Frontiers in digital health最新文献

Introduction: Artificial intelligence (AI) and extended reality (XR)-including virtual, augmented, and mixed reality-are increasingly adopted in health-professions education. However, the educational impact of AI, XR, and especially their combined use within integrated AI-XR ecosystems remains incompletely characterized.

Objective: To synthesize empirical evidence on educational outcomes and implementation considerations for AI-, XR-, and combined AI-XR-based interventions in medical and health-professions education.

Methods: Following PRISMA and PICO guidance, we searched three databases (Scopus, PubMed, IEEE Xplore) and screened records using predefined eligibility criteria targeting empirical evaluations in health-professions education. After deduplication (336 records removed) and two-stage screening, 13 studies published between 2019 and 2024 were included. Data were extracted on learner population, clinical domain, AI/XR modality, comparators, outcomes, and implementation factors, and narratively synthesized due to heterogeneity in designs and measures.

Results: The 13 included studies involved undergraduate and postgraduate learners in areas such as procedural training, clinical decision-making, and communication skills. Only a minority explicitly integrated AI with XR within the same intervention; most evaluated AI-based or XR-based approaches in isolation. Across this mixed body of work, studies more often than not reported gains in at least one outcome-knowledge or skills performance, task accuracy, procedural time, or learner engagement-relative to conventional instruction, alongside generally high acceptability. Recurrent constraints included costs, technical reliability, usability, faculty readiness, digital literacy, and data privacy and ethics concerns.

Conclusions: Current evidence on AI, XR, and emerging AI-XR integrations suggests promising but preliminary benefits for learning and performance. The small number of fully integrated AI-XR interventions and the methodological limitations of many primary studies substantially limit the certainty and generalizability of these findings. Future research should use more rigorous and standardized designs, explicitly compare AI-only, XR-only, and AI-XR hybrid approaches, and be coupled with faculty development, robust technical support, and alignment with competency-based assessment.

Background and objective: Structured clinical data is essential for research and informed decision-making, yet medical reports are frequently stored as unstructured free text. This study compared the performance of BERT-based and generative language models in converting unstructured breast imaging reports into structured, tabular data suitable for clinical and research applications.

Methods: A dataset of 286 anonymised breast imaging reports in Spanish was translated into English and used to evaluate five transformer-based models pre-trained in medical data: BlueBERT, BioBERT, BioMedBERT, BioGPT and ClinicalT5. Two natural language processing approaches were explored: classification of 19 categorical variables (e.g. diagnostic technique, report type, family history, BI-RADS category, tumour shape and margin) and extractive question answering of four entities (patient age, patient history, parenchymal distortion or asymmetries, and tumour size). Multiple fine-tuning strategies and input configurations were tested for each model, and performance was evaluated using accuracy and macro F1 scores.

Results: BioGPT demonstrated the best performance in classification tasks, achieving an overall accuracy of $96.10\%$ and a macro F1 score of $90.30\%$ . This was significantly better than BERT-based models ( $p=0.012$ for accuracy and $p=0.017$ for F1), particularly in underrepresented categories such as tumour descriptors. In extractive question answering tasks, BioGPT achieved an average accuracy of $93.24\%$ , which is slightly lower than that of BioMedBERT and ClinicalT5, but not significantly so. Notably, BioGPT could perform classification and extractive question answering simultaneously, which is a capability unavailable in BERT-like models.

Conclusions: Generative models, particularly BioGPT, offer a robust and scalable approach to automating the extraction of structured information from unstructured breast imaging reports. Their superior performance, combined with their ability to handle multiple tasks concurrently, highlights their potential to reduce the manual effort required for clinical data curation and to enable the efficient integration of imaging data into research and clinical workflows.

This systematic literature review investigates the Google Cloud Healthcare API's role in transforming healthcare delivery through advanced analytics, machine learning, and cloud-based solutions. The study examines current features of cloud-based healthcare platforms in managing heterogeneous healthcare data formats, analyzes the effectiveness of cloud solutions in enhancing clinical outcomes, and compares Google Cloud Healthcare API with alternative platforms. The findings reveal that Google Cloud Healthcare API demonstrates notable advantages through its fully managed, serverless architecture, native support for healthcare standards (e.g., FHIR, HL7v2, DICOM), and seamless integration with advanced AI/ML services. Cloud-based predictive analytics platforms have proven effective in reducing hospital readmissions, addressing physician burnout, and enabling scalable telemedicine solutions. However, significant challenges persist including data privacy concerns, regulatory compliance complexities, infrastructure dependencies, and potential vendor lock-in risks. The research demonstrates that healthcare organizations implementing comprehensive cloud-based solutions achieve measurable improvements in patient outcomes, operational efficiency, and care delivery models. While technical challenges around latency in medical imaging and interoperability remain, the evidence strongly supports cloud adoption for healthcare transformation, provided organizations address security, compliance, and implementation challenges through strategic planning and comprehensive change management approaches.

Introduction: Rural regions often face persistent healthcare access challenges due to geographic isolation, aging populations, limited infrastructure, and seasonal fluctuations in demand. These challenges not only impact resident well-being but also hinder tourism development. While research has addressed rural healthcare or development separately, limited attention has been given to the synergies between telehealth, regional revitalization, and tourism. This study investigates how telehealth can act as a catalyst for both rural development and sustainable tourism in remote settings.

Methods: This pilot study introduces a telehealth framework using a portable diagnostic device integrated into a broader smart village strategy. The initiative was led by the Digital Innovation in Public Health Research Lab at the University of West Attica, in partnership with local authorities and private healthcare providers. Key components included: (a) cross-sector collaboration; (b) a custom-built web platform for monitoring effectiveness; (c) training of local personnel to assist with guided remote consultations; (d) use cases such as chronic disease monitoring, acute symptom triage for tourists, and digital nomad services; and (e) policy alignment at national and European levels.

Results: Preliminary qualitative findings suggest improved healthcare accessibility for residents with chronic conditions and enhanced medical support for visitors and digital nomads. The system demonstrated feasibility even in low-connectivity environments and received positive feedback from community stakeholders.

Discussion: This study contributes both theoretically and practically by advancing literature on the intersection of telehealth and rural tourism development. The framework offers a replicable model for other European rural regions seeking to enhance health equity, promote digital inclusion, and attract long-term visitors. Despite challenges-such as digital literacy, infrastructure limitations, and sustainability-the pilot illustrates the strategic potential of telehealth in underserved areas. Future research will focus on longitudinal outcomes and the policy tools needed for broader scalability.

Introduction: Temper tantrums are common in early childhood. Severe emotional outbursts, however, are transdiagnostic, disruptive, and difficult to measure across settings, highlighting the need for better methods to identify and predict these components of emotion dysregulation. To address major methodological gaps, we propose a multimodal approach combining a retrospective electronic health record (EHR) analysis (Study 1) and a pilot wearable feasibility study (Study 2) to explore new ways of predicting and quantifying emotional outbursts in children enrolled in a therapeutic day program (TDP).

Methods: In Study 1, we explored retrospective data collected from the EHR (historical patient data and hourly behavioral observations), trying to understand which variables might predict an outburst. In Study 2, wearable technology was employed to characterize outbursts leveraging free-living data collected during a typical day at a TDP. Moreover, we used these data to assess the future of possible outburst predictions among a clinical sample by analyzing the feasibility of such a technology.

Results: An EHR analysis of 45 patients aged 4-8 years revealed that observed rough behaviors at the beginning of the day were associated with an increased likelihood of subsequent outbursts (p < .001), from 6% for those with zero rough behaviors to 68% for those with two or more such behaviors. Wearable sensor data demonstrated high tolerability (all four children assented each of 3-5 days of participation for 5 h of wear) and minimal data loss (<4%). Case studies of wearable-derived heart rate, heart rate variability, and skin temperature suggested that these factors might serve as promising indicators for detecting distress and outbursts.

Discussion: Our results suggest that behavioral observation has the potential of predicting outbursts, and that wearable sensors are tolerable and feasible for children to wear. Overall, multiple methodologies should be studied concurrently and may be required to predict outbursts in the future.

Background: Telemedicine has emerged as a transformative tool for remote healthcare, taking advantage of information and communication technologies to simplify the access to healthcare by patients. With the publication of Regulation (EU) 2025/327 on the European Health Data Space (EHDS), telemedicine has gained new momentum, particularly in the context of MyHealth@EU.

Objectives: This article explores the framework for cross-border telemedicine under the EHDS, with a focus on real-world applications. It also aims to identify the enablers and barriers in several key domains, including legal, regulatory, organizational, financial, clinical, cultural, and technical aspects. The article also aims to discuss innovations in the field of telemedicine, namely the application of artificial intelligence.

Methods: This review was conducted, focusing on cross-border telemedicine applications and the impact of the EHDS. The review also included studies related to telemedicine implementations in different medical disciplines, presenting the key successes and the challenges associated with these methods.

Results: It is highlighted the initial progress made in cross-border telemedicine, where various approaches have been used, including teleconsultation, tele-expertise exchange, telemonitoring, telepathology, teleradiology, and remote surgery. Despite challenges such as legal uncertainties, financial constraints, and technical barriers, the integration of EHDS, supported by MyHealth@EU, has proven beneficial in building trust in secure, reliable telemedicine applications. The lessons learned and recommendations offer valuable insights for scaling cross-border telemedicine services. With the implementation of the EHDS and the use of MyHealth@EU, services can significantly improve access to healthcare and clinical outcomes by enabling more informed decision-making. As these services continue to evolve, they will contribute to a more integrated, and patient-centered healthcare system.

Introduction: As digital data collection becomes increasingly integrated into the treatment of patients with affective disorders, the use of dashboards to visualize this information for clinicians is gaining importance. However, the question of which parameters should be prioritized for display remains largely unaddressed. This study aims to identify the parameters that physicians working in psychiatric facilities consider most important for inclusion in dashboard infrastructures supporting the inpatient care of patients with affective disorders.

Methods: From July 2024 to August 2024, we conducted a survey among 57 physicians working in psychiatric facilities at German university centers with varying levels of experience. We asked them to rank the relevance of 22 pre-specified key clinical parameters for digital dashboard displays. Additionally, we assessed whether characteristics such as gender, age, years of professional experience, and professional seniority influenced these preferences.

Results: Forty-six physicians (80%) physicians completed the data entry. Across the sample, current suicidality emerged as the most important parameter to clinicians. Other highly ranked parameters included information on previous pharmacological antidepressant treatment attempts and data on the course of disease such as year of onset and the number of episodes. The influence of clinician-related factors on parameter prioritization was limited, supporting the generalizability of the findings.

Discussion: Our findings provide practical guidance for the refinement of digital dashboards tailored to the clinical needs in the treatment of affective disorders. Future research should incorporate the perspectives of the entire multidisciplinary care team and evaluate the feasibility and clinical integration of such dashboards to ensure their broader applicability and effectiveness in routine practice.

Introduction: Survival after out-of-hospital cardiac arrest decreases by 5%-12% for every minute of delay in treatment. Ambulance response times vary widely across Norway, particularly between urban and rural municipalities. Advances in digital health technologies may encourage earlier patient contact with emergency services, potentially mitigating these delays.

Methods: We analyzed official response time data from four Norwegian municipalities representing diverse geographic contexts (Bergen, Tokke, Lurøy, Sørfold). Using a survival decay function (Equation), we simulated changes in survival probability under scenarios where emergency calls were placed 1, 5, or 10 min earlier than observed.

Results: Baseline survival probabilities varied substantially across municipalities, from 47.7% in Bergen (mean response 10.2 min) to 9.3% in Lurøy (32.8 min). Simulated earlier calls produced marked gains: in Bergen, survival increased from 47.7% to 68.6% with a five-minute advance; in Sørfold, from 19.4% to 27.9%; and in Tokke, from 29.9% to 43.1%. Even modest improvements (1-2 min) yielded meaningful survival benefits.

Conclusions: Geographic disparities in emergency response times strongly influence survival after cardiac arrest. Wearables and AI-based monitoring cannot predict cardiac arrest but may promote earlier recognition of abnormal physiological states and timelier emergency calls. If widely adopted, such technologies could provide substantial survival gains, particularly in rural and remote regions.

Objective: To assess whether synthetic-only first-person clinical self-reports generated by a large language model (LLM) can support accurate prediction of standardized mental-health scores, enabling a privacy-preserving path for method development and rapid prototyping when real clinical text is unavailable.

Methods: We prompted an LLM (Gemini 2.5; July 2025 snapshot) to produce English-language first-person narratives that are paired with target scores for three instruments-PHQ-9 (including suicidal ideation), LSAS, and PCL-5. No real patients or clinical notes were used. Narratives and labels were created synthetically and manually screened for coherence and label alignment. Each narrative was embedded using bert-base-uncased (mean-pooled 768-d vectors). We trained linear/regularized linear (Linear, Ridge, Lasso) and ensemble models (Random Forest, Gradient Boosting) for regression, and Logistic Regression/Random Forest for suicidal-ideation classification. Evaluation used 5-fold cross-validation (PHQ-9/SI) and 80/20 held-out splits (LSAS/PCL-5). Metrics: MSE, $R^2$ , MAE; classification metrics are reported for SI.

Results: Within the synthetic distribution, models fit the label-text signal strongly (e.g., PHQ-9 Ridge: MSE $4.41\pm 0.56$ , $R^{2}0.92\pm 0.02$ ; LSAS Gradient Boosting test: MSE $75.00$ , $R^{2}0.95$ ; PCL-5 Ridge test: MSE $35.62$ , $R^{2}0.85$ ).

Conclusions: LLM-generated self-reports encode a score-aligned signal that standard ML models can learn, indicating utility for privacy-preserving, synthetic-only prototyping. This is not a clinical tool: results do not imply generalization to real patient text. We clarify terminology (synthetic text vs. real text) and provide a roadmap for external validation, bias/fidelity assessment, and scope-limited deployment considerations before any clinical use.

Depression in older adults, often underrecognized and frequently conflated with cognitive symptoms, remains a major challenge in settings such as assisted-living communities. However, the need for scalable, speech-based screening tools extends across diverse populations and is not restricted to older adults or residential care. Depression in older adults is both common and frequently underdiagnosed, and while assisted-living environments represent a high-need deployment context, the present model is population-agnostic and can be validated across multiple real-world settings. Depression often co-occurs with mild cognitive impairment, creating a complex and vulnerable clinical landscape. Despite this urgency, scalable, interpretable, and easy-to-administer tools for early screening remain scarce. In this study, we introduce a transparent and lightweight AI-driven screening model that uses only four linguistic features extracted from brief conversational speech to detect depression with high sensitivity. Trained on the DAIC-WOZ dataset and optimized for deployment in resource-constrained settings, our model achieved moderate discriminative performance (AUC = 0.760) with a clinically calibrated sensitivity of 92%. Beyond raw accuracy, the model offers insights into how affective language, syntactic complexity, and latent semantic content relate to psychological states. Notably, one semantic feature derived from transformer embeddings, emb_1, appears to capture deeper emotional or cognitive tension not directly expressed through lexical negativity. Although the dataset does not contain explicit cognitive-status labels, these findings motivate future research to test whether similar semantic patterns may overlap with linguistic indicators of cognitive-affective strain observed in prior work. Our approach outperforms many more complex models in the literature, yet remains simple enough for real-time, on-device use, marking a step forward in making mental health AI both interpretable and clinically actionable. The resulting framework is population-agnostic and can be validated in assisted-living environments as one of several high-need deployment settings.