bims-aukdir 2026-02-08 papers

bims-aukdir

Biomed News

on Automated knowledge discovery in diabetes research

Issue of 2026–02–08
29 papers selected by
Mott Given

Real-world performance of an AI system for diabetic retinopathy screening.
The artificial intelligence driven on the development of diabetic retinopathy prognostic scoring tool among type 2 diabetes mellitus patients: A review.
Artificial intelligence approaches for non-invasive diabetes prediction using ECG signals: A systematic review.
Multi-feature integrated machine learning prediction model for early nephropathy in elderly living with type 2 diabetes mellitus.
Development of an explainable machine learning model for predicting depression in adults with type 2 diabetes mellitus: A cross-sectional SHAP-based analysis of NHANES 2009-2023.
Validation of a Trained AI Imaging Model for Detecting Diabetic Foot Deformities.
Federated multimodal AI for precision-equitable diabetes care.
Artificial Intelligence-Based Medical Devices for Diabetic Retinopathy Screening in the European Union.
Machine Learning-Based Classification of Diabetes: Model Accuracy, Feature Importance, and Clinical Implications.
Predictive Performance of Artificial Intelligence Algorithms for Gestational Diabetes Mellitus in Pregnant Women: Systematic Review and Meta-Analysis.
RenoTrue: A diabetes-specific machine learning model to estimate glomerular filtration rate for people with diabetes.
A panoramic perspective: application prospects and outlook of multimodal artificial intelligence in the management of diabetic retinopathy.
Explainable reinforcement learning for glucose monitoring based on shapley value analysis.
An enhanced diabetes prediction using an improved hybrid deep learning algorithm with mountain gazelle optimizer.
DGHME:A Hierarchical Hybrid Expert Multi-task Learning Model for Disease Grouping in Diabetes Complication Prediction.
Temporal machine learning framework for diabetic foot ulcer healing trajectory prediction.
A comprehensive biomechanical phenotyping framework for diabetic foot ulcer risk stratification using multi-modal gait analysis and machine learning.
Noninvasive blood glucose level estimation using bioimpedance spectroscopy and machine learning: an integrated optimization approach.
Predictors of Glycemic Response to Sulfonylurea Therapy in Type 2 Diabetes Over 12 Months: Comparative Analysis of Linear Regression and Machine Learning Models.
An explainable ensemble for diabetic retinopathy grading with a novel confidence quality factor and configurable heatmaps.
Integrating Regenerative Therapies and Artificial Intelligence for Precision Management of Diabetic Foot Ulcers.
Identification of severe hypoglycemia in adults with type 1 diabetes using CGM-based machine learning: evidence from the FGM-Japan study.
The Burden of Delayed Diabetic Retinopathy Management and Use of Artificial Intelligence-Driven Screening Tools: A Systematic Literature Review.
Editorial: Smart dietary management for precision diabetes mellitus care.
Double Graph Attention Network for predicting non-alcoholic fatty liver disease in patients with type 2 diabetes.
Type 2 diabetes prediction without labs: a systems-level neural framework for risk and behavioral network reorganization.
Ensemble neural network modelling for stratified HbA1c prediction: Integrating past glucose measurement as a predictor of glycaemic control.
Risk stratification in diabetic kidney disease: a review of prediction models for methodological advances and clinical application.
Intravitreal faricimab in patients with refractory diabetic macular edema: 6-month fluid analysis using artificial intelligence.

Sci Rep. 2026 Feb 06.

Real-world performance of an AI system for diabetic retinopathy screening.

Lina Berrada, Laurent Crenier, Maria Lytrivi, Agnes Burniat, Elie Motulsky, Miriam Cnop.

  Diabetic retinopathy (DR) is a leading cause of preventable blindness, and the growing global burden of diabetes is placing increasing pressure on ophthalmic services. Artificial intelligence (AI)-based retinal image analysis offers a promising strategy to scale up DR screening while reducing reliance on specialist graders. We assessed the performance of an AI-based DR screening system implemented in a real-world endocrinology clinic at the Erasmus Hospital, Belgium. Adult patients with diabetes underwent non-mydriatic fundus photography, and images were analyzed by the AI system for referable DR and diabetic macular edema. All images were independently graded by a retinal specialist using the Early Treatment Diabetic Retinopathy Study (ETDRS) classification as the reference standard. Of 405 patients screened, 353 (86.7%) were included in the primary analysis. The AI system achieved an area under the curve of 96.5%, sensitivity of 88.9%, specificity of 98.7%, and high predictive values for referable DR detection. Subgroup analyses showed consistently high accuracy across demographic and clinical strata. Multivariate analysis identified higher HbA1c at diagnosis and longer diabetes duration as significant predictors of referable DR for both AI and human grading. These findings support the robustness, generalizability and operational feasibility of this AI system for DR screening in routine clinical care.

Keywords:  Artificial intelligence; Deep learning; Diabetic retinopathy; Real-world study; Screening; Teleophthalmology

DOI:  https://doi.org/10.1038/s41598-026-37292-6
Med J Malaysia. 2026 Jan;81(1): 158-162

The artificial intelligence driven on the development of diabetic retinopathy prognostic scoring tool among type 2 diabetes mellitus patients: A review.

M M S Teruna, T R Razak, S M Yasin, A A R Ali, M R Isa.

INTRODUCTION: Diabetic retinopathy, a major microvascular complication of type 2 diabetes mellitus, remains a leading cause of preventable blindness worldwide. Early identification of individuals at high risk is essential, yet conventional screening systems are limited by workforce shortages and delayed detection. Artificial intelligence, particularly machine learning, offers substantial potential to support prognostic scoring tools capable of predicting the development of diabetic retinopathy. This review summarises current evidence on AI-driven prognostic models for diabetic retinopathy among adults with type 2 diabetes mellitus.
MATERIALS AND METHODS: A comprehensive PubMed search using Medical Subject Headings and free-text terms related to "Diabetic Retinopathy," "Type 2 Diabetes Mellitus," "Artificial Intelligence," "Machine Learning," and "Prognostic Model" was conducted. Original studies involving adults with T2DM that developed or evaluated AIor ML-based prognostic or risk-scoring tools for DR were included. Extracted data included study design, sample size, artificial intelligence methods, predictors, and model performance, and were synthesised narratively.
RESULTS: From 759 records, five studies met the inclusion criteria. Extreme Gradient Boosting consistently demonstrated the highest predictive performance, with area under the curve values between 0.803 and 0.966. Support Vector Machine also performed well in smaller cohorts. Key predictors across studies included HbA1c, duration of diabetes, renal function markers, blood pressure, lipid profile, and body mass index.
CONCLUSION: AI-driven prognostic tools show strong potential to enhance early diabetic retinopathy risk prediction. However, broader external validation and population-specific calibration are needed before routine clinical adoption.
Comput Methods Programs Biomed. 2026 Feb 03. pii: S0169-2607(26)00032-5. [Epub ahead of print]278 109264

Artificial intelligence approaches for non-invasive diabetes prediction using ECG signals: A systematic review.

Kiruthika Balakrishnan, Durgadevi Velusamy, Karthikeyan Ramasamy, Hana E Hinkle, Holly J Hudson, Ram Bilas Pachori, Hikmat Khan.

  Diabetes is a major global health challenge, with many individuals remaining undiagnosed due to the limitations of traditional screening methods. Artificial intelligence (AI)-based electrocardiogram (ECG) analysis offers a promising, non-invasive approach for the early detection of diabetes. This systematic review aims to critically evaluate machine learning (ML) and deep learning (DL) models developed for non-invasive prediction of diabetes and prediabetes using ECG signals. A comprehensive literature search was conducted across PubMed, Embase, Web of Science, IEEE Xplore, and ACM Digital Library in accordance with PRISMA 2020 guidelines. Twenty-five studies met the inclusion criteria. Extracted data included ECG input types, model architectures, preprocessing methods, feature sets, validation strategies, and performance metrics. Most studies used small, single-site, cross-sectional datasets, with sample sizes ranging from 24 to over 190,000 individuals. ECG preprocessing methods varied widely, including filtering, normalization, and decomposition. Features were extracted from time, frequency, morphological, and non-linear domains, though formal feature selection was applied inconsistently. ML and DL models reported high internal accuracy (>90%) but most lacked external validation and subgroup performance assessments. Notably, no study specifically focused on rural or underserved populations, and only one provided open-source code. AI-based ECG analysis demonstrates strong potential for detecting diabetes; however, current research is limited by generalizability issues, lack of standardized methods, poor external validation, and insufficient transparency. Future studies should prioritize rigorous validation, reproducibility, fairness audits, and applications in rural and underserved settings to ensure equitable and clinically viable deployment of these models.

Keywords:  Artificial intelligence (AI); Deep learning; Diabetes prediction; Electrocardiogram (ECG); Heart rate variability; Machine learning; Model validation; Non-invasive screening; Rural health; Systematic review

DOI:  https://doi.org/10.1016/j.cmpb.2026.109264
Front Endocrinol (Lausanne). 2025 ;16 1660903

Multi-feature integrated machine learning prediction model for early nephropathy in elderly living with type 2 diabetes mellitus.

Tingting Fang, Yuanyuan Yang, Feng Zhuo, Xinran Xie, Jialun Song, Linghua Kong.

   Aims: To develop and validate a multi-feature machine learning (ML) model for early diabetic nephropathy (DN) prediction in elderly living with type 2 diabetes mellitus (T2DM), incorporating clinical indicators, symptoms of traditional Chinese medicine (TCM), and ultrasonic imaging features.
Methods: The valid data (including clinical indicators, TCM symptoms, and ultrasonic imaging features) of 786 patients was retained, and the data were divided into training and validation set. Three models were constructed to examine the model's performance. The optimal indicators were selected for seven ML. Performance was assessed using accuracy, precision, recall, F1 score, and the area under the receiver operating characteristic curve (AUC). The subgroup analysis was conducted based on age.
Results: The multi-feature model, combining clinical data, TCM symptoms, and ultrasound imaging, demonstrated the best performance. Among the ML algorithms, RF exhibited superior performance with an AUC of 0.894, sensitivity of 0.667, specificity of 0.877, precision of 0.769, recall of 0.667, and F1 score of 0.714 in the validation set. Subgroup analysis revealed that the AUC values exceed 0.7 in each group.
Conclusion: This study is the first to incorporate TCM symptoms and ultrasound imaging features into a predictive model for early DN in elderly living with T2DM. The models demonstrated strong predictive performance across different age groups. These findings underscore the potential of early screening, prevention, and intervention in improving outcomes for elderly living with T2DM, offering a novel approach to managing diabetic nephropathy.

Keywords:  early nephropathy; elderly; machine learning; prediction model; type 2 diabetes mellitus

DOI:  https://doi.org/10.3389/fendo.2025.1660903
Medicine (Baltimore). 2026 Feb 06. 105(6): e47522

Development of an explainable machine learning model for predicting depression in adults with type 2 diabetes mellitus: A cross-sectional SHAP-based analysis of NHANES 2009-2023.

Yan Tang, Lei Jia, Junjun Zhou, Jin Dou, Jingjuan Qian, Xin Yi, Kim Lam Soh.

  Depression (DEP) is a common yet underdiagnosed comorbidity in adults with type 2 diabetes mellitus (T2DM), worsening glycemic control and increasing complication risk. Practical, interpretable risk tools using routine patient data are limited. We conducted a cross-sectional analysis using data from adults with T2DM enrolled in the National Health and Nutrition Examination Survey between 2009 and 2023. DEP was classified based on a Patient Health Questionnaire-9 score of 10 or higher. Twenty-eight candidate predictors encompassing demographic characteristics, clinical and biochemical measurements, and lifestyle factors were initially included. Variable selection was performed using least absolute shrinkage and selection operator regression. Five machine learning algorithms - random forest, extreme gradient boosting (XGBoost), multilayer perceptron, logistic regression, and support vector machine - were trained and evaluated using 5-fold cross-validation. The best-performing model was interpreted through SHapley Additive exPlanations analysis to identify the most influential predictors. A streamlined version incorporating the top 10 predictors was further developed and implemented as a user-friendly web-based risk estimation tool. Among 2837 participants, 449 (15.8%) were identified as having comorbid DEP. The XGBoost model demonstrated the highest discriminative ability, with a validation area under the receiver operating characteristic curve of 0.888, accuracy of 0.834, F1-score of 0.715, sensitivity of 0.577, and specificity of 0.979, surpassing the performance of the other algorithms evaluated. SHapley Additive exPlanations analysis revealed gender, poverty-to-income ratio, sleep duration, smoking status, educational levels, race, age, high cholesterol, hypertension, and insulin use as the most influential predictors. A streamlined XGBoost model incorporating only these 10 variables achieved an area under the curve of 0.886, closely matching the predictive capability of the full model. The deployed web-based tool enables rapid and individualized estimation of DEP risk in patients with T2DM using routinely available clinical and demographic information. Explainable machine learning applied to nationally representative data can accurately identify adults with T2DM at heightened risk of DEP using a small set of noninvasive clinical features. The deployed tool offers a scalable, interpretable, and clinically actionable approach to support early detection and intervention, potentially improving mental health outcomes in this high-risk population.

Keywords:  NHANES; SHapley additive exPlanations; depression; machine learning; type 2 diabetes mellitus

DOI:  https://doi.org/10.1097/MD.0000000000047522
Int Wound J. 2026 Feb;23(2): e70821

Validation of a Trained AI Imaging Model for Detecting Diabetic Foot Deformities.

Kher Li Teoh, Khare Shubham Hari, Li Lian Ng, Zhiwen Joseph Lo.

  Diabetes is a leading cause of morbidity and mortality, contributing to complications such as cardiovascular disease, kidney failure and lower-limb amputations. Diabetic foot complications, such as structural deformities, ulceration and infection, present significant risks, necessitating early detection and intervention. This study explores the development and validation of artificial intelligence (AI) image analysis for diabetic foot screening, focusing on structural deformity identification which includes callus, hallux valgus and hammer toes, because they represent the earliest detectable visual risk markers for ulceration, preceding wound formation. Leveraging datasets comprising over 1000 healthy foot images and 215 diabetic foot deformity images, the model employed YOLOv5 for object detection, a convolutional autoencoder for anomaly detection, and DenseNet201 for anomaly classification. Initial internal validation yielded 91.1% anomaly detection accuracy, while anomaly classification accuracy improved to 88.57% following refinement. External validation using 27 participants achieved an overall accuracy of 85.2% and anomaly classification accuracy of 66.7%. Final evaluation on 35 unlabelled images demonstrated promising performance, with 88.57% accuracy, 90.47% precision and an F1 score of 86.11%. Integrated into the 'Foot at Risk' (FAR) mobile application, this AI-driven solution offers a scalable tool for early diabetic foot deformity detection. With larger dataset input for training and development, it can be utilised as an early screening tool for diabetic foot and integrated into existing community diabetic care model, facilitating timely intervention and improving patient outcomes.

Keywords:  artificial intelligence; diabetic foot; foot deformities; mobile applications

DOI:  https://doi.org/10.1111/iwj.70821
Front Digit Health. 2025 ;7 1678047

Federated multimodal AI for precision-equitable diabetes care.

Bing Bai, Xilin Liu, Hong Li.

  Type 2 diabetes mellitus (T2DM) constitutes a rapidly expanding global epidemic whose societal burden is amplified by deep-rooted health inequities. Socio-economic disadvantage, minority ethnicity, low health literacy, and limited access to nutritious food or timely care disproportionately expose under-insured populations to earlier onset, poorer glycaemic control, and higher rates of cardiovascular, renal, and neurocognitive complications. Artificial intelligence (AI) is emerging as a transformative counterforce, capable of mitigating these disparities across the entire care continuum. Early detection and risk prediction have progressed from static clinical scores to dynamic machine-learning (ML) models that integrate multimodal data-electronic health records, genomics, socio-environmental variables, and wearable-derived behavioural signatures-to yield earlier and more accurate identification of high-risk individuals. Complication surveillance is being revolutionised by AI systems that screen for diabetic retinopathy with near-specialist accuracy, forecast renal function decline, and detect pre-ulcerative foot lesions through image-based deep learning, enabling timely, targeted interventions. Convergence with continuous glucose monitoring (CGM) and wearable technologies supports real-time, AI-driven glycaemic forecasting and decision support, while telemedicine platforms extend these benefits to remote or resource-constrained settings. Nevertheless, widespread implementation faces challenges of data heterogeneity, algorithmic bias against minority groups, privacy risks, and the digital divide that could paradoxically widen inequities if left unaddressed. Future directions centre on multimodal large language models, digital-twin simulations for personalised policy testing, and human-in-the-loop governance frameworks that embed ethical oversight, trauma-informed care, and community co-design. Realising AI's societal promise demands coordinated action across patients, clinicians, technologists, and policymakers to ensure solutions are not only clinically effective but also equitable, culturally attuned, and economically sustainable.

Keywords:  artificial intelligence; digital-twin; federated learning; health-equity; precision medicine

DOI:  https://doi.org/10.3389/fdgth.2025.1678047
Ophthalmol Ther. 2026 Jan 30.

Artificial Intelligence-Based Medical Devices for Diabetic Retinopathy Screening in the European Union.

Andrzej Grzybowski, Kai Jin.

   BACKGROUND: Diabetic retinopathy (DR) remains a leading cause of preventable blindness, yet screening programs across Europe face persistent workforce and capacity constraints amid rising diabetes prevalence. Artificial intelligence (AI)-enabled screening platforms have been developed to support scalable DR detection; however, their regulatory status, validation approaches, and implementation readiness vary considerably.
METHODS: We conducted a targeted scoping review of 13 CE-certified AI systems for autonomous or semi-autonomous DR detection available in the European Union as of October 23, 2025 (IDx-DR, EyeArt, RetCAD, Mona DR, Retmarker DR, SELENA+, Remidio Medios AI, RetinoScan, Aireen DR, OphthAI, LuxIA, Airdoc-Eye DR, and Vistel). Data were charted across predefined domains, including device designation, regulatory classification, evidence sources, validation study design, reported diagnostic performance metrics, and implementation-related considerations. The review aimed to map the extent and nature of available evidence without conducting quantitative synthesis or comparative ranking.
RESULTS: Most systems employed deep-learning-based fundus image analysis, often incorporating automated image-quality assessment. Reported sensitivities and specificities for referable DR (RDR) varied across systems, generally falling within ranges consistent with regulatory expectations; however, reporting standards and study designs were heterogeneous, limiting direct comparison. Several systems were supported by multicenter or prospective evaluations, while others relied primarily on retrospective datasets. A subset of platforms reported multi-disease detection capabilities. Evidence specific to sight-threatening DR (STDR) was less frequently described and demonstrated wider variability. Non-EU regulatory pathways were mentioned in some reports, but were outside the primary scope of this review. Other systems demonstrate high diagnostic accuracy in controlled evaluations, though performance for STDR remains limited (mean ≈ 80%), largely due to reliance on single-modality 2D fundus imaging without optical coherence tomography (OCT) integration. Implementation-related evidence, including workflow integration and monitoring requirements under the EU Medical Device Regulation (MDR), was limited across systems.
CONCLUSIONS: CE-certified AI systems for DR detection represent a diverse and rapidly evolving landscape. While substantial progress has been made in regulatory classification and validation efforts, evidence remains heterogeneous, particularly for STDR detection and real-world implementation. Future research should prioritize consistent reporting standards, evaluation of multimodal approaches, and studies addressing real-world effectiveness to support safe and equitable deployment under the evolving EU regulatory framework.

Keywords:  Artificial intelligence; CE certification; Diabetic retinopathy; EU AI Act; EU Medical Device Regulation; Screening

DOI:  https://doi.org/10.1007/s40123-026-01322-3
Acta Inform Med. 2025 ;33(4): 279-283

Machine Learning-Based Classification of Diabetes: Model Accuracy, Feature Importance, and Clinical Implications.

Nour Obeidat, Maher Obeidat, Malek Zihlif, Suha Khayri Ababneh, Israa Yousef.

   Background: Diabetes mellitus (DM) is highly prevalent and often remains undiagnosed until complications appear, especially in low- and middle-income countries. Simple tools that use routinely collected clinical and demographic variables may support earlier identification of individuals at increased risk.
Objective: This study aimed to build a supervised achine-learning model to classify individuals as diabetic or non-diabetic using a large publicly available dataset, and to identify which variables contributed most to the model decisions.
Methods: We analysed a cleaned subset of 89,540 records from a Kaggle diabetes dataset. A multilayer perceptron artificial neural network (ANN) was trained and tested on separate subsets. Model performance was evaluated by overall accuracy and misclassification rates, and post-hoc variable importance scores were used to summarise the contribution of each predictor.
Results: The ANN achieved an overall prediction accuracy of 96.8% in both the training and testing samples. Most records were correctly classified, although the error pattern suggested that non-diabetic cases were recognised more easily than diabetic cases. Blood glucose, HbA1c and body mass index (BMI) showed the highest importance values, whereas demographic and lifestyle variables contributed less to the classification.
Conclusion: In this dataset, an ANN based on simple clinical and demographic variables was able to distinguish between diabetic and non-diabetic records with high internal accuracy and a plausible pattern of variable importance. The model could form the basis for a practical screening aid, but it requires external validation and further work on handling class imbalance and explainability before use in routine care.

Keywords:  Artificial neural network; Diabetes mellitus; Feature importance; Machine learning; Risk prediction

DOI:  https://doi.org/10.5455/aim.2025.33.279-283
J Med Internet Res. 2026 Jan 30. 28 e79729

Predictive Performance of Artificial Intelligence Algorithms for Gestational Diabetes Mellitus in Pregnant Women: Systematic Review and Meta-Analysis.

Yingni Liang, Anran Dai, Meiyan Luo, Zhuolian Zheng, Jiayu Shen, Yinhua Su, Zhongyu Li.

   Background: Gestational diabetes mellitus (GDM) is a common complication during pregnancy, with its incidence increasing year by year. It poses numerous adverse health effects on both mothers and newborns. Accurate prediction of GDM can significantly improve patient prognosis. In recent years, artificial intelligence (AI) algorithms have been increasingly used in the construction of GDM prediction models. However, there is still no consensus on the most effective algorithm or model.
Objective: This study aimed to evaluate and compare the performance of existing GDM prediction models constructed using AI algorithms and propose strategies for enhancing model generalizability and predictive accuracy, thereby providing evidence-based insights for the development of more accurate and effective GDM prediction models.
Methods: A comprehensive search was conducted across PubMed, Web of Science, Cochrane Library, EMBASE, Scopus, and OVID, covering publications from the inception of databases to June 1, 2025, to include studies that developed or validated GDM prediction models based on AI algorithms. Study selection, data extraction, and risk of bias assessment using the Prediction Model Risk of Bias Assessment Tool were performed independently by 2 reviewers. A bivariate mixed-effects model was used to summarize sensitivity and specificity and to generate a summary receiver operating characteristic (SROC) curve, calculating area under the curve (AUC). The Hartung-Knapp-Sidik-Jonkman method was further used to adjust for the pooled sensitivity and specificity. Between-study standard deviation (τ) and variance (τ²) were extracted from the bivariate model to quantify absolute heterogeneity. The Deek test was used to evaluate small-study effects among included studies. Additionally, subgroup analysis and meta-regression were conducted to compare the performance differences among algorithms and to explore sources of heterogeneity.
Results: Fourteen studies reported on the predictive value for AI algorithms for GDM. After adjustment with the Hartung-Knapp-Sidik-Jonkman method, the pooled sensitivity and specificity were 0.78 (95% CI 0.69-0.86; τ=0.15, τ2=0.02; PI 0.47-1.09) and 0.85 (95% CI 0.78-0.92; τ=0.11, τ2=0.01; PI 0.59-1.11), respectively. The SROC curve showed that the AUC for predicting GDM using AI algorithms was 0.94 (95% CI 0.92-0.96), indicating a strong predictive capability. Deek test (P=.03) and the funnel plot both showed clear asymmetry, suggesting the presence of small-study effects. Subgroup analysis showed that the random forest algorithm exhibited the highest sensitivity (0.83, 95% CI 0.74-0.93), while the extreme gradient boosting algorithm exhibited the highest specificity (0.82, 95% CI 0.77-0.87). Meta-regression further revealed an evaluation in predictive accuracy in prospective study designs (regression coefficient=2.289, P=.001).
Conclusions: Unlike previous narrative reviews, this systematic review innovatively provided a comparative and quantitative synthesis of AI algorithms for GDM prediction. This established an evidence-based framework to guide model selection and identified a critical evidence gap. The key implication for real-world application was the demonstrated necessity of local validation before clinical adoption. Therefore, future work should focus on large-scale, prospective validation studies to develop clinically applicable tools.

Keywords:  PRISMA; Preferred Reporting Items for Systematic Reviews and Meta-Analysis; artificial intelligence; gestational diabetes mellitus; meta-analysis; prediction

DOI:  https://doi.org/10.2196/79729
Diabetes Res Clin Pract. 2026 Feb 02. pii: S0168-8227(26)00056-2. [Epub ahead of print]233 113137

RenoTrue: A diabetes-specific machine learning model to estimate glomerular filtration rate for people with diabetes.

Rodney Kwok, Kartik Kishore, Tina Zafari, Digsu N Koye, Mariam Hachem, Ian H de Boer, Tae-Dong Jeong, Won-Ki Min, Esteban Porrini, Petter Bjornstad, Yih Chung Tham, Richard J MacIsaac, Leonid Churilov, Elif I Ekinci.

   BACKGROUND: Existing methods for estimating GFR in people with diabetes have shown inaccuracies when compared to mGFR measurements. We developed and validated an artificial neural network - RenoTrue to improve estimating GFR in people with diabetes.
METHODS: 5,619 individuals from five international cohorts with type 1 and type 2 diabetes was split into training (70%), validation (10%) and test (20%) datasets. RenoTrue was developed to estimate GFR using age, sex, and serum creatinine. The performance was evaluated in the test dataset by estimating agreement, bias (mean difference), and accuracy (p30), and compared to CKD-EPI estimates through a multi-level mixed effect regression model.
FINDINGS: Median mGFR was 75 ml/ min per 1.73 m2 [IQR: 49, 100] and median age was 59 years [IQR: 38, 69]. RenoTrue demonstrated high agreement (ICC: 0.87 (95% CI: 0.78, 0.93)), low bias (-0.57 (95% CI: -1.59, 0.46) ml/min per 1.73 m2) and p30 of 81% (95% CI: 79%, 83%) compared to mGFR measurements. The 2009 CKD-EPI equation had an ICC of 0.86 (95% CI: 0.77, 0.92), bias of 4.17 (95% CI: 3.14, 5.20) ml/min per 1.73 m2 and p30 of 74% (95% CI: 72%, 77%).
CONCLUSION: For people with diabetes, RenoTrue demonstrated better performance compared to the 2009 CKD-EPI equation in terms of estimating GFR across the full range of GFR.

Keywords:  Algorithm; Chronic; Computer; Creatinine; Diabetes Mellitus, Type 1; Diabetes Mellitus, Type 2; Diabetic Nephropathies; Glomerular Filtration Rate; Machine Learning; Neural Networks; Renal Insufficiency

DOI:  https://doi.org/10.1016/j.diabres.2026.113137
Front Public Health. 2025 ;13 1724001

A panoramic perspective: application prospects and outlook of multimodal artificial intelligence in the management of diabetic retinopathy.

Chun Liu, Yu Duan, Hao Wu, Junguo Duan.

  Diabetic retinopathy (DR) is a leading cause of blindness among the working-age population, and its management is challenged by the disease's inherent heterogeneity. Current management paradigms, based on standardized grading, are inadequate for addressing the significant inter-patient variability in disease progression and treatment response, thereby limiting the implementation of personalized medicine. While artificial intelligence (AI) has achieved breakthroughs in unimodal analysis of retinal images, the single dimension of information fails to capture the complete, complex pathophysiology of DR. Against this backdrop, multimodal AI, capable of integrating heterogeneous data from multiple sources, has garnered widespread attention and is regarded as a revolutionary tool to overcome current bottlenecks and achieve a panoramic understanding for the management of each patient. This review aims to systematically explore the frontier research and developmental potential of multimodal AI in DR management. It focuses on its data sources, core fusion technologies, and application framework across the entire management workflow. Furthermore, this review analyzes future challenges and directions, with the goal of providing a theoretical reference and guidance for the advancement of precision medicine in DR.

Keywords:  artificial intelligence; data fusion; diabetic retinopathy; multimoda; precision medicine

DOI:  https://doi.org/10.3389/fpubh.2025.1724001
Comput Methods Programs Biomed. 2026 Jan 29. pii: S0169-2607(26)00034-9. [Epub ahead of print]278 109266

Explainable reinforcement learning for glucose monitoring based on shapley value analysis.

Arsene Adjevi, Abdiwahab Mohamed Abdirashid, Faruk Aktaş, Mustafa Hikmet Bilgehan Ucar, Serdar Solak.

   BACKGROUND AND OBJECTIVE: Effective diabetes management requires continuous regulation of blood glucose in response to complex factors such as diet, activity, stress, and medication. Advances in continuous glucose monitoring and machine learning have improved short-term glucose prediction. However, preprocessing of signals like insulin, carbohydrate intake, heart rate, and activity to better capture metabolic dynamics remains underexplored. Similarly, the integration of predictive models with preventive strategies for guiding interventions is still limited.
METHODS: We propose a research-only decision-support framework combining signal preprocessing, CNN-based glucose prediction, Shapley Additive Explanations (SHAP) values attribution, and an Actor-Critic Reinforcement Learning (RL) agent. Exponential decay models preprocess inputs, a compact CNN forecasts short-term glucose levels, and SHAP values highlights the most influential input features; however, these attributions reflect associative patterns in the data and do not establish or map to causal clinical mechanisms. These SHAP-derived attributions guide the RL agent, which issues bounded one-step behavioral adjustments. Because SHAP-guided RL remains stochastic and uncertain, the proposed system is exploratory and not clinically safe, serving solely as a simulation framework.
RESULTS: Using the OhioT1DM dataset, the model achieved state-of-the-art RMSE across prediction horizons with a compact size of 7̃4 KB per patient and training under one minute for 1000 epochs. Over 98% of predictions fell within Clarke Error Grid Zones A and B, confirming safe 5-20 min forecasts. The preventive component corrected hyper- and hypoglycemia in 2̃5% of cases within 10 min when predictions were near 80-120 mg/dL (±10 mg/dL). When deviations exceed ±10 mg/dL, the RL agent is unable to fully restore blood glucose to the target range within 10 min but can bring it as close as possible to the defined interval.
CONCLUSIONS: This study presents a significant innovation by bridging predictive accuracy, adaptability, and transparency in diabetes management. The integration of a predictive model with Reinforcement Learning (RL) guided by SHAP values, which are typically used for interpretability but here are employed in the learning process, delivers a powerful decision support framework. This approach advances the field toward next-generation, personalized digital health tools.

Keywords:  Deep reinforcement learning; Explainable AI; Glucose-level prediction; SHAP values; Time series prediction

DOI:  https://doi.org/10.1016/j.cmpb.2026.109266
J Diabetes Metab Disord. 2026 Jun;25(1): 46

An enhanced diabetes prediction using an improved hybrid deep learning algorithm with mountain gazelle optimizer.

Mohammadreza Valilou, Sevda Valilou, Farhad Soleimanin Gharehchopogh.

  Diabetes is one of the global health challenges and requires early detection and an accurate diagnosis for the prevention of serious complications. Traditional methods struggle to handle the complexities of modern data sets. Advanced deep learning techniques can yield better solutions. This paper proposes a novel deep-learning framework optimized for diabetes prediction using the Pima Indian Diabetes Dataset. This is suggested to introduce the CatBoost algorithm and a deep learning architecture involving Convolutional Neural Networks (CNNs) and Bidirectional Long Short-Term Memory (Bi-LSTM) networks. Hyperparameter tuning was performed using the Mountain Gazelle optimizer (MGO) to balance exploration and exploitation in the search space effectively. It achieved the best performance, with an accuracy of 0.955, a precision of 0.96, a recall of 0.95, and an F1-score of 0.95, outperforming traditional algorithms such as Logistic Regression and Naive Bayes, which recorded accuracies of 0.775 and 0.78, respectively. Conversely, this proposed approach outperforms other deep learning methods, including CNNs and Bi-LSTMs, across multiple evaluation metrics, demonstrating strength and potential in clinical diagnostics. This enhances method interpretability; therefore, Recursive Feature Elimination (RFE) is an ideal candidate for medical applications in which clarity in decision-making is crucial.

Keywords:  CatBoost; Deep learning; Diabetes prediction; Mountain gazelle optimizer; Optimization

DOI:  https://doi.org/10.1007/s40200-025-01844-w
IEEE J Biomed Health Inform. 2026 Jan 30. PP

DGHME:A Hierarchical Hybrid Expert Multi-task Learning Model for Disease Grouping in Diabetes Complication Prediction.

Xiaowei Tang, Hongru Li, Bowen Han, Chenglei Wang, Lisheng Xu, Xia Yu.

Accurate identification and prediction of diabetes complications contribute to improved patient health. However, existing prediction models predominantly employ single-task learning (STL) paradigms, failing to fully leverage the intrinsic correlations among different complications that arise from shared underlying pathophysiological mechanisms, thereby limiting predictive accuracy. To address this, we propose a disease-grouping hierarchical mixed expert model (DGHME). This model integrates clinical-pathological grouping knowledge into a multi-task learning (MTL) architecture, constructing a hierarchical network comprising a bottom-layer self-attention shared expert, group-internal shared experts, task-private experts, and a global shared expert. Through an adaptive gating mechanism and uncertainty-based loss weighting strategy, it achieves refined learning of both disease-common and disease-specific features. Finally, comparative experiments against existing multi-task baseline models on a real-world dataset demonstrate the superiority of the proposed DGHME. Our ablation studies further indicate that DGHME aids in accurately identifying high-risk patients and enables more effective complication prediction.

DOI: https://doi.org/10.1109/JBHI.2026.3659609
Biomed Eng Online. 2026 Feb 05.

Temporal machine learning framework for diabetic foot ulcer healing trajectory prediction.

Reza Basiri, Asem Saleh, Shehroz S Khan, Milos R Popovic.

   OBJECTIVES: Diabetic foot ulcer management relies predominantly on reactive treatment adjustments based on current wound status. This study developed an accessible machine learning framework using routinely collected clinical metadata (no imaging required) to predict healing phase transitions at the next clinical appointment, enabling proactive treatment planning with an integrated recommendation system.
METHODS: Longitudinal data from 268 patients with 329 distinct ulcers across 890 appointments were analyzed. Features (n = 103) including temporal measurements normalized by inter-appointment intervals were engineered. An Extra Trees classifier was optimized via Bayesian hyperparameter tuning with impurity-based feature selection and sequential augmentation to predict three transition categories: favorable, acceptable, or unfavorable. Threefold patient-level cross-validation ensured robust performance estimation.
RESULTS: Feature selection identified 30 essential predictors, achieving 70.9% dimensionality reduction. The optimized classifier demonstrated 78% ± 4% accuracy with balanced category performance (per-class F1 scores: 0.72-0.84) and average AUC of 0.90. Historical phase features dominated predictive importance. The integrated treatment recommendation system achieved 88.7% within-category agreement for offloading prescriptions across all chronicity levels. Dressing recommendations demonstrated chronicity-stratified performance, with match rates declining from 83.7% for acute wounds to 5.6% for very chronic wounds, appropriately reflecting clinical reality that treatment-resistant wounds require individualized therapeutic experimentation.
CONCLUSIONS: This framework demonstrates potential for next-appointment trajectory prediction using accessible clinical metadata without specialized imaging, pending prospective validation. The chronicity-dependent recommendation performance appropriately distinguishes wounds amenable to standardized protocols from treatment-resistant cases requiring iterative experimentation.

Keywords:  Clinical decision support; Diabetic foot ulcer; ExtraTrees; Healing phase classification; Longitudinal analysis; Machine learning; Temporal prediction; Treatment optimization

DOI:  https://doi.org/10.1186/s12938-026-01529-2
Clin Biomech (Bristol). 2026 Jan 17. pii: S0268-0033(26)00008-2. [Epub ahead of print]133 106753

A comprehensive biomechanical phenotyping framework for diabetic foot ulcer risk stratification using multi-modal gait analysis and machine learning.

Mohsen Jafarzadeh, Ali Tavakoli Golpaygani, Farhad Tabatabaei Ghomshe.

   BACKGROUND: Current diabetic foot ulcer risk assessment methods lack precision in identifying high-risk biomechanical phenotypes. This study aimed to develop a comprehensive biomechanical profiling framework integrating multi-modal gait analysis with machine learning for enhanced ulcer risk stratification.
METHODS: In this prospective cross-sectional study, we analyzed214 participants: active diabetic foot ulcer patients (n = 68), diabetic controls without ulceration (n = 73), and healthy controls (n = 73). We implemented a multi-modal assessment protocol combining high-resolution plantar pressure mapping, wearable inertial sensors, 3D motion capture, and electromyography. Machine learning algorithms included unsupervised learning for phenotyping and supervised learning for predictive modeling, validated through nested cross-validation.
FINDINGS: Diabetic foot ulcer patients demonstrated significantly elevated forefoot pressures (metatarsal 1: 21.3 ± 4.8 vs 15.2 ± 3.1 N/cm2, p < 0.001), altered pressure-time integrals, and cautious gait patterns (velocity: 1.12 ± 0.14 vs 1.45 ± 0.16 m/s, p < 0.001). K-means clustering revealed four distinct biomechanical phenotypes with differential ulceration risks (OR: 3.2-8.7). The random forest model achieved 94.3% accuracy (95% CI: 91.2-96.8%) in classifying diabetic foot ulcer risk using six key biomechanical features, substantially outperforming conventional methods. Dynamic center of pressure analysis identified previously unrecognized instability patterns predictive of ulcer development 6-8 months before clinical presentation.
INTERPRETATION: We identified and validated novel biomechanical phenotypes with differential ulcer susceptibility. The integration of machine learning with multi-modal gait analysis enables precise risk stratification and personalized prevention strategies, representing a paradigm shift from reactive treatment to proactive, phenotype-specific diabetic foot care.

Keywords:  Biomechanics; Diabetic foot ulcer; Gait analysis; Machine learning; Phenotyping; Wearable sensors

DOI:  https://doi.org/10.1016/j.clinbiomech.2026.106753
Med Eng Phys. 2026 Jan 23. 147(2):

Noninvasive blood glucose level estimation using bioimpedance spectroscopy and machine learning: an integrated optimization approach.

Zhongwei Lu, Tian Zhou, Cong Hu, Chuanpei Xu, Shaorong Zhang, Benxin Zhang.

  Noninvasive blood glucose detection can avoid pricking fingers to collect blood to obtain blood glucose level (BGL), which can greatly alleviate the pain of diabetic patients. In this paper, a new noninvasive blood glucose detection method using bioimpedance spectroscopy combined with machine learning technology is proposed. Specifically, a data generation method is introduced that adaptively increases the number of missing samples based on the sample density distribution, thereby addressing the problems of small sample sizes and large estimation errors in extreme BGLs. Subsequently, sparse group LASSO with a weight ratio threshold is employed to simultaneously select the frequencies and features with the largest contribution based on prior knowledge of the data structure. Finally, the XGBoost algorithm is used to construct a machine learning regression model. Optuna is used to achieve integrated optimization of all hyperparameters and is evaluated with five-fold cross-validation. Blood glucose data of healthy people and type 2 diabetes patients were collected through oral glucose tolerance test in the laboratory environment. The test results of the model are satisfactory with a mean absolute relative difference of 9.55%. The clinically acceptable zone A + B (A) for the Clarke Error Grid is 99.38% (90.63%). Therefore, our method is expected to be developed into wearable devices to replace traditional invasive methods.

Keywords:  bioimpedance spectroscopy; blood glucose estimation; machine learning; noninvasive measurement

DOI:  https://doi.org/10.1088/1873-4030/ae23bf
JMIR Diabetes. 2026 Feb 06. 11 e82635

Predictors of Glycemic Response to Sulfonylurea Therapy in Type 2 Diabetes Over 12 Months: Comparative Analysis of Linear Regression and Machine Learning Models.

Shilpa Garg, Robert Kitchen, Ramneek Gupta, Emanuele Trucco, Ewan Pearson.

   Background: Sulfonylureas are commonly prescribed for managing type 2 diabetes, yet treatment responses vary significantly among individuals. Although advances in machine learning (ML) may enhance predictive capabilities compared to traditional statistical methods, their practical utility in real-world clinical environments remains uncertain.
Objective: This study aimed to evaluate and compare the predictive performance of linear regression models with several ML approaches for predicting glycemic response to sulfonylurea therapy using routine clinical data, and to assess model interpretability using Shapley Additive Explanations (SHAP) analysis as a secondary analysis.
Methods: A cohort of 7557 individuals with type 2 diabetes who initiated sulfonylurea therapy was analyzed, with all patients followed for 1 year. Linear and logistic regression models were used as baseline comparisons. A range of ML models was trained to predict the continuous change in hemoglobin A1c (HbA1c) levels and the achievement of HbA1c <58 mmol/mol at follow-up. These models included random forest, extreme gradient boosting, support vector machines, a conventional feedforward neural network, and Bayesian additive regression trees. Model performance was assessed using standard metrics including R² and root mean squared error for regression tasks and area under the receiver operating characteristic for classification. In a subset of 2361 patients, nonfasting connecting peptide (C-peptide) was analyzed as a proxy for β-cell function. SHAP analysis was performed to identify and compare key predictors driving model performance across methods.
Results: All models exhibited similar performance, with no significant advantages of ML techniques over linear regression. For continuous outcomes, Bayesian additive regression trees demonstrated the highest R² (0.445) and lowest root mean squared error (0.105), though the differences among models were minimal. For the binary outcome, extreme gradient boosting achieved the highest area under the receiver operating characteristic curve (0.712), with CIs overlapping those of other models. Across all models, baseline HbA1c was consistently the primary predictor, explaining the majority of the variance. SHAP analyses confirmed that baseline HbA1c, age, BMI, and sex were the most influential predictors. Sensitivity analyses and hyperparameter tuning did not significantly improve model performance. In the C-peptide subset, higher C-peptide levels were associated with greater glycemic improvement (β=-3.2 mmol/mol per log(C-peptide); P<.001).
Conclusions: In this large, population-based cohort, ML models did not outperform traditional regression for predicting glycemic response to sulfonylureas. These findings suggest that limited gains from ML likely reflect an absence of strong nonlinear or high-order interactions in routine clinical data and that available features may not capture sufficient biological heterogeneity for complex models to confer added benefit. The inclusion of a C-peptide subset provides additional mechanistic insight by linking preserved β-cell function with treatment response.

Keywords:  drug response; glycated hemoglobin; linear regression; machine learning models; treatment response prediction; type 2 diabetes

DOI:  https://doi.org/10.2196/82635
Med Biol Eng Comput. 2026 Feb 05.

An explainable ensemble for diabetic retinopathy grading with a novel confidence quality factor and configurable heatmaps.

Javier Civit-Masot, Francisco Luna-Perejon, Luis Muñoz-Saavedra, José María Rodríguez Corral, Manuel Domínguez-Morales, Anton Civit.



Keywords:  Clinical decision support; Deep learning; Diabetic retinopathy; Explainable AI (xAI); Explainable ensemble; ICDR; Medical imaging

DOI:  https://doi.org/10.1007/s11517-026-03514-2
Adv Wound Care (New Rochelle). 2026 Feb 05. 21621918261422388

Integrating Regenerative Therapies and Artificial Intelligence for Precision Management of Diabetic Foot Ulcers.

Hsiu-Jung Liao, Tzu-Chi Liao, Yu-Feng Wu.

   SIGNIFICANCE: Diabetic foot ulcers (DFUs) represent one of the most devastating complications of diabetes, leading to high rates of amputation and mortality. Their multifactorial pathogenesis-including neuropathy, ischemia, infection, and immune dysfunction-creates a chronic inflammatory microenvironment that impairs tissue repair and regeneration.
RECENT ADVANCES: Emerging regenerative strategies using stem cells and extracellular vesicles (EVs) have demonstrated potential to restore vascularization and modulate inflammation. In particular, miRNA-enriched EVs regulate key wound-healing pathways such as angiogenesis, extracellular matrix remodeling, and oxidative stress response. Meanwhile, small-molecule drugs targeting hypoxia and inflammatory cascades are being explored to enhance re-epithelialization and fibroblast migration. Parallel advances in artificial intelligence (AI) and optical sensing-using visible, infrared, or hyperspectral imaging-enable automated wound detection, tissue classification, and healing prediction with high accuracy.
CRITICAL ISSUES: Despite these developments, translation remains limited by unstable therapeutic efficacy, variable biomarker expression, and the absence of standardized evaluation systems. AI-based wound assessment requires robust datasets and clinical validation to ensure reliability across diverse populations.
FUTURE DIRECTIONS: Integrating molecular-targeted therapies with AI-assisted diagnostic platforms could establish a next-generation DFU management framework-combining precise molecular intervention, automated wound monitoring, and personalized treatment planning-to achieve reliable, real-time, and patient-centered wound care.

Keywords:  artificial intelligence; diabetic wound; therapeutic development

DOI:  https://doi.org/10.1177/21621918261422388
Diabetol Int. 2026 Apr;17(2): 21

Identification of severe hypoglycemia in adults with type 1 diabetes using CGM-based machine learning: evidence from the FGM-Japan study.

Naoki Sakane, Yushi Hirota, Akane Yamamoto, Junnosuke Miura, Hiroko Takaike, Sari Hoshina, Masao Toyoda, Nobumichi Saito, Kiminori Hosoda, Masaki Matsubara, Atsuhito Tone, Satoshi Kawashima, Hideaki Sawaki, Tomokazu Matsuda, Masayuki Domichi, Akiko Suganuma, Seiko Sakane, Takashi Murata.

   Background: Severe hypoglycemia (SH) in adults with type 1 diabetes mellitus (T1DM) is associated with significant morbidity and mortality; however, its underlying causes are often complex and multifactorial. Improved tools to identify individuals at a high risk of SH are critically needed. In this study, machine learning techniques were applied to continuous glucose monitoring (CGM) data to identify distinguishing features between individuals with and without SH episodes.
Methods: We analyzed data from the real-world study of adults with T1DM enrolled in the FGM-Japan study. Eleven machine learning algorithms using continuous glucose monitoring (CGM) metrics were applied to identify SH and assess the relative importance of the contributing features. The CGM metrics included mean glucose/GMI, time above range (TAR > 250 and > 180 mg/dL), time in range (TIR 70-180 mg/dL), time below range (TBR < 70 and < 54 mg/dL), coefficient of variation (%CV), and glycemic risk index (GRI). The model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, precision, recall, and F1 score.
Results: Data from 264 adults with T1DM were analyzed. Across the models, XGBoost showed the highest AUC, significantly outperforming logistic regression, k-NN, and SVM but performed marginally below Naive Bayes. The F1-score analysis showed that logistic regression and neural networks provided a better balance between precision and recall. The model using four CGM variables (TBR < 70, %CV, GMI, and GRI) achieved the highest AUC of 0.794.
Conclusions: XGBoost offers strong overall discrimination; however, simpler models exhibit better F1 performance. Features like 'TBR', '%CV', 'GMI,' and 'GRI' were key features, suggesting their usefulness in identifying individuals at risk for adverse glycemic events.
Trial registration: Clinical Trial Registry No. UMIN000039376.
Supplementary Information: The online version contains supplementary material available at 10.1007/s13340-025-00872-4.

Keywords:  Continuous glucose monitoring; Machine learning; Severe hypoglycemia; Type1 diabetes

DOI:  https://doi.org/10.1007/s13340-025-00872-4
Ophthalmol Ther. 2026 Feb 07.

The Burden of Delayed Diabetic Retinopathy Management and Use of Artificial Intelligence-Driven Screening Tools: A Systematic Literature Review.

Firas Rahhal, Jun Zhang, Munia Mukherjee.

   PURPOSE: Patients with diabetic retinopathy (DR) are at risk of visual deterioration owing to systemic and financial barriers in accessing appropriate care. DR screening tools that implement artificial intelligence (AI) algorithms are gaining recognition due to their accuracy and high-throughput potential. This systematic literature review aimed to understand the economic, humanistic, and clinical burden associated with delayed DR management and the impact of AI-based screening tools for diagnosis and treatment.
METHODS: MEDLINE, Embase, and Cochrane Library databases were searched per Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (1 January 2014 to 28 October 2024). Screening, extraction, and quality assessment were performed by two independent reviewers. Supplementary searching was conducted to evaluate visual outcomes.
RESULTS: In total, 33 records were included. Economic evidence demonstrated that infrequent screening was initially cost-saving but decreased patient quality-adjusted life years, delayed sight-threatening DR diagnosis, and resulted in high treatment-related costs in the long term. Several studies found delayed DR treatment to adversely impact visual acuity, central subfield thickness, and time spent with vision loss. The majority of economic studies evaluating AI-based screening found its use to result in lower overall costs than conventional screening, while two noted higher costs attributable to greater screening uptake and increased specialist referrals. Most studies that modeled clinical impact found AI-based screening to reduce blindness or vision loss versus conventional screening.
CONCLUSIONS: This research underscored the considerable harms associated with delayed DR diagnosis and treatment. AI-based screening tools have the potential to become powerful instruments in supporting improved clinical outcomes for patients and economic benefits for healthcare systems.

Keywords:  Artificial intelligence; Delayed diagnosis; Diabetic retinopathy; Screening programs; Systematic literature review; Vision loss

DOI:  https://doi.org/10.1007/s40123-026-01329-w
Front Nutr. 2026 ;13 1757124

Editorial: Smart dietary management for precision diabetes mellitus care.

Shaker El-Sappagh, Radwa Hassan.



Keywords:  AI in diabetes; diabetes mellitus; digital health (eHealth); mHealth; precision medicine; smart dietary management

DOI:  https://doi.org/10.3389/fnut.2026.1757124
Artif Intell Med. 2026 Jan 29. pii: S0933-3657(26)00021-7. [Epub ahead of print]174 103369

Double Graph Attention Network for predicting non-alcoholic fatty liver disease in patients with type 2 diabetes.

Tianbin Chen, Yongbin Zeng, Jinlin Wang, Xiao Sun, Sihao Liu, Ya Fu, Qiang Yi, Qishui Ou, Kai Yan, Zhiheng Zhou.

  Type 2 diabetes mellitus (T2DM) is a chronic metabolic disease, while non-alcoholic fatty liver disease (NAFLD) is the most prevalent chronic liver disease, which can progress to more severe liver diseases such as liver fibrosis, cirrhosis and hepatocellular carcinoma. Approximately 50%-70% of T2DM patients also have NAFLD. Traditional diagnostic methods like liver biopsy have limitations, making large-scale screening difficult. In the past decade, machine learning have emerged as crucial tools for assisting in NAFLD diagnosis. In this paper, we propose a novel Dual Graph Attention Network (DGAN) for diagnosing NAFLD in T2DM patients. We model the NAFLD diagnosis problem as a node classification task on graph by using features similarity constructed graph. The model includes a Feature Attention Module to capture feature importance through a feature graph and a Patient Attention Module to evaluate patient importance using graph attention mechanisms. These components enhance the model's classification accuracy by leveraging both feature and topological information. The model was trained and tested on clinical data from 2402 T2DM patients, demonstrating superior accuracy in identifying NAFLD compared to other models.

Keywords:  Deep learning; Graph attention neural network; Non-alcoholic fatty liver disease; Type 2 diabetes mellitus

DOI:  https://doi.org/10.1016/j.artmed.2026.103369
Front Digit Health. 2025 ;7 1714545

Type 2 diabetes prediction without labs: a systems-level neural framework for risk and behavioral network reorganization.

Mahreen Kiran, Ying Xie, Graham Ball, Nasreen Anjum, Rudolph Schutte, Barbara Pierscionek.

   Background: Prediction models for Type 2 Diabetes Mellitus (T2DM) often rely on biochemical markers such as glycated hemoglobin, fasting glucose, or lipid profiles. While clinically informative, these indicators typically reflect established dysglycemia, limiting their value for early prevention. In contrast, psychosocial stress, sleep disturbance, tobacco use, and dietary quality represent modifiable, non-clinical factors that can be observed long before metabolic abnormalities are clinically detectable. Yet most studies examine these factors in isolation or as additive lifestyle scores, overlooking how their interdependencies reorganize in the preclinical phase. A systems-level approach is therefore needed to capture how disruptions in behavioral coherence signal emerging vulnerability.
Methods: This study develops a dual-analytic framework that integrates Cox proportional hazards models with artificial neural network (ANN) coherence analysis. Using longitudinal data from the UK Biobank (n=15,774; follow-up up to 17 years), we identified non-clinical predictors of incident T2DM and examined how behavioral networks reorganize across health states. Predictors were screened through multivariate survival analysis and mapped into ANN-derived influence matrices to quantify stability, direction, and systemic coherence of relationships among diet, sleep, psychosocial states, and demographics.
Results: Eighteen significant predictors of T2DM onset were identified. Elevated risk was linked to loneliness, psychiatric consultation, emotional distress, insomnia, irregular sleep, tobacco use, and high intake of processed meat, beef, and refined grains. Protective effects were observed for 7-8 h of sleep, oat and muesli consumption, and fermented dairy. ANN analyses revealed a pronounced breakdown of behavioral coherence in T2DM: foods that stabilized mood in healthy individuals became associated with distress, age and BMI lost their anchoring roles, and emotional states emerged as dominant but erratic drivers of diet. These reversals and destabilizations were consistent across model iterations, suggesting robust signatures of preclinical vulnerability.
Conclusion: T2DM risk is better conceptualized as systemic reorganization within behavioral networks rather than the additive effects of isolated factors. By combining survival models with ANN-derived coherence mapping, this study demonstrates that early prediction is possible from modifiable, everyday behaviors without laboratory measures. The framework highlights leverage points for psychologically informed, personalized prevention strategies.

Keywords:  Type 2 Diabetes Mellitus (T2DM); behavioral coherence breakdown; early disease prediction; machine learning; neural network modeling; psychosocial risk factors; survival analysis

DOI:  https://doi.org/10.3389/fdgth.2025.1714545
Bioinformation. 2025 ;21(10): 3941-3946

Ensemble neural network modelling for stratified HbA1c prediction: Integrating past glucose measurement as a predictor of glycaemic control.

Prakruti Dash, Kasala Farzia, Dharashree Priyadarshini, Saurav Nayak.

  Accurate assessment of glycemic control is crucial for effective diabetes management and the prevention of long-term complications. This study employed an ensemble neural network framework, combining a Multi-Layer Perceptron Regressor (MLPR) and Classifier (MLPC) model, to predict and stratify HbA1c using routine fasting (FBS) and post-prandial (PPBS) glucose values from retrospective e-laboratory data (n = 22,920, 2021-2024). The regressor, trained on mean FBS and PPBS values from the preceding three months, achieved an R2 of 81 ± 3.7%, sMAPE of 9.13 ± 4.01% and RMSE of 1.1 ± 0.01, reflecting high predictive accuracy and minimal bias. Partial Dependence and ICE analyses revealed a strong, consistent positive association of FBS with HbA1c across glycaemic ranges. The classifier, based on predicted HbA1c, achieved 87.4% accuracy, 94.3% precision and a Diagnostic Odds Ratio of 35.26 ± 0.36, as confirmed by ROC analysis, which demonstrated superior discrimination compared to traditional glucose metrics.

Keywords:  HbA1c prediction; diagnostic accuracy; ensemble neural network; fasting blood glucose; glycaemic control; machine learning; multi-layer perceptron; post-prandial blood glucose

DOI:  https://doi.org/10.6026/973206300213941
J Transl Med. 2026 Feb 04.

Risk stratification in diabetic kidney disease: a review of prediction models for methodological advances and clinical application.

Yazhi Wang, Jianzhou Wang, Hui Chen.

   BACKGROUND: Diabetic kidney disease (DKD) represents the leading cause of end-stage renal disease (ESRD) worldwide, characterized by a complex pathophysiology and heterogeneous progression. Accurate prediction of the onset, progression, and adverse outcomes of DKD is critical for early intervention and personalized management.
MAIN BODY: This review systematically summarizes the current research on prediction models in DKD, encompassing both diagnostic and prognostic models. It discusses key methodological considerations in model development and validation, with a specific focus on the application of machine learning (ML) techniques in model construction. Furthermore, this article also evaluates the performance of prediction models based on routine clinical parameters and multimodal models integrating multi-omics, imaging, retinal parameters, and renal pathological features. The primary challenges in clinical translation are analyzed, and future directions for optimizing DKD prediction are proposed.
CONCLUSIONS: In summary, advancing the optimization and clinical translation of DKD prediction models holds significant potential to improve patient care. Future research should focus on addressing the existing challenges, aiming to advance risk-stratified and personalized management and inform future precision medicine approaches in nephrology.

Keywords:  Biomarker; Diabetic kidney disease; Machine learning; Multimodal data; Prediction model; Risk assessment; Risk stratification

DOI:  https://doi.org/10.1186/s12967-026-07746-6
BMC Ophthalmol. 2026 Feb 06.

Intravitreal faricimab in patients with refractory diabetic macular edema: 6-month fluid analysis using artificial intelligence.

Nicolas Owlya, Mickael Barbosa, Nicolò Bartolomeo, Yannic Pannatier-Schuetz, Anna Chiara Nascimbeni, Daniela Gallo Castro, Mamadou Pathé Barry, Aude Ambresin.

   PURPOSE: Diabetic macular edema (DME) is a leading cause of vision loss in patients with diabetes. In developed countries, intravitreal (IVT) anti-vascular endothelial growth factor (VEGF) injections are the standard-of-care first-line treatment for DME. However, despite the efficacy of anti-VEGF and associated improvements in prognosis, some patients show only a partial response and continue to require monthly injections. The aim of this study was to investigate the effect of switching from aflibercept 2.0 mg to faricimab (which targets both angiopoietin-2 [Ang-2] and VEGF-A) on visual function, retinal anatomy and intraretinal fluid (IRF) dynamics in patients with refractory DME.
METHODS: A single-center, observational study of patients with aflibercept-resistant DME who switched to IVT faricimab treatment, comprising a 3-month loading phase, during which faricimab was administered monthly (total of four injections), followed by a treat-and-extend regimen. Visual acuity, anatomical parameters, and fluid dynamics were assessed from baseline to Month 6 in an interim analysis.
RESULTS: Fourteen eyes from 10 patients were included. At Month 6, mean best-corrected visual acuity improved by + 2.7 Early Treatment Diabetic Retinopathy Study (ETDRS) letters (not statistically significant). Mean changes in central macular thickness and outer nuclear layer (ONL) thickness at Month 6 were not significant. However, ONL thickness was significantly reduced in multiples ETDRS macular grid subfields at Month 4. Subretinal fluid volume was negligible through Month 6, with most fluid located in the intraretinal layers (97.8-100%). Total IRF decreased by 22% at Month 4, reaching a nadir of - 37% at Month 2. There was no significant change in mean vascular density from Month 0 to Month 4.
CONCLUSION: Faricimab treatment led to modest early improvements in visual acuity and retinal anatomy overall in patients with refractory DME. The reduction in total IRF at Month 4 may be attributable to Ang-2 inhibition in these patients, who had previously not responded to anti-VEGF treatment alone. Longer-term studies are needed to evaluate the durability and long-term efficacy of faricimab for the treatment of refractory DME.

Keywords:  Diabetic macular edema; Faricimab; Refractory; Switch

DOI:  https://doi.org/10.1186/s12886-026-04651-w