Frontiers in digital health最新文献

{"title":"Associations of eHealth literacy and dietary diversity with depressive symptoms among patients with chronic diseases in rural Western China.","authors":"Yanan Chen, Ting Zhao, Chunsheng Li, Haodong Meng, Hui Qiao","doi":"10.3389/fdgth.2026.1920279","DOIUrl":"https://doi.org/10.3389/fdgth.2026.1920279","url":null,"abstract":"<p><strong>Objective: </strong>The global prevalence of depressive symptoms has been increasing steadily. Both eHealth literacy and dietary diversity are closely linked to mental health. However, the mechanisms by which these factors jointly influence depressive symptoms remain unclear. This study aimed to investigate whether dietary diversity mediates the association between eHealth literacy and depressive symptoms among patients with chronic diseases in rural western China while also evaluating the interaction and combined effects of these two factors.</p><p><strong>Methods: </strong>This cross-sectional study utilized data from the 2024 Ningxia Rural Residents' Health Status and Health Service Utilization Survey. The final analysis included 2,655 patients aged 18 years and older with chronic diseases. Latent profile analysis (LPA) was employed to categorize eHealth literacy levels. A mediation analysis was performed to investigate the mediating role of dietary diversity in the relationship between eHealth literacy and depressive symptoms. To evaluate the interactive and joint effects of eHealth literacy and dietary diversity on depressive symptoms, multiplicative interaction models, additive interaction models, and multivariable logistic regression models were applied.</p><p><strong>Results: </strong>Using LPA, participants were classified into three groups on the basis of eHealth literacy: high (34.43%), medium (19.43%), and low (46.14%). Low eHealth literacy was linked to increased odds of depressive symptoms (OR = 1.64, 95% CI: 1.49 to 1.82), a relationship that diminished after adjusting for dietary diversity (OR = 1.53, 95% CI: 1.38 to 1.72). Dietary diversity served as a partial mediator in this association, explaining 18.79% of the total effect. A significant additive interaction was identified (RERI = 0.65; 95% CI: 0.12 to 1.19), whereas the results of the joint analysis revealed the lowest odds of depressive symptoms in the high-literacy/high-dietary-diversity group (OR = 0.17; 95% CI: 0.05 to 0.56).</p><p><strong>Conclusions: </strong>Dietary diversity may partially mediate the association between eHealth literacy and depressive symptoms. A synergistic association exists between eHealth literacy and dietary diversity with respect to depressive symptoms. Within various eHealth literacy subgroups, dietary diversity is correlated with reduced odds of depressive symptoms, underscoring its potential significance for promoting mental health.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1920279"},"PeriodicalIF":5.2,"publicationDate":"2026-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13649753/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149170334","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A simulation study of differentially private federated learning with zero-trust policy screening for medical IoT devices.","authors":"Ihssan S Masad, Ali Mohammad Alqudah, Shoroq Qazan, Sultan Shoaib","doi":"10.3389/fdgth.2026.1807342","DOIUrl":"10.3389/fdgth.2026.1807342","url":null,"abstract":"<p><p>Artificial intelligence is increasingly used in connected medical-device systems for physiological-signal classification and continuous monitoring. This paper presents a simulation study of the learning and policy layers of a zero-trust federated-learning framework for Internet of Medical Things (IoMT) devices. The implemented prototype combines federated averaging, Gaussian perturbation of client updates, a device-level trust score, and gradient-magnitude screening of anomalous client updates. This is a simulation study of the federated-learning and zero-trust policy layers; cryptographic, encrypted-transport, and hardware-attestation components are modeled as deployment considerations or overheads and are not implemented as complete physical or protocol-level modules. Using the ECG classification pipeline described in the methodology, we compare centralized training, traditional federated learning, differentially private federated learning, and the proposed zero-trust federated configuration. The proposed ZT-FL configuration is reported at 0.928 accuracy, 0.920 F1-score, and 0.968 AUC. These results support feasibility within the stated simulation scope and do not establish clinical validity, formal end-to-end DP accounting, comprehensive adversarial robustness, regulatory compliance, or post-quantum readiness.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1807342"},"PeriodicalIF":5.2,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13646169/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149139621","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Real-world validation of an AI-based non-mydriatic fundus camera in the diagnosis of diabetic retinopathy in Liberia.","authors":"Vickie Fayiah Sam, Mohammad Rizwan, John Chiemerie Douglas, Amrit Jha, Sirisha Senthil, Padmaja Kumari Rani","doi":"10.3389/fdgth.2026.1938702","DOIUrl":"10.3389/fdgth.2026.1938702","url":null,"abstract":"<p><strong>Introduction: </strong>Diabetic retinopathy (DR) is a leading cause of preventable vision loss, yet access to retinal screening remains limited in low-resource settings such as Liberia. This study evaluated the real-world diagnostic performance of an offline artificial intelligence (AI)-enabled smartphone-based non-mydriatic fundus camera for detecting referable diabetic retinopathy in a routine general outpatient screening setting in Liberia.</p><p><strong>Methods: </strong>This prospective study was conducted at the General Outpatient Department of John F. Kennedy Medical Center, Monrovia, Liberia. Adults aged 18 years and older were recruited through convenience sampling and underwent non-mydriatic fundus imaging using the Remidio Fundus on Phone (FOP-NM-10) integrated with an offline AI system. Fundus images were independently assessed by two trained optometrists using the International Clinical Diabetic Retinopathy (ICDR) classification, with disagreements adjudicated by a senior retinal ophthalmologist. Referable DR was defined as moderate non-proliferative DR, severe non-proliferative DR, or proliferative DR. Diagnostic performance was evaluated at the participant level among participants with at least one gradable eye.</p><p><strong>Results: </strong>A total of 1,612 participants were enrolled, of whom 1,357 were included in the final analytical cohort. Of these, 1,198 (88.3%) had at least one gradable eye and were included in the diagnostic-performance analysis. Eleven participants were classified as having referable DR by the human reference standard. The AI system correctly classified 10 of these 11 participants, yielding a sensitivity of 90.9% (95% CI, 58.7%-100%). Among 1,187 participants without referable DR, 1,183 were correctly classified as negative, resulting in a specificity of 99.7% (95% CI, 99.1%-99.9%). The positive predictive value was 71.4% (95% CI, 41.9%-91.6%), and the negative predictive value was 99.9% (95% CI, 99.5%-99.9%).</p><p><strong>Conclusions: </strong>In this prospective real-world study, the offline AI-based non-mydriatic fundus camera showed high sensitivity and specificity for detecting referable diabetic retinopathy among participants with at least one gradable eye in a general outpatient setting in Liberia. These findings support its feasibility for AI-assisted detection of referable DR in resource-limited settings.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1938702"},"PeriodicalIF":5.2,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13646051/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149139456","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Designing healthcare transformation: a hybrid model between the hospital and home for high-risk pregnant women.","authors":"Nirit Putievsky Pilosof, Rakefet Yoeli-Ullman, Shlomi Toussia-Cohen, Esther Galler, Galia Barkai, Eyal Zimlichman, Abraham Tsur","doi":"10.3389/fdgth.2026.1797655","DOIUrl":"10.3389/fdgth.2026.1797655","url":null,"abstract":"<p><strong>Introduction: </strong>Healthcare organizations are developing hybrid models of care between the hospital and home using remote technologies for monitoring and communication, aiming to transform their healthcare service. While most programs focus on technology integration and adaptation, or on clinical safety and quality, sustainable healthcare transformation requires a system design approach that coordinates change across human, technological, and spatial dimensions. This study investigates a novel hybrid model for women with high-risk pregnancy, offering an alternative to prolonged hospital stays. The model enables home hospitalization through intensive telemonitoring and daily remote clinical encounters, complemented by twice-weekly in-person hospital assessment by the same clinical team. Advanced telemedicine technologies enable women to perform essential maternal and fetal assessments at home under remote guidance, whereas inpatient visits support evaluations that are not yet feasible remotely. The People-Process-Place (3P) framework, which conceptualizes healthcare systems as dynamic configurations of social actors, care processes, and physical and digital places, provides a useful lens for examining the hybrid model as a form of healthcare transformation that simultaneously reshapes human roles, organizational workflows, and care environments.</p><p><strong>Methods: </strong>The qualitative study, based on observations and forty interviews of patients, staff, management, and technology directors, examines the transformation of the hybrid model of care by adopting the People-Process-Place (3P) framework to analyze the multidimensional change.</p><p><strong>Results: </strong>The findings demonstrate transformation across all dimensions of the People-Process-Place framework. The hybrid model reshaped professional practice, enhanced patient autonomy, and increased family involvement (People), thereby expanding relational continuity of care. It reconfigured the patient journey, introduced new patterns of technology use, redefined organizational policies and advanced economic models (Process), and expanded care across hospital and home settings, with physical and digital environments provided in diverse locations (Place). At the intersection of People-Process-Place lay safety, trust, and experience, which affect the system's flexibility and ability to manage challenges and scale.</p><p><strong>Conclusions: </strong>The study presents a design framework for healthcare transformation to examine the intersection of human, technological, and spatial change. The case of the hybrid model for high-risk pregnant women demonstrated the complexity of transformation and the need for a design framework to evaluate and develop change processes.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1797655"},"PeriodicalIF":5.2,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13646079/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149139543","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Current status, challenges and future development of self-service appointment platforms in hospital radiology departments in China: a narrative review.","authors":"Peng-Wei Shu, Ke-Ke Lu, Shi-Hao Xie, Zi-Qiao Lei, Qing Fu","doi":"10.3389/fdgth.2026.1940548","DOIUrl":"10.3389/fdgth.2026.1940548","url":null,"abstract":"<p><p>In recent years, China has accelerated the implementation of precision medicine strategy, integrating accuracy, individualized care and evidence-based decision-making into clinical diagnosis and treatment. Meanwhile, the expansion of smart-hospital and the requirement of Level-5 electronic medical record (EMR) certification have made digital upgrading an indispensable component of radiological services. As a result, self-service appointment is increasingly replacing conventional manual registration and has become an important part of the digital infrastructure of modern radiology departments, with an increasing proportion of administrative pressure, and enhance overall operational performance. Nevertheless, their effectiveness remains restricted by several deficiencies, including limited differentiation among examination workflows, insufficient risk assessment before imaging procedures, inadequate accessibility for older adults, and delayed information exchange between heterogeneous systems; therefore, future improvement should focus on three major directions: (a) AI-supported scheduling tailored to individual patient needs, (b) closed-loop management covering the entire examination process, and (c) coordinated appointment mechanisms involving multiple clinical disciplines, thereby advancing radiology services toward a more refined, intelligent, and patient-centered model.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1940548"},"PeriodicalIF":5.2,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13646530/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149139487","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"GRAPHOS re-examined: graph-theoretic features do not improve dyslexia screening, and a demographic confound in the Dytective dataset.","authors":"Sudha M, Murali S","doi":"10.3389/fdgth.2026.1927725","DOIUrl":"https://doi.org/10.3389/fdgth.2026.1927725","url":null,"abstract":"<p><strong>Introduction: </strong>Most machine learning methods for dyslexia screening treat the performance measurements of cognitive tasks as independent feature vectors. It has been hypothesised that this discards information carried by the interdependence among tasks probing shared cognitive processes. This paper tests that hypothesis and does not find support for it.</p><p><strong>Methods: </strong>We evaluate GRAPHOS, a graph-theoretic framework that builds task correlation networks from gamified cognitive assessment data, on the Dytective dataset of 3,644 participants (392 dyslexic). Model, aggregation strategy and decision threshold are selected exclusively on training and validation partitions, and the test partition is read once.</p><p><strong>Results: </strong>On the threshold-free AUC, GRAPHOS does not outperform a strong single-classifier baseline (0.8881 vs. 0.8883; DeLong p = 0.989), and this null replicates across ten random splits. A model using no graph-derived features is statistically indistinguishable from the full framework. A size-matched permutation null shows that none of nine population-level network differences between the dyslexic and non-dyslexic groups survives correction for the 8.3:1 sampling asymmetry. A subgroup audit shows that dyslexia prevalence in this dataset differs 4.7-fold by native-language status, that four demographic columns alone reach AUC 0.724 against 0.744 for all 192 task metrics, and that including them yields a twelve-fold disparity in false-positive rate between language groups.</p><p><strong>Discussion: </strong>A psychometric analysis indicates why the graph representation gains nothing: the battery exhibits little common variance in this sample (KMO 0.497, Cronbach's α 0.171), and exploratory factor analysis recovers 13% of variance with 28 of 32 tasks loading on no factor. We report these as negative results together with the two controls that produced them, and suggest both as routine: the size-matched null wherever correlation networks are compared between groups of unequal size, and the subgroup audit wherever a screening classifier is trained on a sample whose label prevalence may covary with demographic composition.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1927725"},"PeriodicalIF":5.2,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13656794/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149185930","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"AI is already in the theatre of war just not on the side of the patient: clinical chatbots, telemedicine, and an agenda for \"AI for medical humanity\".","authors":"Andrew R Prigg, Henry Robinson, Munzer Alkhalil, James Gough","doi":"10.3389/fdgth.2026.1955671","DOIUrl":"10.3389/fdgth.2026.1955671","url":null,"abstract":"<p><p>Artificial intelligence (AI) is already embedded in modern warfare, mainly through surveillance, targeting, and military decision-making. However, its potential to support clinicians and patients in conflict settings remains underdeveloped. Frontline medical teams must make time-critical, high-stakes decisions with scarce diagnostics, disrupted referrals, insecure communications, changing injury patterns, and irregular access to specialist expertise. Humanitarian medicine needs a proactive clinician-facing AI agenda that strengthens clinical judgment, not replaces it. We propose a two-stage model: a bounded, protocol-anchored chatbot as the \"first mile\" of point-of-care support, and telemedicine as the \"second mile\" for specialist oversight and accountability. The chatbot would structure assessments, guide protocol-based actions, identify missing information and areas of uncertainty, generate standard handovers, and trigger escalations. Remote specialists would provide context and retain responsibility for clinical decisions. Safe deployment requires transparent protocols, constrained scope, hard refusals, awareness of uncertainty, data minimisation, offline-first functionality, adversarial threat modelling, and safeguards against coercive or dual-use applications. Evaluation should use locally informed simulation, silent deployment, prospective monitoring of deviations and harms, predefined safety thresholds, and independent humanitarian oversight before wider clinical use. AI will increasingly shape conflict medicine, with or without humanitarian engagement. The priority is to ensure design, governance, and evaluation are guided by patient safety, medical neutrality, local involvement, and clinical accountability, placing AI on the side of the patient, not the machinery of war.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1955671"},"PeriodicalIF":5.2,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13642742/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149139243","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Evaluating the accuracy, quality, and reliability of septoplasty-related content on TikTok.","authors":"Raika Bourmand, Aayush Shah, Angel Sheu, Reshma Soundharrajan, Sofia Olsson, Viraj Shah","doi":"10.3389/fdgth.2026.1934459","DOIUrl":"10.3389/fdgth.2026.1934459","url":null,"abstract":"<p><strong>Introduction: </strong>TikTok has become a major source of health information, but concerns remain regarding the quality and reliability of medical content. Septoplasty is one of the most performed otolaryngologic procedures and is frequently discussed on the platform. We seek to evaluate the quality, reliability, and accuracy of septoplasty-related content on TikTok.</p><p><strong>Methods: </strong>A cross-sectional content analysis of publicly available TikTok videos was conducted between April and May 2025. Videos were identified using the search terms \"septoplasty,\" \"deviated septum,\" and \"septal reconstruction.\" After screening, 173 videos were included for analysis using the Global Quality Scale (GQS) and modified DISCERN (mDISCERN); 74 videos contained sufficient health information for evaluation with the Accuracy in Digital-health Instrument (ANDI). Descriptive and comparative analyses assessed content quality, reliability, accuracy, and engagement across creator types.</p><p><strong>Results: </strong>Overall, videos demonstrated moderate accuracy (mean ANDI 2.82), low-to-moderate quality (mean GQS 2.31), and low reliability (mean mDISCERN 0.78). Most videos (67.1%) were created by non-health care professionals. Physician-created videos had significantly higher accuracy, quality, and reliability scores than nonphysician videos (all <i>P</i> ≤ .001). Engagement metrics did not differ significantly by creator type (all <i>P</i> > .30). Verified account status was the factor most consistently associated with higher engagement.</p><p><strong>Conclusion: </strong>Septoplasty-related TikTok content demonstrates moderate accuracy but generally low reliability and quality. Content created by physicians and other healthcare professionals demonstrated higher accuracy, quality, and reliability than content created by non-healthcare professionals. These findings support greater participation by qualified healthcare professionals in disseminating evidence-based information on social media, although the effects of such content on viewer understanding and healthcare decision-making were not assessed.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1934459"},"PeriodicalIF":5.2,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13642820/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149139497","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Comparative evaluation of explainable vision models for bone tumour detection on radiographs.","authors":"Daniel Jones Ortega, Khuhed Memon, Syed Saad Azhar Ali, Yasir Hafeez","doi":"10.3389/fdgth.2026.1926330","DOIUrl":"10.3389/fdgth.2026.1926330","url":null,"abstract":"<p><strong>Introduction: </strong>Accurate detection of bone tumours on radiographs can be challenging because lesion appearance varies and interpretation requires specialist expertise. This study evaluates deep learning (DL) models for bone tumour localisation, segmentation, classification, and explainability using the publicly available Bone Tumour X-Ray Radiograph Dataset (BTXRD), comprising 3,746 musculoskeletal radiographs.</p><p><strong>Methods: </strong>A YOLOv8s-seg model was evaluated alongside MobileNetV3-Large, EfficientNet-B3, and ConvNeXt-Tiny classifiers trained under a common-resolution protocol. Grad-CAM, Grad-CAM++, and Layer-CAM activation maps were compared with expert tumour masks using mean Intersection over Union and a thresholded lesion-overlap measure. For the retrospective agreement analysis, EfficientNet-B3 was paired with YOLOv8s-seg. An exploratory external series of 11 tumour-positive radiographs was also examined.</p><p><strong>Results: </strong>ConvNeXt-Tiny achieved the highest binary classification accuracy (0.879), while YOLOv8s-seg achieved an overall box mAP50 of 0.676 and mask mAP50 of 0.668. The components produced concordant binary tumour-presence predictions for 605 of 750 test radiographs, corresponding to 80.7% coverage (95% confidence interval [CI]: 77.7%-83.3%). Binary accuracy within the agreed subset was 93.6% (566/605; 95% CI: 91.3%-95.2%), while the remaining 145 cases were discordant and received no consensus output.</p><p><strong>Discussion: </strong>The XAI analyses quantify spatial correspondence rather than explanation faithfulness or clinical reliability. Because the exploratory external series contained seven benign and four malignant tumour-positive radiographs and no normal radiographs, it cannot assess specificity or false-positive behaviour and is not presented as representative external validation.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1926330"},"PeriodicalIF":5.2,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13642675/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149129249","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Cross-sectional stratification of fall-related vulnerability in Parkinson's disease through multimodal fusion of speech, gait, facial expression, and upper-limb movement.","authors":"Vasileios Skaramagkas, Iro Boura, Georgios Karamanis, Anastasia Pentari, Emilia Michou, Konstantinos Marias, Dimitrios I Fotiadis, Zinovia Kefalopoulou, Cleanthe Spanaki, Manolis Tsiknakis","doi":"10.3389/fdgth.2026.1937343","DOIUrl":"10.3389/fdgth.2026.1937343","url":null,"abstract":"<p><p>Falls constitute a significant cause of morbidity in Parkinson's disease (PD), but current approaches to cross-sectional assessment of fall-related vulnerability frequently depend on unimodal mobility metrics with limited ability to characterize its multidimensional manifestations. This study presents a multimodal deep learning (DL) architecture that incorporates facial expressions, voice, gait dynamics, and upper-limb movements to classify cross-sectional fall-related vulnerability in PD using current postural-instability status and fall history. Data were gathered from 147 patients, and modality-specific models were systematically pruned to reduce model size, inference latency, and computational requirements while maintaining classification performance, providing an initial assessment of computational efficiency within the experimental environment. Various fusion strategies, including late, intermediate, attention-based, transformer-based, and large-language-model-assisted methods, were assessed using leave-one-subject-out cross-validation. The large-language-model-assisted analysis was exploratory and was based on a single recorded API run per participant-modality configuration. Pruning resulted in considerable efficiency improvements without a notable decrease in accuracy, with gait demonstrating the highest robustness. Fusion performance varied across methods and modality combinations. Attention-based fusion with face and gait achieved an observed macro-f1 of 91.67%. An exploratory large-language-model-assisted face-upper-limb configuration produced an observed macro-f1 of 93.75% in a single recorded run; because repeated-run stability and exact model-version reproducibility were not established, this result should not be interpreted as evidence of reproducible superiority. Transformer fusion produced an observed macro-f1 of 85.02% when all four modalities were included. Significantly, speech and particularly face expressivity provided supplementary signals to conventional gait and upper-limb assessments, underscoring the potential of underexplored axial characteristics as complementary digital markers associated with fall-related vulnerability. These findings support further study of multimodal AI for cross-sectional classification of fall-related vulnerability in PD. Prospective fall prediction, external generalizability, and performance on intended devices remain to be evaluated in independent future studies.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1937343"},"PeriodicalIF":5.2,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13642838/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149139369","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}