bims-aukdir 2026-02-01 papers

bims-aukdir

Biomed News

on Automated knowledge discovery in diabetes research

Issue of 2026–02–01
nineteen papers selected by
Mott Given

Construction of an interpretable machine learning model for predicting gestational diabetes mellitus based on 45 dietary nutrients.
Early prediction of diabetic retinopathy using a multimodal deep learning framework integrating fundus and OCT imaging.
Development of a machine learning-based model for prediction of diabetes risk in patients with metabolic dysfunction-associated steatotic liver disease (MASLD).
Development and validation of a two-stage machine learning model for personalised type 2 diabetes screening in the All of Us Research Program and UK Biobank.
Evaluating Machine Learning Models for Classifying Diabetes Using Demographic, Clinical, Lifestyle, Anthropometric, and Environmental Exposure Factors.
Enhancing survival prediction for COVID-19 in diabetic patients in Mexico: integrating RMST, propensity score matching, and ensemble machine learning.
UHPLC-Q-Orbitrap HRMS-Based Machine Learning Constructs the Integrated Biomarker Profiling of Type 2 Diabetes and Diabetic Heart Disease.
Artificial Intelligence-Driven Transformation of Pediatric Diabetes Care: A Systematic Review and Epistemic Meta-Analysis of Diagnostic, Therapeutic, and Self-Management Applications.
Real-world performance of the AI diagnostic system IDx-DR in the diagnosis of diabetic retinopathy and its main confounders.
A Hybrid Closed-Loop Blood Glucose Control Algorithm with a Safety Limiter Based on Deep Reinforcement Learning and Model Predictive Control.
Efficient feature selection with attention based deep cat convolutional stacked sparse autoencoder for diabetes prediction.
DME-RWKV: An Interpretable Multimodal Deep Learning Framework for Predicting Anti-VEGF Response in Diabetic Macular Edema.
Integrating Ophthalmology, Endocrinology, and Digital Health: A Bibliometric Analysis of Telemedicine for Diabetic Retinopathy.
Intelligent technology-driven diabetes prevention and control: From informatization management to artificial intelligence.
Privacy-Preserving Collaborative Diabetes Prediction in Heterogeneous Health Care Systems: Algorithm Development and Validation of a Secure Federated Ensemble Framework.
Identification of Key Genes in Diabetic Retinopathy Using Bioinformatics and Machine Learning Approaches.
Diagnostic accuracy of artificial intelligence to identify diabetic foot ulcers: A systematic review and meta-analysis.
Few-Shot Transfer Learning for Diabetes Risk Prediction Across Global Populations.
Comprehensive blood glucose level prediction from HbA1c levels using machine learning models across the biological range.

BioData Min. 2026 Jan 24.

Construction of an interpretable machine learning model for predicting gestational diabetes mellitus based on 45 dietary nutrients.

Zhihui Xiong, Yi Yuan, Zhouhui Yun, Lijie Li, Yunmeng Chen.



Keywords:  Dietary nutrients; Gestational diabetes mellitus; Machine learning; Predictive model; SHAP analysis; XGBoost

DOI:  https://doi.org/10.1186/s13040-025-00515-z
Front Med (Lausanne). 2025 ;12 1741146

Early prediction of diabetic retinopathy using a multimodal deep learning framework integrating fundus and OCT imaging.

Abdel-Hamid M Emara, Jawad Hasan Alkhateeb, Ghada Atteia, Aiman Turani, Jamal Zraqou, Zeinab Elsawaf, Abid Jameel.

  Diabetic Retinopathy (DR) remains a leading cause of preventable vision impairment among individuals with diabetes, particularly when not identified in its early stages. Conventional diagnostic techniques typically employ either fundus photography or Optical Coherence Tomography (OCT), with each modality offering distinct yet partial insights into retinal abnormalities. This study proposes a multimodal diagnostic framework that fuses both structural and spatial retinal characteristics through the integration of fundus and OCT imagery. We utilize a curated subset of 222 high- quality, modality- paired images (111 fundus + 111 OCT), selected from a larger publicly available dataset based on strict inclusion criteria including image clarity, diagnostic labeling, and modality alignment. Feature extraction pipelines are optimized for each modality to capture relevant pathological markers, and the extracted features are fused using an attention- based weighting mechanism that emphasizes diagnostically salient regions across modalities. The proposed approach achieves an accuracy of 90.5% and an AUC- ROC of 0.970 on this curated subset, indicating promising feasibility of multimodal fusion for early- stage DR assessment. Given the limited dataset size, these results should be interpreted as preliminary, demonstrating methodological potential rather than large- scale robustness. The study highlights the clinical value of hybrid imaging frameworks and AI- assisted screening tools, while emphasizing the need for future validation on larger and more diverse datasets.

Keywords:  EyePACS dataset; artificial intelligence in ophthalmology; attention-based fusion; deep learning; diabetic retinopathy; early diagnosis; fundus photography; medical image analysis

DOI:  https://doi.org/10.3389/fmed.2025.1741146
BMJ Open. 2026 Jan 28. 16(1): e107239

Development of a machine learning-based model for prediction of diabetes risk in patients with metabolic dysfunction-associated steatotic liver disease (MASLD).

Sijie Yang, Ruiqing Mo, Wenqiang Wang, Puxiang Zhen, Wentian Han.

   BACKGROUND: Metabolic dysfunction-associated steatotic liver disease (MASLD) is a leading chronic liver disorder closely linked to diabetes mellitus (DM) and its cardiovascular and renal complications. Early identification of diabetes risk in this population is essential for timely intervention.
OBJECTIVE: To develop machine learning (ML) models to predict diabetes risk in individuals with MASLD and to identify key predictive factors using a nationally representative dataset.
METHODS: Data from 6310 MASLD participants (2007-2018) were analysed and classified into DM and non-DM groups. Feature selection was performed using Random Forest, Least Absolute Shrinkage and Selection Operator and Support Vector Machine Recursive Feature Elimination. Based on selected features, nine ML models were developed. Model performance was evaluated using accuracy, sensitivity, area under the curve, F1 score, Rank Score and Brier Score. SHapley Additive exPlanations (SHAP) were used for interpretability.
RESULTS: Eight key variables (age, urinary albumin (Ualb), total cholesterol (TC), lipid accumulation product (LAP), urinary creatinine, white blood cell count, uric acid and Visceral Adiposity Index) were identified and used for model construction. Among nine algorithms, the Light Gradient Boosting Machine (LightGBM) model showed superior predictive performance. SHAP analysis revealed that Ualb, age, TC and LAP were the most influential predictors.
CONCLUSION: Our ML-based model effectively identifies individuals with MASLD at high risk for developing DM. The LightGBM algorithm outperformed other models in both accuracy and interpretability. Key predictors such as Ualb and LAP highlight the importance of renal and metabolic markers in early diabetes risk prediction, offering a new approach for individualised intervention and clinical decision-making.

Keywords:  Diabetes & endocrinology; Machine Learning; Risk Factors

DOI:  https://doi.org/10.1136/bmjopen-2025-107239
BMJ Open. 2026 Jan 27. 16(1): e103171

Development and validation of a two-stage machine learning model for personalised type 2 diabetes screening in the All of Us Research Program and UK Biobank.

Ahmed Khattab, Shang-Fu Chen, Hossein Javedani Sadaei, Nathan E Wineinger, Ali Torkamani.

   OBJECTIVE: To develop and externally validate a two-stage machine learning framework that integrates polygenic risk and clinical variables for early identification of individuals at risk of developing type 2 diabetes.
METHODS: We conducted a prospective prediction study using data from the All of Us Research Program for model development and the UK Biobank for external validation. Two models were constructed. Stage 1 used gradient boosted decision trees (XGBoost) with cross validation, automated hyperparameter optimisation and class weighting to predict 5-year incident type 2 diabetes using demographic, clinical and polygenic predictors. Stage 2 incorporated glycated haemoglobin or fasting glucose measurements to refine risk estimates. Model interpretation used SHapley Additive exPlanations values and permutation importance, and logistic regression and random forest models served as comparators. Discrimination of all models was compared using the DeLong test.
RESULTS: The Stage 1 model achieved an area under the receiver operating characteristic curve (AUROC) of 0.81 in All of Us and 0.82 in UK Biobank, performing significantly better than the phenotype-only model in UK Biobank (DeLong p=1.05×10⁻⁷⁶). Higher polygenic risk quartiles were associated with increased incidence of type 2 diabetes in both cohorts (global χ2 p<0.001). The Stage 2 model achieved AUROC values of 0.78 in All of Us and 0.77 in UK Biobank. Subgroup performance was consistent across sex and ancestry groups, with CIs reported. Cost analysis suggested potential net savings compared with the American Diabetes Association test.
CONCLUSION: A two-stage machine learning framework that integrates genetic and clinical information can support personalised screening for type 2 diabetes across diverse populations. The approach demonstrated robust performance across cohorts and offers a practical structure for early risk identification.

Keywords:  Diabetes Mellitus, Type 2; Primary Prevention; Risk Assessment

DOI:  https://doi.org/10.1136/bmjopen-2025-103171
Toxics. 2026 Jan 14. pii: 76. [Epub ahead of print]14(1):

Evaluating Machine Learning Models for Classifying Diabetes Using Demographic, Clinical, Lifestyle, Anthropometric, and Environmental Exposure Factors.

Rifa Tasnia, Emmanuel Obeng-Gyasi.

  Diabetes develops through a mix of clinical, metabolic, lifestyle, demographic, and environmental factors. Most current classification models focus on traditional biomedical indicators and do not include environmental exposure biomarkers. In this study, we develop and evaluate a supervised machine learning classification framework that integrates heterogeneous demographic, anthropometric, clinical, behavioral, and environmental exposure features to classify physician-diagnosed diabetes using data from the National Health and Nutrition Examination Survey (NHANES). We analyzed NHANES 2017-2018 data for adults aged ≥18 years, addressed missingness using Multiple Imputation by Chained Equations, and corrected class imbalance via the Synthetic Minority Oversampling Technique. Model performance was evaluated using stratified ten-fold cross-validation across eight supervised classifiers: logistic regression, random forest, XGBoost, support vector machine, multilayer perceptron neural network (artificial neural network), k-nearest neighbors, naïve Bayes, and classification tree. Random Forest and XGBoost performed best on the balanced dataset, with ROC AUC values of 0.891 and 0.885, respectively, after imputation and oversampling. Feature importance analysis indicated that age, household income, and waist circumference contributed most strongly to diabetes classification. To assess out-of-sample generalization, we conducted an independent 80/20 hold-out evaluation. XGBoost achieved the highest overall accuracy and F1-score, whereas random forest attained the greatest sensitivity, demonstrating stable performance beyond cross-validation. These results indicate that incorporating environmental exposure biomarkers alongside clinical and metabolic features yields improved classification performance for physician-diagnosed diabetes. The findings support the inclusion of chemical exposure variables in population-level diabetes classification and underscore the value of integrating heterogeneous feature sets in machine learning-based risk stratification.

Keywords:  ROC–AUC; SMOTE; environmental exposure; exposome; machine learning; multiple imputation (MICE); predictive modeling; random forest

DOI:  https://doi.org/10.3390/toxics14010076
Front Endocrinol (Lausanne). 2025 ;16 1725251

Enhancing survival prediction for COVID-19 in diabetic patients in Mexico: integrating RMST, propensity score matching, and ensemble machine learning.

Mariano Vargas-Santiago, Diana A León-Velasco, Raúl Monroy, Sergio Quezada-García.

   Background: This study evaluates the survival impact of diabetes on hospitalized COVID-19 patients in Mexico by combining traditional survival methods (Restricted Mean Survival Time, RMST) with machine learning (ML) prediction. The goal is to understand how diabetes and associated comorbidities affect short-term survival and to develop accurate, interpretable models that support data-driven decision-making.
Methods: A national dataset of over one million COVID-19 cases was analyzed. Diabetic and non-diabetic cohorts were matched using propensity scores based on key covariates (e.g., age, gender, and comorbidities). RMST differences were estimated using survival curves and statistical testing. Separately, machine learning models (Random Forest (RF) and Variational Deep Neural Network (VDNN)) were trained to predict individual RMST values, and SHapley Additive exPlanations (SHAP) were used for model interpretability.
Results: The RMST for diabetic patients was lower than that for non-diabetic patients, with a difference of 2.32 days (p = 0.0583) after matching. Predictive models achieved strong internal validity (R 2 > 0.60). SHAP analysis revealed obesity, smoking, and hypertension as the top predictors and suggested that temporal variables and comorbidities played a central role in short-term survival.
Conclusion: Combining survival analysis with machine learning provides both inferential and predictive insights into the mortality risk of diabetic COVID-19 patients. More importantly, results show that traditional survival analyzes with modern machine learning yields accurate and interpretable predictions that can support personalized interventions tailored to patients with COVID-19 and comorbid diabetes: such as prioritizing early clinical monitoring, individualized treatment plans, or risk-informed hospital admission decisions, and guide a more efficient allocation of healthcare resources.

Keywords:  COVID-19; RMST; diabetes; ensemble models; machine learning; propensity score matching; survival prediction; viral infections

DOI:  https://doi.org/10.3389/fendo.2025.1725251
Rapid Commun Mass Spectrom. 2026 ;40(9): e70044

UHPLC-Q-Orbitrap HRMS-Based Machine Learning Constructs the Integrated Biomarker Profiling of Type 2 Diabetes and Diabetic Heart Disease.

Jianglan Long, Yueyue Wang, Qiushi Fang, Zhe Shi, Zhenzhen Wang, Dan Yan.

   RATIONALE: Over 70% of diabetic patients die from cardiovascular disease, in which diabetic heart disease (DHD) is an important cause of death in individuals with type 2 diabetes (T2D). It is hence imperative to explore the simple, rapid, and economical method for diagnosing DHD from T2D.
METHODS: T2D and DHD patients were recruited, and their serum samples were used for metabolomic analysis to identify differential metabolites. Logistic regression analysis and receiver operating characteristic curve analysis were performed to identify candidate biomarkers. Moreover, four machine learning methods were used to construct the integrated biomarker profiling (IBP) models with the candidate biomarkers. Gini impurity was employed to select characteristic candidate biomarkers.
RESULTS: Eighty-four differential metabolites were identified in the serum of 58 T2D and 62 DHD patients. Logistic regression analysis indicated that 17 differential metabolites were protective factors, whereas 39 were risk factors for DHD. Further, 29 differential metabolites were identified as the candidate biomarkers of DHD after receiver operating characteristic curve analysis. After comparing the predictive performance of the four machine learning models, the IBP was constructed based on the eXtreme Gradient Boosting (XGBoost) with six candidate biomarkers, which were sphingomyelin (d18:0/16:1), deoxycholic acid, hexadecanedioic acid, phosphatidylcholine (20:5/18:3), L-tryptophan, and N-undecanoylglycine from the ranked results of Gini impurity. The accuracy of the IBP for distinguishing T2D and DHD reached 88.89%, with a 100% accuracy in predicting DHD from T2D patients.
CONCLUSIONS: The IBP, composed of six metabolites, can effectively predict DHD from T2D, and it is expected to become a screening indicator for early-stage DHD.

Keywords:  XGBoost; diabetic heart disease; integrated biomarker profiling; machine learning; type 2 diabetes

DOI:  https://doi.org/10.1002/rcm.70044
Int J Mol Sci. 2026 Jan 13. pii: 802. [Epub ahead of print]27(2):

Artificial Intelligence-Driven Transformation of Pediatric Diabetes Care: A Systematic Review and Epistemic Meta-Analysis of Diagnostic, Therapeutic, and Self-Management Applications.

Estefania Valdespino-Saldaña, Nelly F Altamirano-Bustamante, Raúl Calzada-León, Cristina Revilla-Monsalve, Myriam M Altamirano-Bustamante.

  The limitations of conventional diabetes management are increasingly evident. As a result, both type 1 and 2 diabetes in pediatric populations have become major global health concerns. As new technologies emerge, particularly artificial intelligence (AI), they offer new opportunities to improve diagnostic accuracy, treatment outcomes, and patient self-management. A PRISMA-based systematic review was conducted using PubMed, Web of Science, and BIREME. The research covered studies published up to February 2025, where twenty-two studies met the inclusion criteria. These studies examined machine learning algorithms, continuous glucose monitoring (CGM), closed-loop insulin delivery systems, telemedicine platforms, and digital educational interventions. AI-driven interventions were consistently associated with reductions in HbA1c and extended time in range. Furthermore, they reported earlier detection of complications, personalized insulin dosing, and greater patient autonomy. Predictive models, including digital twins and self-learning neural networks, significantly improved diagnostic accuracy and early risk stratification. Digital health platforms enhanced treatment adherence. Nonetheless, the barriers included unequal access to technology and limited long-term clinical validation. Artificial intelligence is progressively reshaping pediatric diabetes care toward a predictive, preventive, personalized, and participatory paradigm. Broader implementation will require rigorous multiethnic validation and robust ethical frameworks to ensure equitable deployment.

Keywords:  artificial intelligence; closed-loop insulin delivery; continuous glucose monitoring; digital health; machine learning; pediatric diabetes

DOI:  https://doi.org/10.3390/ijms27020802
Sci Rep. 2026 Jan 29.

Real-world performance of the AI diagnostic system IDx-DR in the diagnosis of diabetic retinopathy and its main confounders.

Elisabeth Hunfeld, Allam Tayar, Sebastian Paul, Broder Poschkamp, Rico Großjohann, Eva Morawiec-Kisiel, Beathe Bohl, Johanna M Pfeil, Martin Busch, Merlin Dähmcke, Tara Brauckmann, Sonja Eilts, Marie-Christine Bründer, Milena Grundel, Bastian Grundel, Frank Tost, Jana Kuhn, Jörg Reindel, Petra Augstein, Wolfgang Kerner, Andreas Stahl.

  The escalating prevalence of diabetes mellitus (DM) emphasizes the critical need for early detection of diabetic retinopathy (DR). This study assesses the performance of the autonomous AI-based diagnostic system IDx-DR in detecting DR and its associated confounders in a real-world clinical setting. This prospective cross-sectional study involved 875 diabetic patients with a mean age of 52 years (range: 8-92). Retinal images were captured by trained assistants. IDx-DR results were compared with mydriatic fundus examination (gold standard) and Ophthalmologists' image analysis. Factors impacting image acquisition or analyzability were examined. Among all patients, 10.5% yielded no image in miosis, and 26.1% were unanalyzable by IDx-DR. Confounders affecting image acquisition were examiner, pupil size, patient age and patients' visual acuity. When good quality images were achieved, IDx-DR performed well, particularly in detection of severe DR (sensitivity 94.4%; specificity 90.5%). IDx-DR results exactly matched Ophthalmologists' mydriatic fundoscopy gradings in 54.2% if images of sufficient quality were obtainable. Undergrading of DR severity by IDx-DR was rare (4.8%). IDx-DR shows promise in detecting DR, especially in resource-limited settings and in detecting severe DR. One remaining challenge is good image acquisition in miotic patients.

Keywords:  AI-based diagnostics; Diabetic retinopathy; IDx-DR; Retinal Imaging

DOI:  https://doi.org/10.1038/s41598-026-36970-9
Biosensors (Basel). 2026 Jan 06. pii: 47. [Epub ahead of print]16(1):

A Hybrid Closed-Loop Blood Glucose Control Algorithm with a Safety Limiter Based on Deep Reinforcement Learning and Model Predictive Control.

Shanyong Huang, Yusheng Fu, Shaowei Kong, Yuyang Liu, Jian Yan.

  Due to the complexity of blood glucose dynamics and the high variability of the physiological structure of diabetic patients, implementing a safe and effective insulin dosage control algorithm to keep the blood glucose of diabetic patients within the normal range (70-180 mg/dL) is currently a challenging task in the field of diabetes treatment. Deep reinforcement learning (DRL) has proven its potential in diabetes treatment in previous work, thanks to its strong advantages in solving complex dynamic and uncertain problems. It can address the challenges faced by traditional control algorithms, such as the need for patients to manually estimate carbohydrate intake before meals, the requirement to establish complex dynamic models, and the need for professional prior knowledge. However, reinforcement learning is essentially a highly exploratory trial-and-error learning strategy, which is contrary to the high-safety requirements of clinical practice. Therefore, achieving safer control has always been a major challenge for the clinical application of DRL. This paper addresses this challenge by combining the advantages of DRL and the traditional control algorithm-model predictive control (MPC). Specifically, by using the blood glucose and insulin data generated during the interaction between DRL and patients in the learning process to learn a blood glucose prediction model, the problem of MPC needing to establish a patient's blood glucose dynamic model is solved. Then, MPC is used for forward-looking prediction and simulation of blood glucose, and a safety controller is introduced to avoid unsafe actions, thus restricting DRL control to a safer range. Experiments on the UVA/Padova glucose kinetics simulator approved by the US Food and Drug Administration (FDA) show that the time proportion of adult patients within the healthy blood glucose range under the control of the model proposed in this paper reaches 72.51%, an increase of 2.54% compared with the baseline model, and the proportion of severe hyperglycemia and hypoglycemia events is not increased, taking an important step towards the safe control of blood glucose.

Keywords:  artificial pancreas; blood glucose control; model predictive control; reinforcement learning

DOI:  https://doi.org/10.3390/bios16010047
Comput Methods Biomech Biomed Engin. 2026 Jan 27. 1-30

Efficient feature selection with attention based deep cat convolutional stacked sparse autoencoder for diabetes prediction.

G Thilagavathi, N K Karthikeyan.

  Diabetes, one of the most serious diseases in the world, however, early detection can prevent diabetes. This work proposes a novel approach to identifying early signs of diabetes based on deep learning methods. First, the input data is pre-processed and the features are selected using an improved Cheetah Optimization (ICO). Finally, diabetes is classified using a dual attention-based deep cat convolutional stacked sparse autoencoder model (DA_DCC_SSAE). The proposed study improves the results and proves that the proposed method produces better results in terms of accuracy (98.4% - dataset-1, 98% - dataset-2, 97.4% - dataset-3, and 96.8% - dataset-4.

Keywords:  Diabetes prediction; convolutional layer; dual attention module; enhanced cat swarm optimization (ECSO); improved cheetah optimization (ICO); stacked sparse autoencoder

DOI:  https://doi.org/10.1080/10255842.2026.2613708
Bioengineering (Basel). 2025 Dec 24. pii: 12. [Epub ahead of print]13(1):

DME-RWKV: An Interpretable Multimodal Deep Learning Framework for Predicting Anti-VEGF Response in Diabetic Macular Edema.

Yan Liu, Xieyang Xu, Jiaying Zhang, Hui Wang, Ao Shen, Xuefei Song, Xiaofang Xu, Yao Fu.

  Diabetic macular edema (DME) is a leading cause of vision loss, and predicting patients' response to anti-vascular endothelial growth factor (anti-VEGF) therapy remains a clinical challenge. In this study, we developed an interpretable deep learning model for treatment prediction and biomarker analysis. We retrospectively analyzed 402 eyes from 371 patients with DME. The proposed DME-Receptance Weighted Key Value (RWKV) integrates optical coherence tomography (OCT) and ultra-widefield (UWF) imaging using Causal Attention Learning (CAL), curriculum learning, and global completion (GC) loss to enhance microlesion detection and structural consistency. The model achieved a Dice coefficient of 71.91 ± 8.50% for OCT biomarker segmentation and an AUC of 84.36% for predicting anti-VEGF response, outperforming state-of-the-art methods. By mimicking clinical reasoning with multimodal integration, DME-RWKV demonstrated strong interpretability and robustness, providing a promising AI framework for precise and explainable prediction of anti-VEGF treatment outcomes in DME.

Keywords:  causal attention; curriculum learning; diabetic macular edema; epiretinal membrane; multimodal deep learning; optical coherence tomography; ultra-widefield imaging

DOI:  https://doi.org/10.3390/bioengineering13010012
Healthcare (Basel). 2026 Jan 12. pii: 183. [Epub ahead of print]14(2):

Integrating Ophthalmology, Endocrinology, and Digital Health: A Bibliometric Analysis of Telemedicine for Diabetic Retinopathy.

Theofilos Kanavos, Effrosyni Birbas.

  Background/Objectives: Telemedicine has emerged as a pivotal approach to improving access to diabetic retinopathy (DR) screening, diagnosis, management, and monitoring. Over the past two decades, rapid advancements in digital imaging, mobile health technologies, and artificial intelligence have substantially expanded the role of teleophthalmology in DR, resulting in a large volume of pertinent publications. This study aimed to provide a scientific overview of telemedicine applied to DR through bibliometric analysis. Methods: A search of the Web of Science Core Collection was conducted on 15 November 2025 to identify English-language original research and review articles regarding telemedicine for DR. Bibliographic data from relevant publications were extracted and underwent quantitative analysis and visualization using the tools Bibliometrix and VOSviewer. Results: A total of 515 articles published between 1998 and 2025 were included in our analysis. During this period, the research field of telemedicine for DR exhibited an annual growth rate of 13.14%, with publication activity markedly increasing after 2010 and peaking in 2020-2021. Based on the number of publications, United States, China, and Australia were the most productive countries, while Telemedicine and e-Health, Journal of Telemedicine and Telecare, and British Journal of Ophthalmology were the most relevant journals in the field. Keyword co-occurrence analysis revealed three major thematic clusters within the broader topic of telemedicine and DR, namely, public health-oriented work, telehealth service models, and applications of artificial intelligence technologies. Conclusions: The role of telemedicine in DR detection and care represents an expanding multidisciplinary field of research supported by contributions from multiple authors and institutions worldwide. As technological capabilities continue to evolve, ongoing innovation and cross-domain collaboration could further advance the applications of teleophthalmology for DR, promoting more accessible, efficient, and equitable identification and management of this condition.

Keywords:  artificial intelligence; diabetes mellitus; diabetic retinopathy; fundus photography; monitoring; screening; telecare; telehealth; telemedicine; teleophthalmology

DOI:  https://doi.org/10.3390/healthcare14020183
Sci Prog. 2026 Jan-Mar;109(1):109(1): 368504261418899

Intelligent technology-driven diabetes prevention and control: From informatization management to artificial intelligence.

Geer Deng, Wang Chengshi.

  The escalating prevalence of diabetes, along with its complications and mortality risks, imposes a substantial disease burden worldwide. The current suboptimal medical conditions and poor self-management among diabetic patients have exacerbated the deterioration of diabetes globally, particularly in economically underdeveloped countries. However, this situation may now be approaching a turning point. With the constantly advancement of intelligent technologies, the widespread adoption of information management systems and the rise of artificial intelligence have made it possible to enhance the efficiency of diabetes treatment and reduce management costs. Therefore, we have reviewed the relevant literature and conducted a narrative review following the guidance of the Scale for the Assessment of Narrative Review Articles (SANRA). The present paper provides a narrative review of research advances from information management to artificial intelligence in the field of diabetes treatment and management, while also discussing the opportunities and challenges in clinical translation and application. The present review offers a conceptual framework to inform future research and development in intelligent diabetes care.

Keywords:  Diabetes; artificial intelligence; diagnosis; informatization management; precision prevention

DOI:  https://doi.org/10.1177/00368504261418899
JMIR Diabetes. 2026 Jan 26. 11 e79166

Privacy-Preserving Collaborative Diabetes Prediction in Heterogeneous Health Care Systems: Algorithm Development and Validation of a Secure Federated Ensemble Framework.

Md Rakibul Hasan, Juan Li.

   Background: Diabetes prediction requires accurate, privacy-preserving, and scalable solutions. Traditional machine learning models rely on centralized data, posing risks to data privacy and regulatory compliance. Moreover, health care settings are highly heterogeneous, with diverse participants, hospitals, clinics, and wearables, producing nonindependent and identically distributed data and operating under varied computational constraints. Learning in isolation at individual institutions limits model generalizability and effectiveness. Collaborative federated learning (FL) enables institutions to jointly train models without sharing raw data, but current approaches often struggle with heterogeneity, security threats, and system coordination.
Objective: This study aims to develop a secure, scalable, and privacy-preserving framework for diabetes prediction by integrating FL with ensemble modeling, blockchain-based access control, and knowledge distillation. The framework is designed to handle data heterogeneity, nonindependent and identically distributed distributions, and varying computational capacities across diverse health care participants while simultaneously enhancing data privacy, security, and trust.
Methods: We propose a federated ensemble learning framework, FedEnTrust, that enables decentralized health care participants to collaboratively train models without sharing raw data. Each participant shares soft label outputs, which are distilled and aggregated through adaptive weighted voting to form a global consensus. The framework supports heterogeneous participants by assigning model architectures based on local computational capacity. To ensure secure and transparent coordination, a blockchain-enabled smart contract governs participant registration, role assignment, and model submission with strict role-based access control. We evaluated the system on the PIMA Indians Diabetes Dataset, measuring prediction accuracy, communication efficiency, and blockchain performance.
Results: The FedEnTrust framework achieved 84.2% accuracy, with precision, recall, and F1-score of 84.6%, 88.6%, and 86.4%, respectively, outperforming existing decentralized models and nearing centralized deep learning benchmarks. The blockchain-based smart contract ensured 100% success for authorized transactions and rejected all unauthorized attempts, including malicious submissions. The average blockchain latency was 210 milliseconds, with a gas cost of ~107,940 units, enabling secure, real-time interaction. Throughout, patient privacy was preserved by exchanging only model metadata, not raw data.
Conclusions: FedEnTrust offers a deployable, privacy-preserving solution for decentralized health care prediction by integrating FL, ensemble modeling, blockchain-based access control, and knowledge distillation. It balances accuracy, scalability, and ethical data use while enhancing security and trust. This work demonstrates that secure federated ensemble systems can serve as practical alternatives to centralized artificial intelligence models in real-world health care applications.

Keywords:  AI; artificial intelligence; blockchain; decentralized health care; diabetes prediction; ensemble learning; federated learning; knowledge distillation; privacy-preserving AI

DOI:  https://doi.org/10.2196/79166
Cureus. 2025 Dec;17(12): e99936

Identification of Key Genes in Diabetic Retinopathy Using Bioinformatics and Machine Learning Approaches.

Xiaoyan Zhu, Lulu Lian.

  Objective This study aimed to identify key genes associated with diabetic retinopathy (DR) by applying bioinformatics and machine learning techniques to publicly available transcriptomic datasets. We further evaluated their diagnostic performance and explored their potential biological functions and upstream regulatory mechanisms, providing a theoretical basis for the early diagnosis and molecular-targeted therapy of DR. Methods DR-related transcriptomic datasets GSE94019 and GSE60436 were obtained from the Gene Expression Omnibus (GEO) database, with GSE94019 serving as the training set and GSE60436 as the validation set. The data were then subjected to normalization and differential expression analysis. Feature genes were selected using the Least Absolute Shrinkage and Selection Operator (LASSO) regression and Support Vector Machine-Recursive Feature Elimination (SVM-RFE) algorithms. Overlapping genes were identified as key candidates. Diagnostic performance was evaluated by plotting receiver operating characteristic (ROC) curves using the R package pROC. Functional enrichment analysis, including Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses, was performed on differentially expressed genes (DEGs) associated with the key gene. Potential upstream miRNAs and lncRNAs were predicted using the miRanda, miRDB, TargetScan, and spongeScan databases, and a lncRNA-miRNA-mRNA regulatory network was constructed. Results A total of 790 DEGs were identified, including 370 upregulated and 419 downregulated genes. Cross-validation using LASSO and SVM-RFE identified Collagen Type VI Alpha 2 Chain (COL6A2) and LINC01247 as key genes. COL6A2 was significantly upregulated in the DR group. ROC analysis revealed high diagnostic accuracy, with area under the curve (AUC) values of 1.00 (training set) and 0.89 (validation set). In contrast, LINC01247 was significantly downregulated, but its AUC values were 1.00 (training set) and 0.52 (validation set), indicating limited diagnostic value; thus, it was excluded from further analysis. Functional enrichment centered on COL6A2 suggested that its associated DEGs were involved in aberrant extracellular matrix (ECM) organization, cell adhesion, angiogenesis, and inflammatory responses. Moreover, regulatory network analysis indicated that hsa-miR-762 and hsa-miR-29a-3p may indirectly regulate COL6A2 expression by competitively binding multiple lncRNAs (e.g., PABPC1L2B-AS1 and RP11-223P11.3), forming a potential ceRNA regulatory axis. Conclusion This study identifies COL6A2 as a key gene in DR, characterized by significant upregulation in DR tissues and close involvement in ECM remodeling, cell adhesion, and angiogenesis. These findings provide novel molecular targets and theoretical insights for elucidating the molecular mechanisms of DR and for improving early diagnostic strategies.

Keywords:  bioinformatics; col6a2; diabetic retinopathy; extracellular matrix; machine learning

DOI:  https://doi.org/10.7759/cureus.99936
J Tissue Viability. 2025 Nov 27. pii: S0965-206X(25)00123-8. [Epub ahead of print]35(1): 100975

Diagnostic accuracy of artificial intelligence to identify diabetic foot ulcers: A systematic review and meta-analysis.

Zhigang Chen, Xinliang Liu, Simeng Li, Zhenheng Wu, Haifen Tan, Fuqian Yu, Dongmei Wang, Changhua Zhu, Pengfei Lin.

   BACKGROUND: The accuracy of artificial intelligence (AI) in diagnosing diabetic foot ulcers (DFUs) in dermatology remains uncertain.
OBJECTIVE: To summarize the diagnostic accuracy of AI for DFUs and to provide specific theoretical basis for clinical diagnosis.
METHODS: From the inception of the database up to November 3, 2024, we performed an extensive search across several databases, including PubMed, Web of Science (WoS), Embase, Scopus, the Cochrane Library, Wanfang, and the China National Knowledge Infrastructure (CNKI). To assess the overall efficacy of AI in diagnostic testing, we utilized combined metrics of sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and the area under the curve (AUC). Finally, we assessed the presence of publication bias using the Deeks' funnel plot asymmetry test.
RESULTS: In this meta-analysis, a total of 16 references were identified. The summary diagnostic performance is as follows: sensitivity, 0.89 (95 % CI, 0.85-0.92); specificity, 0.93 (95 % CI, 0.90-0.95); PLR, 6.31 (95 % CI, 5.67-7.02); NLR, 0.14 (95 % CI, 0.12-0.15); DOR, 58.22 (95 % CI, 50.18-67.55); and AUC, 0.97 (95 % CI, 0.95-0.98). Subgroup analysis showed the best performance observed in studies with 300 to 1000 samples. Furthermore, the Fagan plot indicates an increase in post-test probability from 10 % pre-test to 59 % post-test.
CONCLUSION: In summary, our results suggest that AI has high accuracy in diagnosing DFUs.

Keywords:  AUC; Accuracy; Artificial intelligence; Diabetic foot ulcers (DFUs)

DOI:  https://doi.org/10.1016/j.jtv.2025.100975
Medicina (Kaunas). 2025 Dec 19. pii: 7. [Epub ahead of print]62(1):

Few-Shot Transfer Learning for Diabetes Risk Prediction Across Global Populations.

Shrinit Babel, Sunit Babel, John Hodgson, Enrico Camporesi.

  Background and Objectives: Type 2 diabetes mellitus (T2DM) affects over 537 million adults worldwide and disproportionately burdens low- and middle-income countries, where diagnostic resources are limited. Predictive models trained in one population often fail to generalize across regions due to shifts in feature distributions and measurement practices, hindering scalable screening efforts. Materials and Methods: We evaluated a few-shot domain adaptation framework using a simple multilayer perceptron with four shared clinical features (age, body mass index, mean arterial pressure, and plasma glucose) across three adult cohorts: Bangladesh (n = 5288), Iraq (n = 662), and the Pima Indian dataset (n = 768). For each of the six source-target pairs, we pre-trained on the source cohort and then fine-tuned on 1, 5, 10, and 20% of the labeled target examples, reserving the remaining for testing; a final 20% few-shot version was compared with threshold tuning. Discrimination and calibration performance metrics were used before and after adaptation. SHAP explainability analyses quantified shifts in feature importance and decision thresholds. Results: Several source → target transfers produced zero true positives under the strict source-only baseline at a fixed 0.5 decision threshold (e.g., Bangladesh → Pima F1 = 0.00, 0/268 diabetics detected). Few-shot fine-tuning restored non-zero recall in all such cases, with F1 improvements up to +0.63 and precision-recall gains in every zero-baseline transfer. In directions with moderate baseline performance (e.g., Bangladesh → Iraq, Iraq → Pima, Pima → Iraq), 20% few-shot adaptation with threshold tuning improved AUROC by +0.01 to +0.14 and accuracy by +4 to +17 percentage points while reducing Brier scores by up to 0.14 and ECE by approximately 30-80% (suggesting improved calibration). All but one transfer (Iraq → Bangladesh) demonstrated statistically significant improvement by McNemar's test (p < 0.001). SHAP analyses revealed population-specific threshold shifts: glucose inflection points ranged from ~120 mg/dL in Pima to ~150 mg/dL in Iraq, and the importance of BMI rose in Pima-targeted adaptations. Conclusions: Leveraging as few as 5-20% of local labels, few-shot domain adaptation enhances cross-population T2DM risk prediction using only routinely available features. This scalable, interpretable approach can democratize preventive screening in diverse, resource-constrained settings.

Keywords:  Type 2 diabetes mellitus; domain adaptation; few-shot learning; global health; transfer learning

DOI:  https://doi.org/10.3390/medicina62010007
Sci Rep. 2026 Jan 27. 16(1): 3675

Comprehensive blood glucose level prediction from HbA1c levels using machine learning models across the biological range.

Tugba Muhlise Okyay, Ibrahim Yilmaz, Macit Koldas.



Keywords:  Data augmentation; Glucose prediction; HbA1c ; Machine learning; Transformation

DOI:  https://doi.org/10.1038/s41598-025-20234-z