East West University

BackgroundForecasting influenza outbreaks remains a significant challenge due to the complexity of disease transmission and the influence of environmental and behavioral factors. Traditional models based solely on the basic reproduction number often fall short in capturing the full scope of outbreak dynamics.MethodsIn this study, we employ a seasonally adjusted SEIRT model incorporating stochastic differential equations (SDEs), including Brownian motion and Lévy jump processes, to simulate random and abrupt fluctuations in transmission. A branching process approximation is used to evaluate the probability of an epidemic under the influence of seasonal variability and stochastic perturbations. The model is calibrated using weekly influenza case data from Mexico, with noise components estimated from publicly available CDC [1] and WHO [2] surveillance data.ResultsSimulation results show that the inclusion of stochastic effects and periodic transmission rates significantly enhances the model's accuracy in reflecting real-world epidemic dynamics. Numerical comparisons between deterministic, Brownian-based, and Lévy-based scenarios reveal that both the initial state of the exposed or infectious subpopulation and the seasonal transmission patterns are critical to determining outbreak probabilities. Results indicate that seasonal transmission rates and stochastic effects significantly alter epidemic probabilities, with Lévy processes capturing abrupt outbreak dynamics more accurately than deterministic models.ConclusionsThe findings underscore that deterministic models may underestimate epidemic risk when they overlook random and sudden changes in contact rates or disease introduction. The proposed stochastic modeling framework yields a deeper understanding of influenza transmission dynamics by incorporating uncertainty and seasonal variability, thereby supporting more informed and effective public health decision-making.

Abstract Bangladesh, a highly climate-vulnerable nation, is increasingly confronting climate change challenges, especially in the coastal areas. This research investigates the knowledge, abilities, and practices of primary teachers related to Climate Change Education (CCE) within the national curriculum framework. Drawing upon a qualitative phenomenology, in-depth interviews with 30 teachers in Shyamnagar sub-district, Satkhira district, a highly climate-hazard-affected area reveal that climate change content is lacking chronological integration in discipline-based learning areas like Science, Bangladesh & Global Studies, and Religion. The teachers’ understanding was found to vary, with a predisposition towards learning areas based on patchy textbook materials and sparse professional development. Even with acknowledgment of the significance of CCE in promoting climate resilience and environmental literacy in students, teachers point to the lack of systematic guidance, localized materials, and capacity development programs. This research emphasizes the necessity of revamping the curriculum to embed integrated, location-based, and age-relevant climate education. It stresses the need to provide targeted teaching capacity building to ensure pedagogic competence to deliver CCE. The findings of the research have significant policy implications for aligning the National Education Policy of Bangladesh with actual implementation for enabling the empowerment of future generations to address climate challenges. The paper contributes to policy deliberations for including sustainable development and disaster risk reduction practices in the primary school curriculum and textbooks.

We present MobiPhysio, a 2D video-based dataset designed to support AI-driven physiotherapy assessment and monitoring. The dataset contains 3686 segmented videos of 9 Active Range of Motion physiotherapy exercises performed by 58 male and female participants. The recordings are done under the variations in lighting, camera angles, occlusion, and jitter in order to mimic real-world conditions. Data collection occurred in two phases: first from non-expert participants at Stamford University Bangladesh, and later from expert participants at the Department of Physiotherapy and Rehabilitation, Jashore University of Science and Technology. The entire process was conducted under the guidance of certified physiotherapists. Each video is further annotated with assessment scores derived from the exercise-specific Exercise Accuracy Assessment Questionnaire (EAAQ), developed under expert guidance. This dataset would enable researchers to build and test AI-powered physiotherapy and rehabilitation systems, examine human motion, and create exercise monitoring solutions using available 2D camera devices like mobile phones without the need of external body-reliant sensors.

Introduction: Chronic kidney disease (CKD) represents a significant disorder that affects multiple vital organs and physiological systems, with endocrine dysfunction being a common associated complication. The altered metabolic milieu in CKD affects the secretion of hormones and the response of target tissues, causing endocrine dysfunctions. Early identification of patients with chronic kidney disease (CKD) may help implement interventions to decrease progression and eventual morbidity and mortality. Most previous studies investigating testosterone abnormalities and renal disease have concentrated on patients with end-stage renal disease (ESRD), and little is known about the problem’s pervasiveness in patients in the pre-dialysis phase. Aim of the Study: This study aimed to evaluate the serum testosterone level of adult males with different stages of CKD and its association with body mass index, and comparison with age-matched healthy individuals. Methods: This cross-sectional study was conducted in the Department of Nephrology, Dhaka Medical College Hospital, Dhaka, Bangladesh. All CKD patients were designated as group ‘A’ with 90 cases; the healthy control group were grouped as group ‘B’ with 88 cases. Statistical analyses were conducted using the Statistical Package for Social Sciences version 16.0. Results: In this study, the majority 40(44.4%) of patients had glomerulonephritis, 24(26.8%) had DM and 11(12.2%) had HTN. The mean BMI was found to be 20.85±2.7 kg/m2. More than two-thirds (66.7%) of patients had normal (3.2-14.6 ng/ml) Serum testosterone, and their mean Serum testosterone was found at 4.29±1.9 ng/ml. Negative Spearman’s rank correlation (r=-0.893; p=0.001) was observed between different stages of CKD and S. Testosterone of CKD patients. A positive Pearson correlation existed (r=0.517; p=0.001) between BMI and S. Testosterone of CKD patients. Conclusion: A significant positive correlation existed between BMI and serum testosterone in CKD patients. Future studies are needed to determine the potential adverse effects of low serum testosterone levels in patients with chronic kidney disease and to determine whether the therapeutic intervention of low serum testosterone in males with chronic kidney disease.

<title>Abstract</title> <bold>The development of high-efficiency and stable per- ovskite solar cells (PSCs) requires both suitable material selection and advanced predictive modeling. In this study, Rb</bold> _{<bold>2</bold>} <bold>LiGaI</bold> _{<bold>6</bold>} <bold>, a halide perovskite with a bandgap of 1.13 eV, is explored as the absorber material due to its high absorption coefficient (</bold> <italic> <bold>&gt;</bold> </italic> <bold>105 cm</bold> −1 <bold>), low carrier effective masses (∼0.2–0.4</bold> <italic> <bold>m</bold> </italic> <bold>0), and thermodynamic stability. Device architecture optimization was performed using a machine learning (ML) framework to predict photovoltaic performance metrics, including</bold> <italic> <bold>J</bold> </italic> <italic>sc</italic> <bold>,</bold> <italic> <bold>V</bold> </italic> <italic>oc</italic> <bold>, FF, and PCE, under varying ETL/HTL interface configurations. Among the tested models, Random Forest achieved the best predictive accuracy R</bold> 2 <bold>of 0.9974 surpassing XGBoost, HistGBR, and KNN. Correlation and SHAP analyses revealed that HTL bandgap, conduction band offset, electron affinity, and hole mobility play critical roles in governing efficiency. Device simulations showed that absorber thickness strongly influences PCE, to a maximum of 22% at 1000 nm which is very close approximation of the ML predicted PCE of 22.13% with only a prediction error of 0.6%. Optimization of transport layers demonstrated that a thin ETL (∼50 nm) avoids parasitic absorption losses, while moderate HTL thickness (∼100–150 nm) enhances</bold> <italic> <bold>V</bold> </italic> <italic>oc</italic> <bold>. Temperature- dependent studies further confirmed that device efficiency drops drastically due to elevated dark current. These findings highlight the potential of combining Rb</bold> _{<bold>2</bold>} <bold>LiGaI</bold> _{<bold>6</bold>} <bold>absorbers with ML-guided optimization to design efficient and stable PSC architectures.</bold>

Background: Antenatal care (ANC) plays a critical role in improving maternal and neonatal health by enabling early detection and management of pregnancy-related complications. However, ANC utilization remains inadequate in many developing countries due to insufficient knowledge, unfavorable attitudes, and poor health-seeking practices. Objective: To assess knowledge, attitude, and practices related to antenatal care and to identify factors influencing ANC utilization among pregnant women attending a tertiary-level hospital in Dhaka, Bangladesh. Methods: A hospital-based cross-sectional study was conducted from January to June 2025 at the Antenatal Outpatient Department of Monowar Sikder Medical College and Hospital, Dhaka. Using consecutive sampling, 350 pregnant women were enrolled. Data were collected through a structured questionnaire and analyzed using descriptive statistics and Chi-square tests (p &lt; 0.05). Results: Most participants were aged 21–30 years (77.1%), and 66.3% were literate. Knowledge of nutritional supplementation during pregnancy was high (93.1%), while awareness of gestational diabetes (11.4%) and recommended investigations (21.7%) was low. ANC utilization was suboptimal, with 40.0% reporting no ANC visits and only 2.3% completing four or more visits. Although 79.7% reported good accessibility, satisfaction with ANC services was moderate (48.3%), and 82.0% required spousal permission to attend clinics. Conclusion: Significant gaps in ANC knowledge and utilization persist despite generally positive attitudes. Targeted health education, improved service delivery, and greater family involvement are essential to enhance ANC uptake and maternal–neonatal outcomes.

Federated fine-tuning of large language models (LLMs) allows different participants to fine-tune a shared model, without any of these participants accessing the raw data. Rather than having sensitive information concentrated, parties train locally and provide only model updates, which enables the system to learn through diverse sources of data without breaching privacy and ownership of data; however, it is severely vulnerable to two potential attacks, namely (1) malicious clients may actively inject poisoned gradients in training and hence undermine integrity of the global model. and (2) accidental memorization and leakage of personally identifiable information (PII) during model updates. Traditional defense methods such as powerful aggregation schemes (e.g. Krum), and differential privacy can either protect against adaptive adversaries with limited protection or introduce substantial utility loss in large scale training of LLM. We can resolve this by indicating a privacy related federated framework called VeriFed with proposing a client to suitprove that it satisfies, under agreed upon conditions of privacy and safety, succinct cryptographic non-interactive arguments of knowledge (SNARKs). These are ℓ2 sensitivity bounded (to be DP-compliant), minimal embedding of PII tokens and a structured parameter efficient architecture (e.g., LoRA). Noteworthy, actual evidence is generated without disclosing the crude gradients or data, and secrecy is thus kept. We have a verification algorithm, using gradient drawing, hierarchical break down of restrictions, allowing scale reductions of up to billions of constraints in high-dimensional, high-order verification. VeriFed, evaluated on PubMedQA and Financial PhraseBank across five state-of-the-art LLMs, shows VeriFed achieves close utility, with a performance loss of less than 1% compared to FedAvg, and backdoor attacks success rates of 1.8%, corresponding to a suppression of over 98%. The leakage of privacy is reduced to a minimum, exposure to personally identifiable information (PII) = 0.08, which is 89% lower (LiRA = 0.08). Generation of proofs is still computationally viable, and takes less than 19 seconds to generate proofs with models of up to 33 million parameters. On the whole, VeriFed makes collaboration between LLM fine-tuning and jointly held accountability verifiable, because it is based on cryptographic assurances of reliability and not solely on statistical credibility.