IntroductionPrecisely grading renal cell carcinoma (RCC) through histopathology slides is a requisite to predict the cancer prognosis and select treatments, however, there is a considerable variability in the assessment between observers. While some AI systems mainly rely on transformer-based and nuclei-centric models, they are computationally very demanding so their clinical use is thus limited. Hence, there is a need for practical and easy-to-understand solutions that can be integrated into digital pathology workflows.MethodsWe built a hybrid framework that integrates U-Net-based tumor segmentation, convolutional feature extraction, nuclei-aware descriptors, stain normalization, and attention-based multiple instance learning for slide-level RCC grading. The framework was tested on 3 public datasets (TCGA-ccRCC, RCdpia, MMIST-ccRCC) with a cross-dataset validation approach. The metrics used for the performance evaluation were macro-F1 and quadratic weighted kappa (QWK).ResultsThe designed method yielded a macro-F1 of 0.94, QWK of 0.92, and accuracy of 0.95. Extracting tumor patches and performing aggregation based on attention resulted in the best improvements. The method was equally effective when tested with different datasets. Most of the errors made by the model were those within the clinical grading range variability. Average time for inference was around one minute and ten seconds per slide.ConclusionBy fine-tuning the convolutional pipeline, one can obtain a RCC grading capability that can be rivaled by very few, yet the model is efficient and interpretable, hence it will continue to be a strong candidate decision-support tool in digital pathology to be clinically deployed.

ObjectiveTo optimize incision placement for uniportal endoscopic lumbar fusion using computational methods to maximize disc and endplate preparation.MethodsWe analyzed 605 axial lumbar MRI images from 121 patients in an institutional development cohort and 120 MRI images from 40 patients in an independent external validation cohort. A parametric MRI-based anatomical delineation system was developed to define the relevant surgical anatomy, and genetic algorithm optimization was used to determine optimal incision placement. Machine learning was then used to predict optimal incision positions from MRI-derived anatomical features, including disc morphology, Kambin’s triangle width, dural sac width, and posterior longitudinal ligament-to-skin distance.ResultsGenetic algorithm optimization identified level-specific optimal incision distances: 58.93 ± 10.19 mm (L1-2), 63.31 ± 9.84 mm (L2-3), 67.79 ± 12.82 mm (L3-4), 83.89 ± 13.69 mm (L4-5), and 95.28 ± 19.38 mm (L5-S1). Under a patient-level framework, the artificial neural network (ANN) showed the best performance in predicting the GA-derived optimal incision position from MRI-derived anatomical features (internal validation R2 = 0.935; external validation R2 = 0.918 [95% CI, 0.869–0.947] and MAE = 6.08 mm [95% CI, 5.17–7.10]). The model was deployed online for real-time patient-specific planning.ConclusionA web-based computational planning tool was developed to optimize incision placement and maximize disc and endplate preparation in uniportal endoscopic lumbar fusion. This approach may support patient-specific preoperative planning.

BackgroundConsidering rising dementia rates, voice assistants are being investigated to support people with dementia (PWD) and reduce the burden on caregivers (CGs). GRACE, an intervention-delivering voice assistant, provides cognitive support and assistance with daily living for people with early-stage dementia. Initial pilot studies showed promising user experience among healthy adults but required further evaluation by care professionals working with PWD as a next reiteration process.ObjectiveThis study investigated three main research questions: (1) Which intervention components of GRACE do healthcare professionals (HCPs) and CGs find most valuable for supporting the daily lives of PWD? (2) What key constraints do HCPs and CGs expect when implementing GRACE in dementia care settings? and (3) What changes or additions do HCPs and CGs recommend for optimizing the functionality of GRACE?MethodsThirteen semi-structured interviews were conducted (CG: n = 4, HCP: n = 9). Anonymized transcripts were coded and analyzed thematically to directly answer the research questions.ResultsRespondents found the onboarding activity and memory exercise to be most valuable, due to its relevance in everyday life. Identified barriers for implementation included practical, technical, and ethical challenges. Recommendations for successful implementation and further development of GRACE included improving PWD acceptance and greater artificial intelligence-supported personalized content. Finally, a list of future exercises for incorporation and other use cases of GRACE beyond dementia care were also identified.ConclusionsGRACE has great potential as a complementary tool in dementia care. Ensuring personalization, ease of use, and alignment with care processes will be critical for adoption and long-term success.

ObjectiveOur aim was to explore the impact of wearable devices on rehabilitation of stroke patients.MethodsWe searched the CINAHL, Cochrane, Embase, PsycINFO, PubMed, Web of Science, China National Knowledge Internet, Wanfang, SineMed, and VIP databases to identify articles from inception through Janu 15th, 2026. The primary endpoint was limbs function, comprising upper limb function, lower limb function, and overall motor function, while secondary endpoints included quality of life and activities of daily living. The standard mean difference was taken as the effect value and the random effects method was adopted when I2 was greater than 50%. The risk bias assessment employed the Cochrane Collaboration’s tool, while the sensitivity analysis and publication bias were respectively evaluated through the sequential exclusion method and the Egger’s method.ResultsThis systematic review included 20 studies and a total sample of 1007 participants. Significant statistical differences were observed in upper limb function (SMD 0.27, 95%CI 0.09 to 0.45, P=0.004), lower limb function (SMD 1.58, 95%CI 0.31 to 2.85, P=0.01), overall motor function (SMD 4.59 95% CI, 0.80 to 8.38, P=0.02) and activities of daily living (SMD 2.61, 95%CI 1.18 to 4.04, P<0.01) among stroke patients.ConclusionWearable devices are found to have positive impact on the limb function and activity of daily living of stroke patients, but have no significant intervention effect on quality of life.

BackgroundPrimary gastric diffuse large B-cell lymphoma (PG-DLBCL) exhibits heterogeneous outcomes, and conventional prognostic systems often fail to capture its unique clinicopathological features. We aimed to develop a robust, interpretable prognostic model to improve risk stratification and guide individualized management.MethodsData from 3773 PG-DLBCL patients (2000–2021) were extracted from the SEER database. Four complementary feature selection strategies—LASSO, Boruta, backward stepwise elimination, and best subset regression (BSR)—were employed to identify stable prognostic variables. Four machine learning (ML) algorithms (logistic regression, support vector machine, k-nearest neighbor, and XGBoost) were trained using these selected variables. Model performance was evaluated through discrimination (AUC), calibration, decision curve analysis, and internal validation. Shapley Additive Explanations (SHAP) were applied for interpretability, and patients were stratified into high- and low-risk groups.ResultsAge, stage, chemotherapy, marital status, and income emerged as key prognostic determinants. The XGBoost model based on BSR-selected predictors achieved the highest performance with good calibration and net clinical benefit. SHAP analysis demonstrated that older age, advanced stage and absence of chemotherapy increased predicted risk, whereas marital status and higher income were protective. Risk stratification effectively distinguished survival outcomes in training and testing cohorts (p < 0.001). A web-based tool was developed for individualized risk assessment.ConclusionsWe established an interpretable, high-performing ML-based prognostic model for PG-DLBCL that integrates clinical, treatment, and socioeconomic factors. This tool enables precise risk stratification, supports individualized therapeutic decisions, and provides a methodological framework for prognostic modeling in this rare extranodal lymphomas.

BackgroundConversational agents (chatbots) are increasingly used as digital health interventions to support chronic disease self-management. Advances in natural language processing (NLP) have improved their capacity for interactive dialogue and personalization, yet evidence regarding their implementation and clinical impact remains limited.ObjectivesThis systematic review identifies and synthesizes studies implementing NLP-based chatbots for chronic disease self-management.MethodsWe searched seven electronic databases (PubMed, Embase, CINAHL, Web of Science, Scopus, Cochrane Library, and IEEE Xplore) and Google Scholar for studies published between January 2010 and November 2025. Studies evaluating NLP-based chatbots designed to support chronic disease self-management were deemed eligible. Study quality and risk of bias were assessed using the Mixed Methods Appraisal Tool and the Quality Assessment with Diverse Studies instrument.ResultsSix studies met the inclusion criteria; most were published in 2023 and targeted conditions such as cancer, diabetes, and hypertension. Chatbot functions primarily focused on symptom monitoring and disease-related education. Reported outcomes included improvements in disease-related knowledge, symptom burden, mental well-being, and self-care adherence. Usability and acceptability were generally favorable, with high satisfaction, perceived usefulness, and engagement. However, evidence of objective clinical benefits, including laboratory outcomes, was limited. Technical architectures varied widely, and advanced NLP capabilities—such as free-text natural language understanding—were rarely implemented.ConclusionsNLP-based chatbots show promise for supporting chronic disease self-management, particularly for psychosocial and behavioral outcomes. However, evidence of clinical efficacy remains limited. Future research should prioritize adaptive, context-aware designs and standardized outcome frameworks aligned with real-world self-management needs.

ObjectiveThe integration of Artificial Intelligence (AI) in healthcare offers opportunities to transform patient care, improve efficiency, and support clinical decision-making. Yet, its implementation is hindered by technical, organizational, and collaborative challenges. This study explores the key success factors for sustainable AI adoption from the perspectives of healthcare professionals (HCPs) and AI experts, with the aim of identifying alignments and gaps that influence collaboration and sustainability.MethodsGuided by the NASSS framework, semi-structured interviews were conducted with four HCPs and five AI experts with experience related to AI technologies in the healthcare sector. Thematic analysis was applied to examine stakeholder perspectives, focusing on gaps and alignments between stakeholders and identifying common patterns.ResultsTransparency leading to trust was the most emphasized factor by both AI experts and HCPs. Alignments were also found in recognizing the importance of interorganizational cooperation, demand-side value, scope of use, usability, responsibility and redefining the roles. Gaps included challenges in cooperation with HCPs, misunderstandings between stakeholders and the need for interdisciplinary experts, insufficient training, concerns over data quality and privacy, and limited attention to usability. Based on the found patterns, a framework is proposed to strengthen collaboration between AI experts and HCPs, enabling effective and sustainable AI implementation in healthcare.ConclusionThis study extends the NASSS framework to a dual-stakeholder context, offering insights into the alignments and gaps between HCPs and AI experts. The proposed framework supports improved collaboration, guiding healthcare professionals, AI developers, and managers toward effective, sustainable AI implementation and fostering mutual understanding for successful adoption.

To conduct a comprehensive health systems policy analysis of the Sindh Telemedicine and Telehealth Act 2021 by examining its contextual drivers, governance and stakeholder dynamics, implementation barriers and facilitators, and policy content, and by comparing it with telemedicine policies in neighboring low- and middle-income countries (LMICs). Our study applied Walt and Gilson's policy triangle to examine context, content, process, and actors. A desk review of relevant policy documents from Sindh and other LMICs (India, Bangladesh, and Nepal) was conducted, followed by in-depth interviews with 28 stakeholders (July-October 2023) until data saturation was reached. Participants were recruited using purposive and snowball sampling strategies. Thematic analysis was done using Qualitative Data Analysis Software (QSR NVivo v12) to identify barriers and facilitators to policy implementation. The contextual drivers behind the formulation of the Act included the coronavirus disease 2019 pandemic, limited healthcare access, and the growth of telehealth providers in private sector. The Act provides a regulatory framework for healthcare providers registration and training requirements, guidelines for patient consent, emergency care, and data privacy. However, it lacks clarity on the cost of telemedicine services and measures for cross-border telemedicine. Stakeholder analysis showed government bodies hold the highest power in telemedicine policymaking. Telemedicine service providers and healthcare providers expressed support for the Act but were excluded from policymaking process. Service users also had minimal involvement in the policymaking process. Policy implementation varies, facing financial and technological barriers, while strategic financial planning and public-private partnerships, identified as facilitators in overcoming these challenges. The study may be subject to limitations, including potential selection bias while identifying stakeholders and the constraints of desk review in capturing details. The Act is an important step towards addressing healthcare disparities, with its success contingent on overcoming implementation barriers through inclusive and strategic policymaking and investment in digital infrastructure.

Background Fatigue, impaired sleep quality and daytime sleepiness, common in neurodegenerative and immune-mediated diseases, are debilitating and have serious societal and economic implications. Currently, measurement of these symptoms largely relies on self-reported questionnaires, which are burdensome for patients and lack sensitivity, granularity and reliability. Methods Building on a preceding feasibility study and qualification advice of the European Medicines Agency, the Clinical Observational Study of the European project Identifying Digital Endpoints to Assess FAtigue, Sleep and acTivities of daily living in Neurodegenerative disorders and Immune-mediated inflammatory diseases (IDEA-FAST) investigates the relationship between digital and clinical parameters of the target concepts of fatigue, reduced sleep quality and daytime sleepiness. Results Between 2022 and 2025, 2000 people are being recruited at 24 European sites – 500 with Parkinson's disease, 500 with inflammatory bowel disease, 200 with each of the following diseases: Huntington's disease, rheumatoid arthritis, systemic lupus erythematosus, primary Sjögren's syndrome and 200 healthy volunteers. Participants are followed over a 24-week period with four visits, each including a 1-week assessment phase at home using CE-certified digital health (including active and passive) technologies. The latter collect information on physical activity, physiology, cognition as well as social interaction and behaviour as core dimensions of the target concepts. Conclusion This study will help to develop reliable, valid and efficient digital endpoints of fatigue, impaired sleep quality and daytime sleepiness for use in future clinical studies and trials.

To examine the cross-level sociotechnical linkages between societal risk perception of medical artificial intelligence, institutional adoption patterns, and clinical safety outcomes. Specifically, this study aims to explore how social pressure shapes hospital technology strategies and to rigorously assess the association between AI usage intensity and diagnostic errors using an acute imaging-dependent condition as a specific tracer. A cross-level analytical framework was constructed based on the Technology Acceptance Model and Institutional Theory. We integrated three heterogeneous data streams from the Federal District of Brazil: a stratified probability survey of residents (N = 4764), longitudinal hospital operational panels (1728 hospital-month observations), and a validating index of social media sentiment. A "Catchment Area Ecological Linkage" protocol was employed to merge micro-level psychometric data with meso-level organizational metrics. Structural Equation Modeling was employed to test the direct, mediating, and moderating effects among variables, with robustness and endogeneity checks conducted via time-lag analysis and double-validation. Moderators included public trust and hospital geographical remoteness. Structural equation modeling revealed a significant negative association between aggregated public risk perception and hospital AI application frequency (β = -0.34, p < 0.001), consistent with the theory of "algorithmic aversion" at the institutional level. Within the specific context of the tracer condition, higher AI usage intensity was positively associated with misdiagnosis rates (β = 0.28, p < 0.001), suggesting a pattern of "automation bias" in time-sensitive acute triage. These inhibitory effects are attenuated by high public trust and geographical remoteness. Public risk perception functions as an institutional constraint that throttles technology deployment. While social pressure limits adoption, the uncritical reliance on AI in high-stakes acute settings may compromise diagnostic vigilance. This study highlights the necessity of using precise tracer conditions to evaluate digital health safety and suggests that governance must balance social legitimacy with rigorous clinical oversight.