance. Multi-omics integration identified coordinated in-teractions among microbial, metabolic, immune, and lipid pathways, highlighting inter-connected gut-oral biological responses. Combined vitamin D and probiotic supplementation was associated with clinical im-provement and coordinated oral-intestinal multi-omics changes, supporting a sys-tems-level understanding of OLP remission. These findings suggest that modulating the microbiota-metabolism-immunity axis could represent a promising therapeutic strategy for achieving sustained disease control and clinical remission.

Abstract: Molar-Incisor Hypomineralization (MIH) is a developmental defect of the enamel that affects molars and incisors, with white, yellowish or brownish spots as its main charac-teristics, which may present characteristics of fragility in these regions, with a chance of fractures. This observational study investigates the efficacy of imaging tests requested in orthodontic documentation, in this case, cone beam computed tomography (CBCT) as an additional and profound possibility for identifying and diagnosing HMI. Children aged between 6 and 12 years were selected, who had clinically diagnosed HMI, and who participated in another study, where the request for CBCT was made possible. Of the 26 examinations evaluated, out of a total of 13 patients, 59 teeth were identified, of which 81% were white spots, 9% yellow/brown spots, and 10% had fractures. The results concluded that there is a visual possibility by the professional of the defects of the imaging exams, especially in CBCT, with the possibility of extension measurements; however, knowledge of the clinical condition is still the preponderant factor for the differential diagnosis of this enamel defect. It can be concluded that it is possible to observe HMI lesions on CBCT images, but the orthodontist needs to be qualified in this identification, through the additional tests that already make up his diagnostic arsenal. With this, relationships of difficulties in your clinic can be identified earlier, such as loss of space in the arch, dif-ficulty in bonding accessories, among others.

Abstract: Aim: To describe and propose the implementation of a 1:1 coaching model. Background: The nursing faculty shortage has become a significant factor in maintaining qualified educators, but little is known about specific strategies to retain them. A Coach/Player model can this relationship, creating zones where the experienced and novice faculty can share in each other's successes equally. Innovation: The coach and player worked together, assigned to teach multiple sections of a nursing course. The coach and player met 1-2 times per week virtually to prepare material and debrief on the previous week's course lecture, promoting confidence and ensuring continuity between the course sections. Implications: The player reported high satisfaction with their work environment and felt supported and valued, while the coach perceived satisfaction in passing their knowledge on to the player. Conclusion: Implementing the coach/player model is a promising strategy to promote retention, confidence, and success in novice nurse educators.

Abstract: Reliable monitoring of power system equipment is essential for ensuring operational stability, minimizing unexpected outages, and improving grid reliability. Among various condition monitoring techniques, optical sensing technologies have attracted significant attention due to their high sensitivity, electromagnetic interference immunity, and suitability for harsh electrical environments. This paper presents the design and application of a photonic crystal fiber (PCF)-based sensing system for real-time monitoring in power systems, with particular emphasis on dissolved gas detection in oil-immersed transformers. The proposed sensing approach employs hollow-core photonic crystal fiber (HC-PCF) as an optical absorption chamber, enabling enhanced light–gas interaction while maintaining a compact and flexible sensor configuration. Based on infrared absorption spectroscopy and Beer–Lambert theory, the system is designed to achieve high-sensitivity detection of characteristic fault gases generated during transformer insulation degradation. The diffusion characteristics of gases inside the HC-PCF are theoretically analyzed and experimentally verified to evaluate sensor response performance. Experimental investigations demonstrate that the proposed PCF-based sensing system provides excellent linearity, strong selectivity, and improved detection sensitivity for low-concentration acetylene monitoring. Allan variance analysis indicates that the optimal signal-to-noise ratio is achieved with a 29 s averaging time, resulting in a minimum detection limit of 4.5 ppm. Furthermore, the compact structure and extended optical interaction length offered by the HC-PCF significantly improve the practicality of online transformer condition monitoring. The results confirm that photonic crystal fiber-based sensing technology offers a promising solution for next-generation real-time power system monitoring applications. Owing to its high sensitivity, compactness, and capability for continuous online operation, the proposed system demonstrates strong potential for deployment in intelligent grid monitoring and predictive maintenance of high-voltage electrical equipment.

Abstract: Background: Non–small cell lung cancer (NSCLC) is one of the most commonly diagnosed malignancies and a leading cause of cancer-related mortality worldwide. Although immune checkpoint inhibitors such as nivolumab have become a standard option in the second-line and later-line settings, identifying patients who derive meaningful benefit remains challenging. Established biomarkers, including PD-L1 expression, have shown limited predictive performance. The Lung Immune Prognostic Index (LIPI), a simple score reflecting systemic inflammation and tumor burden, has emerged as a potential prognostic tool. However, its real-world prognostic value in patients receiving nivolumab after first-line therapy remains insufficiently defined. Methods: This retrospective, single-center study included 109 patients with advanced NSCLC who received nivolumab following disease progression after first-line therapy. LIPI was calculated based on derived neutrophil-to-lymphocyte ratio (dNLR ≥3) and lactate dehydrogenase (LDH) levels above the upper limit of normal. Patients were stratified into three groups (LIPI 0-1-2). Progression-free survival (PFS) and overall survival (OS) were analyzed using the Kaplan–Meier method and compared with the log-rank test. Multivariable Cox proportional hazards models were used to identify independent prognostic factors. Results: Patients were categorized as LIPI 0 (n=32), LIPI 1 (n=45), and LIPI 2 (n=32). A clear stepwise deterioration in outcomes was observed across increasing LIPI categories. Median PFS was not reached in the LIPI 0 group, and was 340 days and 109 days in the LIPI 1 and LIPI 2 groups, respectively (p < 0.001). Similarly, median OS was not reached in the LIPI 0 group, while it was 820 days and 201 days in the LIPI 1 and LIPI 2 groups, respectively (p < 0.001). In multivariable analysis, LIPI remained independently associated with both PFS and OS. In contrast, PD-L1 expression and other clinicopathological variables were not independently associated with survival outcomes, suggesting that the prognostic impact of LIPI is independent of PD-L1. Conclusions: LIPI appears to be a robust, readily available prognostic marker in patients with advanced NSCLC treated with nivolumab beyond the first-line setting. Its independent association with survival outcomes, irrespective of PD-L1 expression, suggests that it captures distinct biological processes related to systemic inflammation. Incorporation of LIPI into routine clinical practice may improve risk stratification and support more individualized treatment decisions. Notably, LIPI may provide complementary prognostic information beyond tumor-specific biomarkers such as PD-L1.

Abstract: Social service professionals operate in legally sensitive, administratively intensive, and context-dependent environments in which decision-making requires the simultaneous interpretation of regulatory norms, institutional procedures, and individual case circumstances. This paper proposes a conceptual model of a Personal Legal and Social Artificial Intelligence (AI) Assistant intended to support professional decision-making in social services, and demonstrates its functionality through a working prototype. The model is formulated as a domain-specific retrieval-augmented generation (RAG) framework in which a controlled legal and social document corpus is processed through text extraction, chunking, semantic indexing via SentenceTransformer embeddings, top-k retrieval through cosine similarity, and bounded large-language-model reasoning to produce grounded and explainable responses. The proposed framework is informed by three successive prototype versions and by observed sensitivity to corpus scope, document prioritization, and prompt constraints. The current prototype version operates on a prioritized corpus of sixteen Bulgarian normative acts complemented by three supplementary resources, comprising 883 indexed fragments, and uses DeepSeek as the reasoning model accessed through the OpenRouter API. The functionality of the model is validated through a representative use case concerning child protection, in which the prototype identifies the applicable legal provisions, exposes the retrieved documentary evidence, and generates a four-part structured analysis comprising legal qualification, applicable provisions, legal consequences, and recommendations for action. The main contribution lies in the formalization and prototype-level demonstration of a domain-specific AI assistant that combines legal grounding, social-context awareness, and bounded language-model reasoning for trustworthy decision support in regulated social-service practice.

Abstract: The particular microstructure of hot-deformed Nd-Fe-B magnets leads to difficulties in finding a direct recycling route. In this work, a combination of field assisted sintering technology/spark plasma sintering (FAST/SPS) and spark plasma texturing (SPT) are used as pre-compaction and deformation techniques, respectively, for the consolidation of crushed, hot-deformed Nd-Fe-B scrap. Field assisted sintering has the unique advantage of maintaining fine microstructures during material densification, making it an ideal candidate for direct recycling of this material. Recycled magnets, made from 100 wt% crushed magnet scrap, were able to achieve energy products of over 200 kJ m^-3 after FAST/SPS pre-compaction and SPT deformation. These recycled magnets could then be smoothed and cut to the size of industrial bar magnets for testing in the motor of a water pump. When tested, the recycled magnets could achieve 95% of the electromotive force compared to industrial standard magnets.

Abstract: Network intrusion detection systems (NIDS) play a crucial role in modern network environments where diverse and rapidly evolving traffic patterns are observed. Although deep learning-based NIDS have demonstrated strong performance within specific datasets, their effectiveness significantly degrades when applied to unseen network environments due to domain discrepancies. In this paper, we first experimentally demonstrate the performance degradation of time-series-based NIDS under cross-domain conditions using multiple benchmark datasets. Then, we propose a LoRA-based domain adaptation framework for time-series-based NIDS models. Instead of retraining the entire model, the proposed approach freezes the backbone network and applies low-rank updates to selected layers, enabling parameter-efficient adaptation to new domains. Experimental results show that the proposed method consistently improves cross-domain detection performance across multiple dataset combinations, particularly in terms of recall, while requiring only a small number of additional parameters.

Abstract: This study assessed the effect of dietary supplementation with Kluyveromyces lactis (K. lactis) hydrolysate (HKL) on growth performance, apparent nutrient digestibility, systemic immune–antioxidant status, and fecal microbiota in weaned piglets. A total of fifty-four piglets, with an initial body weight of 6.07 ± 0.086 kg and age of 25 ± 1 days, were randomly assigned to three dietary treatments over 28 days (6 replicates per treatment; 3 piglets per replicate): a control diet (CON), CON supplemented with 5 g/kg HKL (HKL1), or CON supplemented with 10 g/kg HKL (HKL2). Throughout the trial, growth performance was monitored, apparent total tract digestibility of nutrients was determined, serum samples were collected for immune and antioxidant assessments, and fecal samples were gathered for microbiota analysis. The results indicated that compared to the CON group, both HKL1 and HKL2 groups exhibited improved growth performance, by as evidenced by increased average daily gain (ADG) and average daily feed intake (ADFI) from day 0 to 28 (P < 0.05). Furthermore, HKL2 significantly enhanced ADG from day 0 to 14 and reduced the feed-to-gain ratio (F:G) during the same period (P < 0.05). Diarrhea incidence was markedly decreased by HKL supplementation at both day 14 and day 28 (P < 0.001). HKL supplementation increased the apparent digestibility of dry matter, ash, calcium and phosphorus (P < 0.05). On day 14, serum immunoglobulin A (IgA) levels were elevated, while malondialdehyde (MDA) levels were reduced in HKL-supplemented piglets (P < 0.05). By day 28, serum total protein and immunoglobulin G (IgG) concentrations, as well as superoxide dismutase (SOD) and glutathione peroxidase (GSH-Px) activities, were increased, while interleukin-6 (IL-6) levels were decreased in HKL-supplemented piglets (P < 0.05), suggesting HKL possesses immunomodulatory and antioxidant regulatory capacities. HKL also enriched several health-associated commensal bacteria, including [Eubacterium]_xylanophilum_group, unclassified_f_Peptostreptococcaceae, Candidatus Saccharimonas, Erysipelotrichaceae_UCG-003, and Negativibacillus, suggesting a micro-biota-modulatory effect in weaned piglets. These results indicate that dietary supple-mentation with HKL could improve growth performance and nutrient utilization, reduce post-weaning diarrhea, and promote a more favorable immune–antioxidant status and microbial profile in weaned piglets, with the 10 g/kg dosage demonstrating greater overall efficacy. These findings provide a theoretical basis for the development of novel K. lactis products and the application of K. lactis hydrolysate in weaned piglets.

Abstract: The purpose of this study is to investigate the extent to which tourists appreciate sus-tainable tourism and what effect eco-friendly practices have on the decision-making process of selecting a hotel. Through the use of large-scale analysis of online reviews of hotels and the application of sentiment analysis techniques, the research investigates the impact of environmental factors (e.g., energy usage reduction, minimizing waste, and promoting nature experiences) on customer perspectives and decision-making processes for lodging. This research adopts an approach that utilizes machine-learning based sentiment analysis as its source of understanding. The results of this research demonstrate that as more individuals become aware of sustainable tourism; however, sustainability often plays a secondary role in determining whether or not to stay at a specific hotel when compared to such lodging attributes, as comfort, price and quality service. Based upon these findings, this research indicates that while many tourists’ value sustainable tourism and make an effort to choose eco-friendly lodging establishments, the influence of sus-tainability on tourists' lodging decisions is not as strong as other attributes. These results indicate important implications for hotel managers that will help them to balance envi-ronmental stewardship with a competitive stance.

Abstract: Reef-building corals form the calcium-carbonate frameworks that underpin tropical coral reefs, yet global coral cover has declined by ~50% in recent decades, due to marine heatwaves and other stressors. Identifying refugia environments, such as upwelling systems, that buffer stress, promote recovery, and enhance resilience by promoting physiological plasticity that supports thermotolerance is therefore critical. Here, we compared benthic community composition, coral percent cover, and photo-physiology between an upwelling location in the Gulf of Papagayo and a non-upwelling location in Sámara on the Pacific coast of Costa Rica. Waters in Papagayo were cooler, more acidic, and richer in chlorophyll a. Reefs at this location exhibited higher crustose coralline algae, higher sea urchin cover, and lower macroalgae cover, compared to Sámara. Papagayo also showed higher stony coral cover, driven by Pocillopora spp., while Sámara was dominated by massive, heat-tolerant Porites spp.. When significant, photophysiological measurements showed 9.7 - 44.5% higher photosynthetic efficiency (Fv'/Fm') in Papagayo corals and 19.94 - 42.75 % higher maximum photosynthetic rates (P_max) in Sámara corals. These results highlight how contrasting environmental regimes within a relatively small geographic area can shape distinct coral community compositions and photophysiological strategies, with implications for identifying areas of reef persistence or refugia.

Abstract: In this paper, the dynamic modeling of the anisotropic non-ideal gyroscopic rotor system is considered. The equations of nonstationary transitions are derived from the motion differential equations, from there the control equation and stationary frequency dependencies, force and energy relations. When the rigidity of the elastic support is anisotropic in orthogonal directions, two critical velocities and, accordingly, two resonance regions are found. Because of the strong interaction of the rotor system with a non-ideal DC motor, slopes of the resonance curves are observed in the regions of critical speeds even in the absence of a non-linear component of the reference stiffness, and loops. It has been proven that the cubic nonlinearity of damping strongly suppresses the resonant amplitudes of the rotor, reduces the size of the loops even more, and strongly attenuates the Sommerfeld effects until they are completely eliminated than linear damping. It is shown that an increase in the magnitude of the cubic nonlinearity of damping greatly facilitates the passage of the resonance region and expands the range of operating speeds. This proves that the amplification of linear damping with cubic nonlinearity of damping is one of the methods for controlling resonant passages and an effective damping model.

Abstract: A 79-year-old man with prostate cancer underwent initial staging prior to prostatectomy with 99mTc-methylene diphosphonate (99mTc-MDP) bone scintigraphy. Anterior and pos-terior images showed focal uptake overlying the pubic symphysis. Lateral views showed that the activity was extraosseous. Follow-up CT urography showed a bladder hernia as cause of the abnormality on bone scan. Prostatectomy and inguinal hernia repair were performed as a combination case. Four years postoperatively, follow-up ⁶⁸Ga-PSMA-11 positron emission tomography/computed tomography (PET/CT) showed no recurrence. The CT component of the exam show an intermediate density focus at the right inguinal hernia repair site, corresponding to a polypropylene mesh plug, and a hyperattenuating Gore-Tex mesh repair of the left inguinal hernia. This case highlights the importance of lateral projections in resolving scintigraphic pitfalls and recognizing mesh-related im-aging appearances to prevent misinterpretation.

Abstract: This work formulates semiclassical gravity within a causal-diamond framework where a finite-resolution boundary provides the edge structure for a local Wheeler--DeWitt description. Because the diffeomorphism-invariant Hilbert space does not factorize, each diamond is equipped with a boundary-completed algebra $\mathfrak{A}_O$, ensuring the operational state $\rho_O$ and the semiclassical reference family $\sigma_O[\Lambda]$ share identical operator content. Dynamics are posed as local statistical inference: the relative-entropy functional $S_{\rm rel}(\rho_O\|\sigma_O[\Lambda])$ quantifies the mismatch between data and reference. This yields the minimal operational axioms defining subsystems, intrinsic clocks, and regulated observables in a finite-resolution, background-independent setting.The topology-locked boundary capacity budget fixes an effective channel multiplicity $N \approx 1.23\times10^{11}$. Calibrating its coherent fraction to Newton's constant determines a matching scale $M_s \approx 3.02\times10^{13}\,{\rm GeV}$. In the modular/KMS regime, the relative-entropy Hessian (Kubo--Mori metric) block-diagonalizes into orthogonal tensor, vector, and scalar sectors. A quasi-local heat-kernel expansion on the fixed $S^3\times S^1$ history manifold maps this near-equilibrium response to a matching-scale EFT, yielding the Einstein--Hilbert tensor structure, Yang--Mills susceptibilities, and leading mass deformations. Vector and scalar responses remain intensive, while the tensor response scales extensively.The reduced scalar-curvature sector fixes the $R^2$ plateau coefficient, equivalently locking the scalaron scale to $M_R=M_s$. The fixed modular protocol and quantized boundary currents give the matching-scale gauge condition $\alpha^{-1}(M_s)=4\pi k$ for normalized current blocks. In the selected Standard Model boundary inventory this yields integer current levels for the strong and electromagnetic sectors, a weak-mixing projection, and a separate static electromagnetic susceptibility on $S^3\times S^1$. The same boundary capacity gives linked plateau targets $n_s\simeq0.965$, $r\simeq0.0038$, and $A_s\simeq2.1\times10^{-9}$, while single-pixel saturation gives electroweak-scale consistency checks for the Higgs, vacuum expectation, and top-sector scales.Translations between causal diamonds act as completely positive trace-preserving (CPTP) updates. The resulting open-modular Walsh filtration algebraically isolates a threefold matter-generation carrier and lepton-mass accessibility invariants. Treating continuum fields as the response of a finite boundary, the framework yields correlated, falsifiable relations for gravitational stiffness, gauge response, plateau cosmology, electroweak saturation, and threefold matter-sector organization from one minimal operational architecture.

Abstract: Automatic detection of stress from social-media text holds promise for digital mental health, but most existing Transformer-based approaches are opaque and computationally demanding. This work presents DR-Transformer, a Dual-Regularized Transformer that combines two complementary mechanisms: (i) a group sparsity penalty (L_2,1/L₂ elastic net) applied to the query and key projection matrices of every attention head, which encourages whole-row sparsity for token-level interpretability; and (ii) a supervised contrastive loss on the [CLS] projection, which organizes the latent space according to the stress label. The architecture is intentionally lightweight (6 layers, 8 heads, 256-dim embeddings; ∼9.5M parameters) and runs entirely on consumer-grade hardware (NVIDIA GTX 1660, 6 GB). Experiments on the publicly available Dreaddit dataset (binary stress classification, 2,838 train / 715 test segments) compare DR-Transformer against logistic regression, BiLSTM, a standard Transformer of identical architecture, and MentalBERT. Across five seeded runs, DR-Transformer (Full) reaches F₁ = 0.876 (bootstrap 95% CI 0.852–0.898), outperforming the Standard Transformer (F₁ = 0.842; McNemar p < 0.001 with Bonferroni correction) and performing comparably to the much larger MentalBERT (F₁ = 0.879; p = 0.421). Sparse regularization increases the fraction of near-zero attention weights (below 0.01) from 0.215 to 0.682, yielding visibly focused attention maps, while the supervised contrastive loss improves the silhouette score of [CLS] embeddings from 0.312 to 0.483. Dual regularization thus combines accuracy, interpretability, and efficiency in a single model trainable without specialized infrastructure.

Abstract: Artificial intelligence systems for pneumonia detection often achieve strong predictive performance but remain insufficiently calibrated, weakly interpretable, and poorly aligned with clinically meaningful decision-support requirements. This paper presents a diagnostic-field extension of the Unified Variational Intelligence Framework (UVIF) for trustworthy and decision-centric pneumonia screening using chest X-ray imaging. The proposed framework models diagnosis as a variational process in which imaging patterns and latent feature representations are treated as diagnostic fields that must be sensed, filtered, interpreted, and evaluated before clinical decision support is produced. The study combines compact convolutional neural network modeling, embedding-based machine learning classifiers, calibration-aware reliability analysis, threshold-sensitive decision control, and multi-level explainability using Grad-CAM, LIME, and SHAP. Experimental evaluation is conducted on the publicly available PneumoniaMNIST benchmark dataset from the MedMNIST collection. The compact CNN achieved strong discrimination performance with ROC-AUC of 0.9666 and pneumonia recall of 0.9974, while the UVIF-guided diagnostic layer supported reliability-aware model selection and threshold optimization under screening-oriented constraints. Calibration analysis further revealed deviations between predicted probabilities and empirical outcomes, emphasizing the importance of reliability-aware evaluation in medical AI systems. The proposed framework demonstrates that integrating prediction, calibration, explainability, and diagnostic decision control within a unified variational framework can support more transparent, interpretable, and clinically meaningful AI-assisted pneumonia screening systems.

Abstract: Clinical large language model (LLM) agents are increasingly engineered as systems that combine a language model backbone with memory, tools, orchestration loops, and feedback mechanisms. In this setting, the key engineering question is no longer only what the backbone model can answer, but how the surrounding harness stores experience, retrieves context, orchestrates tools, and converts feedback into reusable knowledge. Existing reviews of LLM agents in healthcare primarily emphasise prompting strategies, task capabilities, and benchmark performance, leaving these harness-level mechanisms insufficiently synthesised. This review addresses that gap by organising the emerging literature under the term Self-Evolving Agent Engineering (SEAE), defined here as a harness-level design paradigm centred on three recurring mechanisms: persistent cross-session memory, autonomous skill or experience synthesis, and closed-loop feedback-driven improvement. We review 148 references and 23 representative clinical systems across six task categories, using radiology as the main translational focus. Rather than treating these systems as isolated applications, we map how persistent memory, skill synthesis, tool orchestration, and feedback-driven improvement are implemented across current healthcare agents, and examine the technical, clinical, and regulatory challenges that arise when clinical agents are designed to evolve across sessions.

Abstract: The 2030 Agenda highlights education and entrepreneurship as critical drivers of sustainable development, yet little is known about how sustainability literacy translates into green entrepreneurial confidence among Gen Z students in developing and post-conflict economies. This study examines the direct and mediated effects of sustainability literacy on green entrepreneurial self-efficacy (GESE) through biospheric values, and the moderating roles of university support, digital literacy, and family support. A cross-sectional survey was conducted with 417 Gen Z undergraduate students from Universitas Negeri Padang (Indonesia) and Universidade Nacional Timor Lorosa’e (Timor-Leste). Data were analyzed using PLS-SEM and Importance-Performance Map Analysis (IPMA). Results show that sustainability literacy directly enhances GESE (β = 0.342, p < 0.001) and indirectly through biospheric values (indirect effect = 0.156, p < 0.001). University support moderates the values→efficacy pathway (β = 0.148, p < 0.05), while digital literacy moderates the literacy→efficacy pathway (β = 0.198, p < 0.01). However, family support did not moderate any relationship, and digital literacy exhibited a ceiling effect among Gen Z respondents. IPMA reveals biospheric values and sustainability literacy as high-importance, high-performance priorities, with no urgent intervention needed. We conclude that cultivating biospheric values matters more than transmitting knowledge alone, and university support should strategically target value-driven students rather than compensate for low literacy.