With the accelerating aging of the global population, the importance of health monitoring and assistive technologies for the elderly is becoming increasingly prominent. This study aims to propose and validate a biomorphic design framework based on vascular biomimetic structures to address the challenges of mechanical adaptability, signal stability, and cognitive interaction in age-friendly wearable devices. Using microfluidic lithography and differential capillary self-assembly, a liquid metal flexible circuit with a fractal topology mimicking human microvasculature is constructed on a SEBS/TPU composite substrate. A biomimetic microhinge array is used to disperse stress, and micropits are laser-engraved on the surface to enhance breathability. The display interface, based on a poly(3,4-ethylenedioxythiophene) microcapsule array, combines electrochromic and photoluminescent effects to map parameters such as blood pressure and blood oxygen into dynamic fractal growth patterns. Experimental data show that the biomimetic circuit exhibits a resistance change of only 11.72% under a strain of only 15%, and the substrate stiffness pressure is 0.82±0.07 kPa, ensuring signal stability and comfortable wear during exercise. The user experience is significantly improved, with a low discomfort feedback rate of 7.2%. The conclusion shows that biomorphic design based on vascular bionic structure provides a suitable and systematic solution for the physiological adaptation and cognitive interaction of elderly-friendly wearable devices.

Arrhythmia stands as a primary contributor to cardiovascular disease-associated mortality. Therefore, the classification and monitoring of abnormal electrocardiogram (ECG) signals are of paramount importance for preventive purposes. Although deep - learning - based ECG classification methods have yielded promising outcomes, they frequently encounter challenges in optimizing performance across diverse patient datasets. To overcome these limitations, this research endeavors to enhance the generalization ability of deep - learning models for ECG signal classification. It achieves this by integrating structural risk minimization principles and incorporating RR interval information into the classification process. A convolutional neural network (CNN) founded on structural risk minimization is proposed. Instead of employing the traditional cross-entropy loss, this study adopts a loss function inspired by support vector machine (SVM) classifiers to optimize the CNN. Moreover, the RR interval information, which is often lost during beat segmentation, is manually extracted and integrated into the CNN network to improve classification accuracy. The proposed method attains an accuracy, specificity, and sensitivity of 88.2% respectively, demonstrating superior performance when compared to traditional and existing methods. This improvement underscores the efficacy of the structural risk minimization approach and the integration of RR interval information in enhancing the model's generalization across patient datasets. The method's convenience and effectiveness render it particularly well-suited for real-time application in wearable devices, facilitating the early detection of abnormal ECG patterns and potentially preventing cardiovascular disease-related fatalities.

OBJECTIVE: Leveraging multimodal data from the 2005-2023 National Health and Nutrition Examination Survey (NHANES) database, this study aims to develop a predictive method for the geriatric depression that combines high predictive accuracy with good interpretability, thereby providing support for in-depth exploration of the pathogenesis and risk factors of geriatric depression. METHODS: Data from 8760 participants aged 65 and older in the NHANES database from 2005-2023 are utilized to develop and validate the stacking ensemble predictive model. Depression is assessed using the Patient Health Questionnaire-9 (PHQ-9) total score meeting or exceeding 10. Before the model construction, this work employs the normalization of training data and test data, Synthetic Minority Over-sampling Technique - Random Under-Sampling (SMOTE-RUS) hybrid sampling strategy to address the class imbalance, and the recursive feature elimination method based on the random forest (RFE-RF) for feature selection. A stacking ensemble predictive framework for depression is constructed based on the primary learners (Random Forest, SVM, XGBoost, and Logistic Regression) and meta-learners (SVM and Logistic Regression). Finally, the interpretable machine learning technique SHapley Additive exPlanations (SHAP) is used to visualize the model predictive outputs. RESULTS: The XGBoost model demonstrated outstanding performance on the test set in terms of AUC (83.92%), while the Random Forest (RF) model excelled in sensitivity (71.05%). Subsequently, a specifically designed RFE-Stacking ensemble model, using RF and XGBoost as the primary learner and the SVM as the meta-learner, is developed. In comparison, this stacking ensemble model exhibits the best predictive performance with the biggest AUC (85.14%) and the highest sensitivity (78.71%). The SHAP interpretation reveals that general health condition, frequency of oral pain in the past year, marital status, history of mental health consultations in the past year, and frequency of urine leakage are the top five most influential factors in predicting the depression risk. CONCLUSION: This stacking ensemble model enhances the performance of both the primary learners and the meta-learners. This verifies the feasibility and effectiveness of the proposed model in predicting the geriatric depression. This work integrating the stacking ensemble model with SHAP offers valuable clinical references for assessing the risk of depressive symptoms, which is beneficial to develop the personalized depression interventions and preventions in the elderly.

This study investigates the impact of integrating knowledge graph prompt engineering (KGPE) with large language models in the context of medical question answering. The Hugging Face MedQA dataset (N = 5,000) was utilised for the extraction of key medical entities via the implementation of named entity recognition, and the construction of SPARQL-based relational prompts from the knowledge base of Wikipedia to guide the reasoning process. Two models, Llama-2-7B-chat-hf and Qwen-2-7B-Instruct, are evaluated through a weighted aggregation of BLEU, ROUGE, and cosine similarity metrics. The findings demonstrate that Qwen-2-7B-Instruct attains substantial enhancements under KGPE—BLEU escalating from 0.366 to 0.531 (+0.165) and cosine similarity rising from 0.763 to 0.820 (+0.057). Conversely, Llama-2-7B-chat-hf exhibits a modest decrease, signifying divergent responsiveness to structured knowledge. These findings demonstrate that integrating structured knowledge through KGPE enhances factual accuracy and semantic coherence in medical reasoning without modifying model architecture

The escalating prevalence of chronic stress necessitates development of highly personalized non-pharmacological interventions. Traditional art therapy lacks real-time adaptability to match individuals' fluctuating physiological states. This paper introduces the Bio-Aesthetic Resonator (BAR), a novel closed-loop immersive art therapy system driven by real-time physiological feedback. The BAR integrates biosensors for continuous monitoring of Heart Rate Variability (HRV) and Galvanic Skin Response (GSR), utilizing deep reinforcement learning (DRL) to dynamically generate immersive visual and auditory artscapes. We conducted a pilot randomized controlled trial (pRCT) with 60 participants with mild-moderate anxiety. The BAR intervention significantly reduced perceived stress scores (PSS-10: Cohen's d = 0.65, 95% CI [0.22, 1.08], p = 0.003) and increased high-frequency HRV (HF-HRV: Cohen's d = 0.92, 95% CI [0.48, 1.36], p < 0.001) compared to a sham-adaptive control. These results support bio-aesthetic resonance as a viable framework for personalized digital therapeutics.

INTRODUCTION: Pathological complete response (pCR) following neoadjuvant chemotherapy (NAC) is a validated surrogate endpoint for long-term survival in breast cancer patients. However, conventional biomarkers exhibit limited predictive accuracy, with approximately 60-80% of patients failing to achieve pCR. Dynamic contrast-enhanced MRI (DCE-MRI) provides high-resolution information on tumor vascularization and heterogeneity, but prior radiomics models have predominantly relied on single-feature paradigms, which may not fully capture complex tumor phenotypes. METHODS: We developed a multimodal deep-learning radiomics (DLR) pipeline using the publicly available ACRIN 6657/I-SPY1 dataset (n=163). After rigorous preprocessing (bias-field correction, isotropic resampling, Z-score normalization), we extracted a comprehensive set of 1,702 standardized radiomics features compliant with the Image Biomarker Standardization Initiative (IBSI), which quantitatively capture tumor morphology, texture, and intensity patterns. Additionally, 8,576 deep learning features were derived from five convolutional neural networks (ResNet50, DenseNet-169, InceptionV3, InceptionResNetV2, EfficientNetB0), enabling the model to learn complex, data-driven representations beyond human-defined features. The fusion of these complementary feature types provides a more holistic characterization of tumor phenotype, significantly enhancing predictive performance compared to single-modality approaches. A two-stage feature-selection strategy utilizing univariate analysis and the Least Absolute Shrinkage and Selection Operator (LASSO) algorithm was applied, followed by linear signature construction. Ten classifiers were evaluated under stratified cross-validation and independent testing. RESULTS: The fusion of handcrafted radiomics and deep learning features significantly enhanced predictive performance. The best-performing model, a multilayer perceptron (MLP), achieved an area under the receiver operating characteristic curve (AUC) of 0.98 on the independent test set, with an accuracy of 95.92%, sensitivity of 92.86%, and specificity of 97.14%. Logistic regression also demonstrated strong performance (AUC = 0.980). Decision curve analysis confirmed the clinical utility of all models across a wide range of threshold probabilities. CONCLUSIONS: The integration of radiomics and deep learning features within a machine learning framework provides a robust, non-invasive tool for predicting pCR to NAC in breast cancer. This multimodal approach outperforms single-modality models and offers potential for clinical translation to personalize treatment strategies and avoid ineffective chemotherapy. Further multi-center validation is warranted to confirm its generalizability.

INTRODUCTION: Contemporary orthodontic treatment planning relies heavily on individual practitioner experience, leading to significant variability in clinical decisions for similar malocclusion presentations and limiting standardized evidence-based care. OBJECTIVES: This research aimed to develop an intelligent treatment recommendation system integrating medical big data analytics with specialized orthodontic knowledge extraction to enhance clinical decision-making accuracy and efficiency. METHODS: The study integrated 1,106 cases from multiple public orthodontic datasets, including ISBI 2015 Grand Challenge, GitHub repositories, PubMed Central case reports, and Kaggle dental imaging competitions. Graph Attention Networks were applied alongside collaborative filtering methods to process these cases and construct orthodontic knowledge graphs that map diagnostic data to treatment outcomes. RESULTS: When tested on extraction decisions, the hybrid system correctly identified treatment needs in 94.2% of cases, while manual evaluation achieved 78.8% accuracy. Processing required only 2.3±0.4 seconds, compared to 35-45 minutes for traditional cephalometric analysis. Different malocclusion categories showed varying results, with Class I cases reaching 96.5% accuracy and Class II Division 2 cases achieving 91.2%. Processing speed improved by 99.8%, sensitivity increased 24.7%, and clinical reliability improved by 28.3% compared to standard diagnostic procedures. CONCLUSION: Big data analytics can enhance orthodontic decision-making while preserving the personalized treatment planning that remains fundamental to achieving optimal treatment outcomes.

Healthcare accessibility challenges disproportionately affect underserved populations, with communication barriers between patients and providers contributing to diagnostic errors and suboptimal outcomes. This study develops and validates a transformer-based lightweight mobile health text analytics system for intelligent symptom monitoring in pervasive healthcare environments. The system employs a DistilBERT-based architecture compressed to 45MB, integrated with medical knowledge graphs incorporating ICD-10 and SNOMED CT standards, and trained on 15,000 medical records from ten hospitals. A three-tier pervasive computing architecture enables cross-platform deployment across iOS, Android, and HarmonyOS, while a four-tier risk stratification framework classifies conditions into self-observation (70%), community consultation (20%), hospital evaluation (8%), and emergency intervention (2%) categories. Privacy preservation utilizes federated learning with differential privacy mechanisms. Clinical effectiveness was evaluated through a randomized controlled trial involving 1,500 participants across diverse demographics. Results demonstrated 86.8% diagnostic concordance versus 70.2% in controls, achieving 93.7% sensitivity and 98.4% specificity for critical symptoms, while reducing emergency department visits by 35.7% and achieving $847 cost savings per patient. Patient experience improvements included 82.7 System Usability Scale scores and 78.4% sustained engagement. This research establishes a paradigm for responsible AI deployment in healthcare that prioritizes clinical effectiveness and social responsibility, contributing to universal health coverage through innovative, accessible, and ethically sound technologies.

The comorbidity rate between Autism Spectrum Disorder (ASD) and depression reaches 50%, yet there is a lack of an integrated theoretical framework to guide digital health interventions. Existing research on sensory processing, emotion regulation, and meaning-making is fragmented, which limits the development of technology-enabled treatment solutions. This study integrates embodied cognition theory, the somatic marker hypothesis, and the principle of decreasing entropy and compensation, proposing a three-layer cognitive structure theory of "sensory awareness - emotional awareness - sense of meaning". This framework reveals that the comorbidity of ASD and depression stems from impairments in the three-layer cognitive pathway. Art therapy operates through dual mechanisms: sensory integration activates bottom-up neuroplasticity, while narrative reconstruction achieves top-down cognitive reappraisal. This theoretical model provides clear guidelines for the development of intelligent assessment tools, adaptive intervention algorithms, and data-driven personalized strategies in mobile health applications.

The rising prevalence of adolescent mental health issues underscores the limitations of traditional counselling services in terms of scalability, timeliness, and accessibility. This paper presents RAG-TPC, a Retrieval-Augmented Generation framework built upon the DeepSeek language model for teenage psychological counselling. The system incorporates intent classification, semantic retrieval, and structured prompt-based generation to produce safe, empathetic, and contextually appropriate responses. We construct a domain- specific dataset spanning general distress, mental illness, and SOS emergencies, and employ LoRA-based fine-tuning to enhance intent recognition. Experimental results show that RAG-TPC consistently outperforms competitive LLMs in both classification and response quality. Evaluations by psychological professionals further validate the system’s practical effectiveness and ethical reliability, highlighting its potential for scalable AI-assisted mental health support.