bims-aukdir 2025-11-16 papers

bims-aukdir

Biomed News

on Automated knowledge discovery in diabetes research

Issue of 2025–11–16
twenty-two papers selected by
Mott Given

D-TNet: a hybrid Dense Net-transformer model for robust diabetic retinopathy detection.
Association of HbA1c and an updated glucose management indicator (uGMI) with incident diabetic retinopathy in adults with type 1 diabetes: a longitudinal study.
Interpretable machine learning model for predicting recurrence in patients with diabetic foot ulcers.
A machine learning model for hypoglycemia risk prediction in hospitalized patients with diabetes: development and validation.
GLM-DM: language model boosted neural networks for HbA1c trend prediction in diabetes mellitus.
Generative artificial intelligence in predictive analysis of diabetes and its complications: a narrative review.
Prevalence of Diabetic Retinopathy Among Patients With Diabetes and Its Correlation With Inflammatory Marker Levels in Tear Fluid.
Deep Learning Network with Illuminant Augmentation for Diabetic Retinopathy Segmentation Using Comprehensive Anatomical Context Integration.
Artificial intelligence methods in gestational diabetes mellitus prediction: A systematic literature review.
Prediction of the risk of diabetic foot from corneal nerve images using deep learning algorithms.
Decoding diabetic retinopathy risk through hybrid modeling: integrative assessment of machine learning algorithms and advanced regression frameworks leveraging estimated glucose disposal rate (eGDR).
Using artificial intelligence to predict posttransplant diabetes mellitus: a new step toward personalized medicine.
Assessing adoption of human and AI-enabled diabetic retinopathy screening in primary healthcare settings: findings from a pragmatic trial.
Precision integrated identification of predictive first-trimester metabolomics signatures for early detection of gestational diabetes mellitus.
Artificial intelligence identifies individuals with prediabetes using single-lead electrocardiograms.
Correction: Eckart, A.C.; Sharma Ghimire, P. Exploring Predictors of Type 2 Diabetes Within Animal-Sourced and Plant-Based Dietary Patterns with the XGBoost Machine Learning Classifier: NHANES 2013-2016. J. Clin. Med. 2025, 14, 458.
Explainable machine learning for differential diagnosis of diabetic foot infection and osteomyelitis: a two-center study and clinically applicable web calculator using routine blood biomarkers.
Recent advances in glycated hemoglobin test methods: From lab to point of care testing devices.
Machine Learning-Driven E-Nose-Based Diabetes Detection: Sensor Selection and Feature Reduction Study.
Screening of the key single nucleotide polymorphisms in type 2 diabetes mellitus complicated with lower extremity arterial disease by machine learning.
Segmentation and Clustering Analysis of Scanned Foot Models for Intelligent Diabetic Foot Ulcer Offloading Boots.
Exploring Relationships Between Maternal Characteristics, Continuous Glucose Monitoring Data, and Neonatal Hypoglycemia in Gestational Diabetes Pregnancies Using Probabilistic Modeling.

Sci Rep. 2025 Nov 12. 15(1): 39594

D-TNet: a hybrid Dense Net-transformer model for robust diabetic retinopathy detection.

Priyadharshini Sekar, Ramasubramanian Bhoopalan, N Nagaprasad, Tadesse Regassa Mamo, S P Dhanabal, Ramaswamy Krishnaraj.

One of the main causes of blindness in the globe and a dangerous side effect of diabetes is Diabetic Retinopathy (DR). Preventing permanent vision loss and guaranteeing prompt medical intervention depend on early identification and precise DR severity assessment. Conventional techniques for diagnosing DR depend on experts manually examining retinal pictures, which takes time and is vulnerable to subjectivity. Although Artificial Intelligence (AI) technologies have become attractive substitutes, current approaches frequently suffer from unbalanced classification performance, poor generalisability, and trouble differentiating various DR severity levels. This work offers a Hybrid Deep Learning (DL) strategy D-TNet, which combines DenseNet121 for spatial feature extraction with a Transformer architecture for modeling long-range contextual dependencies to overcome these issues. The model detects key DR indicators like microaneurysms, hemorrhages, and neovascularization. A collection of retinal images from APTOS2019 and Messidor-2 datasets was used to train and test this hybrid model for DR severity grading in five different categories namely healthy retina, Mild, Moderate, Proliferative, and Severe. The model detects key DR indicators like microaneurysms, hemorrhages, and neovascularization. Evaluated on two benchmark datasets, the model achieved 97% accuracy, 0.94 F1-score, and 0.93 kappa score on APTOS2019, and 86% accuracy, 0.79 F1-score, and 0.80 kappa score on Messidor-2. These results demonstrate robust and balanced classification across all five DR severity stages. The model overcame the sensitivity and specificity issues frequently seen in traditional AI-based techniques by exhibiting stable and balanced classification throughout all DR phases. This method has the potential to greatly improve diabetic eye care by offering dependable and scalable DR detection, especially in environments with limited resources. Future work includes domain adaptation for cross-dataset validation and real-world deployment, along with the integration of multimodal data such as blood sugar level, fasting glucose, and HbA1c to enhance diagnostic precision.

DOI: https://doi.org/10.1038/s41598-025-23234-1
Diabetologia. 2025 Nov 11.

Association of HbA1c and an updated glucose management indicator (uGMI) with incident diabetic retinopathy in adults with type 1 diabetes: a longitudinal study.

Viral N Shah, Yongjin Xu, Yaghoub Dabiri, Hemanth P Mohanadas, Alan Cheng, Timothy C Dunn.

   AIMS/HYPOTHESIS: This study aimed to compare the predictive performance of HbA1c and a continuous glucose monitoring (CGM)-based updated glucose management indicator (uGMI) in assessing incident diabetic retinopathy risk.
METHODS: We used the data from a previously published longitudinal case-control study that collected CGM data for up to 7 years prior to diagnosis of incident diabetic retinopathy or no retinopathy (control participants) among adults with type 1 diabetes. Mutual information scores (MIS), receiver operating characteristics (ROC) curves and machine learning models were used to assess the associations of diabetic retinopathy with HbA1c, uGMI and CGM-derived metrics.
RESULTS: The uGMI demonstrated a stronger association with incident diabetic retinopathy (MIS 0.148) compared with HbA1c (MIS 0.078). ROC analysis showed that uGMI had a modestly higher AUC (AUC 0.733) than HbA1c (AUC 0.704). Decision tree models incorporating both HbA1c and uGMI did not improve clinically significant diabetic retinopathy risk prediction. Machine learning models confirmed the better predictive value of uGMI, especially for HbA1c values between 54 mmol/mol (7.1% NGSP) and 58 mmol/mol (7.5% NGSP), where diabetic retinopathy risk escalated significantly.
CONCLUSIONS/INTERPRETATION: The uGMI is a slightly stronger predictor of diabetic retinopathy risk compared with HbA1c. HbA1c and uGMI do not appear to be complementary for diabetic retinopathy risk prediction.

Keywords:  CGM; Diabetic retinopathy; HbA1c ; Type 1 diabetes; Updated GMI

DOI:  https://doi.org/10.1007/s00125-025-06599-w
BMJ Open Diabetes Res Care. 2025 Nov 12. pii: e005242. [Epub ahead of print]13(6):

Interpretable machine learning model for predicting recurrence in patients with diabetic foot ulcers.

International "Flow and Toe" Research Team (iFORT)

   BACKGROUND: Diabetic foot ulcer (DFU) is a severe complication of diabetes mellitus, often characterized by a chronic disease course and a high recurrence rate, posing significant challenges to patient management. Accurately predicting DFU recurrence is essential for enhancing patient care and outcomes through timely treatment and intervention. This study aimed to develop a machine learning (ML) model to predict the 3-year recurrence risk in patients with DFU.
METHODS: A total of 494 patients with DFU were included and assigned to a training set and a test set at a 4:1 ratio. Four feature selection methods-least absolute shrinkage and selection operator, minimum redundancy maximum relevance, Fisher score and recursive feature elimination-were applied to the training set, and intersecting features were selected to construct the final feature set. Seven ML algorithms, including logistic regression, support vector machine, random forest, gradient boosting decision tree, AdaBoost, extreme gradient boosting (XGBoost) and light gradient boosting machine, were employed to develop predictive models. The models' parameters were optimized using fivefold cross-validation. Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC). The best-performing model was calibrated using Platt scaling, with calibration performance assessed by the Brier score. ML model interpretability was enhanced using SHapley Additive exPlanations (SHAP) analysis.
RESULTS: The XGBoost model demonstrated superior predictive performance, achieving an AUROC of 0.924 (95% CI 0.867 to 0.967). Following calibration with Platt scaling, the model exhibited a Brier score of 0.096, indicating good calibration. SHAP analysis identified key risk factors that aligned with existing literature and clinical expertise, further validating the model's interpretability and clinical relevance.
CONCLUSION: The XGBoost model demonstrated strong predictive accuracy and clinical relevance in assessing DFU recurrence risk. However, further multicenter validation with a larger sample size is needed to improve its generalizability and clinical applicability.

Keywords:  Diabetic Foot

DOI:  https://doi.org/10.1136/bmjdrc-2025-005242
Diabetes Res Clin Pract. 2025 Nov 08. pii: S0168-8227(25)01008-3. [Epub ahead of print]230 112993

A machine learning model for hypoglycemia risk prediction in hospitalized patients with diabetes: development and validation.

Liren Li, Jiayu Chen, Tongshu Guan, Ze Yu, Jinyuan Zhang, Ronghong Ji, Zike Li, Ming Lei, Fei Gao, Yilei Li, Ping Zheng.

   BACKGROUND: Hypoglycemia is a major obstacle for optimal glucose management in patients with diabetes. Existing prediction models primarily target non-hospitalized settings or use static variables, limiting applicability for hospitalized patients. We aimed to developed and validated a machine learning model predicting inpatient hypoglycemia using electronic medical records data, integrating dynamic clinical variables to improve accuracy and clinical utility.
METHODS AND FINDINGS: We conducted a retrospective study of 37,966 inpatients with diabetes mellitus at Nanfang Hospital (2021-2022). After applying inclusion and exclusion criteria, 2,845 patients were included. Data preprocessing focused on analyzing potential predictors, including demographic characteristics, medication use, comorbidities, and laboratory parameters. Using a stepwise forward variable selection method based on XGBoost, we identified 10 optimal predictors. The cohort was split into training and testing sets at an 8:2 ratio. Predictive performance was assessed via AUC. Ten ML algorithms were evaluated, with CatBoost demonstrating the best performance (AUC = 0.85, PPV = 0.75, NPV = 0.89).
CONCLUSIONS: Our ML-based predictive model for inpatient hypoglycemia shows robust performance and integrates readily available clinical parameters, offering significant potential for early risk identification and preventive intervention. Future research should focus on multicenter validation and seamless integration into clinical workflows to support dynamic risk assessment.

Keywords:  Hypoglycemia; Inpatient; Machine-learning models; Model; Risk prediction

DOI:  https://doi.org/10.1016/j.diabres.2025.112993
Future Sci OA. 2025 Dec;11(1): 2567166

GLM-DM: language model boosted neural networks for HbA1c trend prediction in diabetes mellitus.

Yikun Ban, Xinrui He, Patricia M Verona, Curtiss B Cook, Jingrui He.

   AIM: Predict Hemoglobin A1c (HbA1c) trends, a key metric in diabetes mellitus (DM) management, using readily available patient variables and language models (LMs).
METHODS: We propose GLM (Language Model Boosted Neural Network) -DM, which leverages data augmentation and language model-driven feature encoding to predict HbA1c trends using easily accessible patient-level variables. Our model captures complex relationships among patient characteristics and enhances predictive performance through Generative Adversarial Networks (GANs) for synthetic data augmentation and LMs for feature embedding. By transforming patient profiles into rich latent representations, our approach enables a more comprehensive analysis of how patient-level variables correlate with HbA1c trends over time.
RESULTS: Using clinical data from 257 DM patients, GLM-DM achieves 70.2% accuracy of HbA1c trend prediction, outperforming classic classifiers and transformer-based models. Ablation studies confirm the effectiveness of GAN-based augmentation and LM-driven embedding. Our model achieves 68.2% prediction accuracy for Type 1 DM and 72.7% for Type 2 DM.
CONCLUSION: Proposed approach learns the underlying complex function of HbA1c using clinical variables easily available at the patient visit and leveraging the power of LMs to accurately predict the trend of HbA1c in a period. The model can enhance patient advisories for daily diabetes management without the need for continuous glucose monitoring.

Keywords:  blood glucose; deep learning; diabetes mellitus; generative adversarial network; hemoglobin A1c; language models

DOI:  https://doi.org/10.1080/20565623.2025.2567166
Ann Transl Med. 2025 Oct 31. 13(5): 59

Generative artificial intelligence in predictive analysis of diabetes and its complications: a narrative review.

Ayokunle Osonuga, David B Olawade, Manisha Gore, Oluwayomi B Rotifa, Eghosasere Egbon, Viviane Chinwah, Stergios Boussios.

   Background and Objective: Diabetes mellitus (DM), particularly type 2 diabetes (T2D), represents a significant global health crisis, often complicated by severe and progressive conditions such as retinopathy, neuropathy, and cardiovascular disease. Traditional diagnostic approaches frequently detect these complications at advanced stages, limiting the opportunity for early, effective intervention. This review aims to examine how recent advancements in generative artificial intelligence (AI), particularly large language models (LLMs), can transform diabetes management by enabling earlier detection and more personalized interventions.
Methods: A narrative review was conducted to evaluate the current literature on the application of generative AI and LLMs in diabetes care. The review focused on how these technologies analyse multi-dimensional datasets, including medical imaging, electronic health records (EHRs), genetic profiles, and lifestyle factors, and how they process both structured and unstructured data to enhance predictive analytics and risk stratification for diabetes complications.
Key Content and Findings: Generative AI models have demonstrated significant promise in detecting hidden trends and early risk factors for complications such as diabetic retinopathy and neuropathy, often before clinical symptoms manifest. LLMs enhance predictive performance by synthesising unstructured data sources, such as physician notes and patient-reported outcomes, with clinical datasets. Despite limitations concerning data quality, model transparency, and ethical concerns surrounding data privacy, these technologies offer powerful tools for proactive disease monitoring and personalized care.
Conclusions: Generative AI and LLMs are poised to redefine diabetes management by enabling earlier detection of complications and personalised treatment strategies. Their integration into clinical decision support systems (CDSS) and precision medicine frameworks may reduce the global burden of diabetes, improve patient outcomes, and shift care from reactive to preventative.

Keywords:  Generative artificial intelligence (generative AI); diabetes mellitus (DM); large language models (LLMs); personalized medicine; predictive analytics

DOI:  https://doi.org/10.21037/atm-25-62
Cureus. 2025 Nov;17(11): e96692

Prevalence of Diabetic Retinopathy Among Patients With Diabetes and Its Correlation With Inflammatory Marker Levels in Tear Fluid.

Neha Shibu Elenjickal, Austin Joju Akkara, Pradeoth M Korambayil, Prashanth Varkey Ambookan.

  Background Diabetic retinopathy (DR) is a leading microvascular complication of diabetes mellitus (DM) and a primary cause of vision loss worldwide. Markers such as NF-κB, TNF-α, and IL-6 have been proposed as potential early indicators on the basis of Inflammation playing a critical role in DR pathogenesis. This study aimed to determine the prevalence of DR among asymptomatic diabetic patients and evaluate correlations between tear fluid inflammatory markers and DR severity. Methods A cross-sectional observational study was conducted over two months at a tertiary care hospital in Kerala, India. Although an initial target of 120 participants was set, logistical and technical challenges in tear sample isolation limited recruitment to 50 asymptomatic diabetic patients. Fundus images were captured using a non-mydriatic handheld fundoscopy device with artificial intelligence (AI)-assisted grading of DR. Tear fluid samples, collected using Schirmer paper, were analysed for nuclear factor kappa B (NF-κB), tumor necrosis factor alpha (NF-α), and interleukin-6 (IL-6) via reverse transcription polymerase chain reaction (RT-PCR). Patient demographics, diabetes duration, comorbidities, and treatment history were documented. Associations between the presence of DR, clinical variables, and tear marker levels were analysed. Results DR was detected in 38% of participants, predominantly among patients aged >60 years and with longer diabetes duration. Insulin use and disease duration were significantly associated with DR, while age, gender, and oral antidiabetic use were not. Tear fluid inflammatory markers did not show a significant correlation with DR status. NF-κB was undetectable, and TNF-α and IL-6 levels were inconsistently elevated in DR-positive patients. Methodological factors, including tear sample over-dilution in 2 mL universal transport medium (UTM) and ocular surface conditions, likely contributed to these findings. Conclusion DR prevalence among asymptomatic diabetic patients is high, underscoring the importance of routine retinal screening. Non-mydriatic AI-assisted fundus imaging is an effective and practical tool for early detection. Tear fluid inflammatory markers, as measured in this study, were not predictive of DR, likely due to sample collection and amplification limitations. Future studies optimizing tear collection and RNA stabilization may establish reliable non-invasive biomarker-based screening, enabling earlier intervention and reducing vision-related morbidity in high-risk populations.

Keywords:  diabetes mellitus; diabetic retinopathy; il-6; inflammatory markers; nf-κb; non-mydriatic fundoscopy; tear fluid; tnf-α

DOI:  https://doi.org/10.7759/cureus.96692
Diagnostics (Basel). 2025 Oct 31. pii: 2762. [Epub ahead of print]15(21):

Deep Learning Network with Illuminant Augmentation for Diabetic Retinopathy Segmentation Using Comprehensive Anatomical Context Integration.

Sakon Chankhachon, Supaporn Kansomkeat, Patama Bhurayanontachai, Sathit Intajag.

  Background/Objectives: Diabetic retinopathy (DR) segmentation faces critical challenges from domain shift and false positives caused by heterogeneous retinal backgrounds. Recent transformer-based studies have shown that existing approaches do not comprehensively integrate the anatomical context, particularly training datasets combining blood vessels with DR lesions. Methods: These limitations were addressed by deploying a DeepLabV3+ framework enhanced with more comprehensive anatomical contexts, rather than more complex architectures. The approach produced the first training dataset that systematically integrates DR lesions with complete retinal anatomical structures (optic disc, fovea, blood vessels, retinal boundaries) as contextual background classes. An innovative illumination-based data augmentation simulated diverse camera characteristics using color constancy principles. Two-stage training (cross-entropy and Tversky loss) managed class imbalance effectively. Results: An extensive evaluation of the IDRiD, DDR, and TJDR datasets demonstrated significant improvements. The model achieved competitive performances (AUC-PR: 0.7715, IoU: 0.6651, F1: 0.7930) compared with state-of-the-art methods, including transformer approaches, while showing promising generalization on some unseen datasets, though performance varied across different domains. False-positive returns were reduced through anatomical context awareness. Conclusions: The framework demonstrates that comprehensive anatomical context integration is more critical than architectural complexity for DR segmentation. By combining systematic anatomical annotation with effective data augmentation, conventional network performances can be improved while maintaining computational efficiency and clinical interpretability, establishing a new paradigm for medical image segmentation.

Keywords:  anatomical context; data augmentation; diabetic retinopathy; domain generalization; semantic segmentation; transformer

DOI:  https://doi.org/10.3390/diagnostics15212762
Int J Med Inform. 2025 Oct 30. pii: S1386-5056(25)00375-2. [Epub ahead of print]206 106158

Artificial intelligence methods in gestational diabetes mellitus prediction: A systematic literature review.

Valentina Ivanovic, Md Abu Jafar Sujan, Ole Jakob Mengshoel, Trine Moholdt.

  Gestational diabetes mellitus (GDM) is the most common metabolic disorder in pregnancy, posing risks to both maternal and neonatal health. Artificial intelligence (AI) and machine learning (ML)-based solutions hold the promise of improving GDM prediction, thus enabling earlier and more personalized care. The main objective of this systematic review is to provide a comprehensive overview of AI/ML methods used for GDM prediction, leveraging the data from both the preconception and pregnancy periods. We conducted a PRISMA-guided search across databases including PubMed, Scopus, IEEE, and Web of Science from their inception to May 27th 2024. Studies were included if they applied AI/ML methods to predict GDM and were published and peer-reviewed. We extracted data across 30 dimensions. We performed a dual-framework quality assessment of included studies using PROBAST and the IJMEDI checklist. A total of 78 studies met the inclusion criteria. Logistic regression (46 studies), tree-based models (41 studies), and support vector machines (29 studies) were the most frequently used AI methods. Neural networks were most often reported as best-performing (15 studies), followed by boosting (14 studies), and tree-based methods (13 studies). Twelve studies included preconception data. Clinically relevant metrics such as sensitivity, specificity, and calibration were frequently underreported, with decision-curve analysis rarely applied. Thirteen studies performed external validation, and very few employed causal or explainable modeling approaches. Risk of bias was high in most studies. According to the IJMEDI checklist, most studies insufficiently addressed data preparation, validation, and deployment aspects. AI-based GDM prediction is rapidly evolving, with strong potential for earlier and more personalized interventions. Future work should prioritize transparent reporting, external validation, and development of trustworthy, explainable models using diverse, longitudinal data. Closer collaboration among data scientists, clinicians, and healthcare systems is needed to close the loop from AI innovation to clinical practice.

Keywords:  Artificial intelligence; Gestational diabetes mellitus; Machine learning; Preconception; Pregnancy; Systematic literature review

DOI:  https://doi.org/10.1016/j.ijmedinf.2025.106158
Diabetes Res Clin Pract. 2025 Nov 06. pii: S0168-8227(25)01005-8. [Epub ahead of print]230 112991

Prediction of the risk of diabetic foot from corneal nerve images using deep learning algorithms.

Chang Liu, Lei Zhu, Gavin Patrick O'Donnell, Mui Hui Koh, Mingyi Yu, Isabelle Xin Yu Lee, Ching-Yu Cheng, Jun Zhou, Xinxing Xu, Rick Siow Mong Goh, Yong Liu, Yu-Chi Liu.

   AIMS: We aimed to develop a deep learning algorithm (DLA) for predicting the risk categories of diabetic foot using corneal nerve images.
METHODS: A total of 23,550 images from 942 eyes of 471 participants were included. We first developed a DLA based on corneal nerve images alone. We then combined classic clinical risk factors of diabetic peripheral neuropathy and quantitative corneal nerve parameters to develop hybrid DLAs. Model performances were assessed based on the area under the receiver operating characteristic curve (AUC).
RESULTS: For the DLA using corneal nerve images alone, the AUC was 0.69 for the prediction of diabetic foot and 0.76 for the identification of patients at high risk. For the hybrid DLAs, the algorithm achieved an AUC of 0.94 for the prediction of diabetic foot with the only addition of HbA1c, and an AUC of 0.93 for the identification of patients with high-risk diabetic foot with the incorporation of serum creatinine. When using quantitative corneal nerve parameters, the performance was not improved compared to using corneal nerve images alone.
CONCLUSIONS: Our DLAs integrating corneal nerve images with HbA1c or serum creatinine have good performance in predicting and stratifying diabetic foot risk, providing a new screening approach.

Keywords:  AI-based; Corneal nerves; Diabetic foot ulcers; IVCM; Predictive model

DOI:  https://doi.org/10.1016/j.diabres.2025.112991
J Diabetes Metab Disord. 2025 Dec;24(2): 260

Decoding diabetic retinopathy risk through hybrid modeling: integrative assessment of machine learning algorithms and advanced regression frameworks leveraging estimated glucose disposal rate (eGDR).

Seyed Sobhan Bahreiny, Mohammad Ali Khaksar, Razieh Kazemzadeh, Reza Mohammadpour Fard, Vahid Radmehr, Mehdi Zahedian.

   Objectives: Diabetic retinopathy (DR) is a major microvascular complication of diabetes mellitus and a leading cause of irreversible visual loss. Insulin resistance plays a central role in DR pathogenesis, but direct measurement is challenging in large-scale studies. The estimated glucose disposal rate (eGDR) is a validated surrogate of insulin sensitivity, yet its relationship with DR in population-based cohorts remains underexplored.
Methods: Data from 954 adults with diabetes in the 2015-2020 National Health and Nutrition Examination Survey were analyzed. Sociodemographic, behavioral, and cardiometabolic factors were compared between groups, followed by multivariable logistic regression, restricted cubic spline (RCS) modeling, and threshold effect analyses. Machine learning algorithms were applied for predictive modeling and feature ranking.
Results: Each one-unit increase in eGDR was associated with 21% lower odds of DR (OR = 0.79; p < 0.001). Quartile analysis revealed a graded inverse association, most pronounced in the third quartile (OR = 0.53; p = 0.020), with a significant linear trend (p < 0.001). RCS confirmed a robust association with non-linearity (p = 0.029) and a plateau at higher eGDR values. Two-piecewise regression identified an inflection at eGDR = 3.92, below which the association weakened (p = 0.010). Among eight algorithms, XGBoost achieved the highest discrimination (AUC = 0.894), with Shapley Additive Explanations ranking eGDR among the most influential predictors.
Conclusions: These findings support eGDR as an independent, non-linear, and potentially threshold-dependent biomarker for DR risk stratification, with clinical implications for targeted screening and intervention.
Graphical Abstract:
Supplementary Information: The online version contains supplementary material available at 10.1007/s40200-025-01765-8.

Keywords:  Diabetic retinopathy; Estimated glucose disposal rate; Insulin resistance; Machine learning; NHANES; Restricted cubic spline; Threshold modeling

DOI:  https://doi.org/10.1007/s40200-025-01765-8
Kidney Res Clin Pract. 2025 Nov;44(6): 860-862

Using artificial intelligence to predict posttransplant diabetes mellitus: a new step toward personalized medicine.

Byung Ha Chung.

DOI: https://doi.org/10.23876/j.krcp.25.135
Arch Public Health. 2025 Nov 10. 83(1): 268

Assessing adoption of human and AI-enabled diabetic retinopathy screening in primary healthcare settings: findings from a pragmatic trial.

Anshul Chauhan, Luke Vale, Ankita Kankaria, Vishali Gupta, Gagandeep Kaur, Neha N, Nayanshi Sood, Basavaraj Tigari, Sanjay Bhadada, Mona Duggal.

   INTRODUCTION: Diabetic retinopathy (DR) screening with defined referral pathways is essential for early detection and effective management of DR. This study assessed the adoption of three DR screening (DRS) models in primary healthcare settings, focusing on referral adherence rates and stakeholders' perceptions of the interventions.
METHODS: A cross-sectional study was conducted in the Mohali district of Punjab, India, between February 2023 and January 2024. This pragmatic study compared three DRS arms (n = 200 each): I) facility-based screening at health and wellness centres (HWCs) by non-ophthalmologists, II) community-based AI DRS screening at home, and III) standard care involving counselling and referral to district hospitals (DHs). Participants with referable DR or ungradable images were advised for ophthalmology opinion, and their follow-up status and reasons for any non-compliance were assessed after one month. The adoption (acceptability and scalability) of the DRS was assessed via in-depth interviews with stakeholders involved in providing diabetes and DR care in public health settings.
RESULTS: Among the 600 participants screened, the average age was 58.22 years (SD ± 11.52). Most participants, 300 (59.57%), were aged 51-70 years, comprising 245 (40.77%) males and 355 (59.23%) females. The referral adherence rates were low, ranging from 13% to 17% across Arms I, II, and III. Barriers to follow-up included lack of awareness, financial limitations, health concerns, perceived good eye health, and transportation challenges. Qualitative findings reveal that DRS, implemented through HWCs and community-based models, is feasible and well-accepted by patients. Stakeholders largely supported the implementation of DRS within primary healthcare settings, though responses varied. Likewise, DRS through HWCs and community-based models is feasible and well-accepted among patients.
CONCLUSION: The non-adherence to referrals among study participants is mainly attributable to economic constraints and knowledge gaps. Enhanced point-of-care counselling targeting groups at higher risk of non-adherence for follow-ups, along with a streamlined referral process, can improve the uptake of referral recommendations.
TRIAL REGISTRATION: Clinical Trial Registry of India (CTRI): 2022/10/046283.

Keywords:  Adoption; Artificial intelligence; Diabetic retinopathy screening; Primary healthcare settings

DOI:  https://doi.org/10.1186/s13690-025-01757-3
Cardiovasc Diabetol. 2025 Nov 14. 24(1): 434

Precision integrated identification of predictive first-trimester metabolomics signatures for early detection of gestational diabetes mellitus.

Sapna Sharma, Yalamanchili Venkata Subrahmanyam, Payal Gupta, Sangeetha Vadivel, Mohan Deepa, Ansh Tandon, Sreekumar Sreedevi, Uma Ram, Priyanka Narad, Dharmeshkumar Parmar, Ranjit Mohan Anjana, Anu Raghunathan, Muthuswamy Balasubramanyam, Viswanathan Mohan, Abhishek Sengupta, Jerzy Adamski, Ponnusamy Saravanan, Venkateswarlu Panchagnula, Dandamudi Usharani, Kuppan Gokulakrishnan.

   BACKGROUND AND AIM: Gestational diabetes mellitus (GDM), a common pregnancy-related metabolic disorder, often goes undiagnosed until the second trimester, limiting early intervention opportunities. Given the higher prevalence of GDM in India, there is a critical need to investigate metabolomic biomarkers among Asian Indians, who exhibit greater insulin resistance and are predisposed to developing type 2 diabetes at an earlier age. This study aimed to identify early pregnancy metabolomic signatures predictive of GDM.
METHODS: Among 2115 pregnant women from the STratification of Risk of Diabetes in Early pregnancy (STRiDE) study, we performed untargeted metabolomic profiling using UPLC-MS/MS at early pregnancy (< 16 weeks) plasma samples from 100 women-comprising 50 with GDM and 50 normal (without GDM) based on oral glucose tolerance test (OGTT) at 24-28 weeks. Statistical and machine learning approaches, including logistic regression and random forest (RF), were applied to identify GDM-associated metabolites and construct predictive models. Pathway enrichment analysis was conducted using KEGG database annotations.
RESULTS: A total of 49 metabolites were significantly associated with GDM, primarily involving lipid classes such as phosphatidylcholines, sphingomyelins, and triacylglycerols. RF analysis identified a panel of eight metabolites that achieved best predictive performance (AUC 0.880; 95% CI: 0.809-0.951) for GDM. When combined with conventional clinical risk factors, the integrated model showed comparable prediction of GDM with AUC 0.88;: 95% CI: 0.810-0.952). Enrichment analysis highlighted dysregulated pathways including glycerophospholipid and sphingolipid metabolism, autophagy, and insulin resistance.
CONCLUSION: This study demonstrates the utility of early-pregnancy metabolomic profiling for predicting GDM in Indian women. The eight-metabolite panel offers a promising tool for early risk stratification of GDM, warranting validation in diverse populations.

Keywords:  First trimester; Gestational diabetes mellitus; Indian women; Mass spectrometry; Metabolomics; Prediction

DOI:  https://doi.org/10.1186/s12933-025-02978-0
Cardiovasc Diabetol. 2025 Nov 11. 24(1): 415

Artificial intelligence identifies individuals with prediabetes using single-lead electrocardiograms.

Daisuke Koga, Ryo Kaneda, Chikara Komiya, Satoshi Ohno, Akira Takeuchi, Kazunari Hara, Masato Horino, Jun Aoki, Rei Okazaki, Ryoko Ishii, Masanori Murakami, Kazutaka Tsujimoto, Kenji Ikeda, Hideki Katagiri, Hideyuki Shimizu, Tetsuya Yamada.

   BACKGROUND: Early detection of prediabetes is crucial for diabetes prevention, yet it remains challenging due to its asymptomatic nature and low screening rates. This study aimed to develop and rigorously validate artificial intelligence (AI) models to identify individuals with prediabetes solely using electrocardiograms (ECGs).
METHODS: We defined prediabetes/diabetes based on fasting plasma glucose ≥ 110 mg/dL, hemoglobin A1c ≥ 6.0%, or ongoing diabetes treatment. From a primary cohort of 16,766 health checkup records, 269 ECG features were extracted to develop a novel AI model. The final model was subsequently evaluated using an internal held-out test dataset and an independent external validation cohort (n = 2,456). SHAP (SHapley Additive exPlanations) was applied to assess feature importance and clinical interpretability.
RESULTS: The best-performing model, a LightGBM-based algorithm we termed DiaCardia, achieved an area under the receiver operating characteristic curve (AUROC) of 0.851 in the internal test dataset (sensitivity: 85.7%, specificity: 70.0%). The model demonstrated robust generalizability, achieving an AUROC of 0.785 in the external validation cohort. Furthermore, DiaCardia maintained substantial predictive ability (AUROC: 0.789) after adjustment for six major confounders using propensity score matching. Higher R-wave amplitude in leads aVL and I, and smaller peak interval dispersion were prominent predictors. Notably, a version of DiaCardia using only single-lead (lead I) ECG data achieved a comparable AUROC of 0.844 (sensitivity: 82.3%; specificity: 70.2%).
CONCLUSIONS: This study establishes that an AI model, DiaCardia, can accurately identify individuals with prediabetes from an ECG alone, with performance that is robust across different patient cohorts and independent of major clinical confounders. Our highly generalizable, single-lead DiaCardia model offers a promising solution for scalable prediabetes screening via wearable devices, potentially enabling early, home-based detection and transforming diabetes prevention strategies.

Keywords:  Artificial intelligence; Electrocardiogram; Machine learning; Prediabetes

DOI:  https://doi.org/10.1186/s12933-025-02982-4
J Clin Med. 2025 Oct 27. pii: 7600. [Epub ahead of print]14(21):

Correction: Eckart, A.C.; Sharma Ghimire, P. Exploring Predictors of Type 2 Diabetes Within Animal-Sourced and Plant-Based Dietary Patterns with the XGBoost Machine Learning Classifier: NHANES 2013-2016. J. Clin. Med. 2025, 14, 458.

Adam C Eckart, Pragya Sharma Ghimire.

In the original publication [...].

DOI: https://doi.org/10.3390/jcm14217600
BMC Med Inform Decis Mak. 2025 Nov 11. 25(1): 420

Explainable machine learning for differential diagnosis of diabetic foot infection and osteomyelitis: a two-center study and clinically applicable web calculator using routine blood biomarkers.

Parhat Yasin, Shiming Dong, Zubaidanmu Aizezi, Yasen Yimit, Alimujiang Yusufu, Maihemuti Yakufu, Xinghua Song.



Keywords:  Blood biomarkers; Diabetic foot infection; Explainable artificial intelligence (XAI); Machine learning; Osteomyelitis

DOI:  https://doi.org/10.1186/s12911-025-03236-9
Int J Biol Macromol. 2025 Nov 12. pii: S0141-8130(25)09299-2. [Epub ahead of print] 148742

Recent advances in glycated hemoglobin test methods: From lab to point of care testing devices.

Praveen Lakhera, Vikas Chaudhary, Preeti Kush, Parveen Kumar, Yash Ughade, Labi Agrawal, Gautam Patel, Kalim Deshmukh.

  Globally, the threat of diabetes mellitus causes health issues and economic burdens on families. Glycated hemoglobin (HbA1c) is an internationally recommended and reliable gold-standard marker to assess the presence and severity of diabetes. It can be measured using both lab-based standard tests and point-of-care testing (POCT) devices. This review explores published literature from 2018 to July 2025 across Scopus, PubMed Central, Google Scholar, Science Direct, and PubMed, using various keywords such as HbA1c detection, diabetes, POCT devices, artificial intelligence (AI), and biosensors. Some sources, including letters to editors, encyclopedias, conference materials, abstracts, and proceedings, were excluded. It covers the history and standardization of HbA1c, as well as recent advances in testing techniques, including standard laboratory methods, various biosensors (electrochemical, optical, electrochemiluminescent, mass-based, and colorimetric), and cutting-edge approaches like colorimetric, fluorescent assays, and chip-based techniques. Additionally, AI-based methods (deep learning and machine learning) are discussed for predicting HbA1c levels. The review highlights technological developments and concludes with a comparative evaluation of publicly available POCT devices. It also details the process flow from ideation to lab testing, approval, and recognition by medical agencies worldwide. Furthermore, this work can serve as a useful resource for understanding different technology readiness levels. Based on this study, POCTs are increasingly essential, but a solid understanding of detection methods is necessary for working in this field. Moreover, integrating mobile apps with deep machine learning algorithms and AI, microfluidics/lab-on-chip systems, various methods, wearable sensors, and the Internet of Wearable Things (IoWT) can enhance analytical performance and automation.

Keywords:  Artificial intelligence; Biosensors; Diabetes; HbA1c; POCT devices

DOI:  https://doi.org/10.1016/j.ijbiomac.2025.148742
Sensors (Basel). 2025 Oct 27. pii: 6607. [Epub ahead of print]25(21):

Machine Learning-Driven E-Nose-Based Diabetes Detection: Sensor Selection and Feature Reduction Study.

Yavuz Selim Taspinar.

  Diabetes is a major global health problem, with a rapidly increasing prevalence and long-term health complications in both developed and developing countries. If not diagnosed early, it can lead to cardiovascular diseases, kidney failure, vision loss, and nervous system disorders. This study aimed to classify individuals with diabetes or healthy individuals using e-nose sensor data obtained from breath samples taken from 1000 individuals. Six sensor features and one class feature were used in the analysis. Machine learning methods included Artificial Neural Networks (ANN), Decision Trees (DT), Gradient Boosting (GB), Naive Bayes (NB), and AdaBoost (AB). ANOVA and Information Gain analyses were conducted to determine the effectiveness of the sensor data, and the TGS2610 and TGS2611 sensors were found to be critical for classification. Principal Component Analysis (PCA) reduced data size and saved processing time. Experimental results showed that the ANN model provided the most successful classification, with 100% accuracy. AB and GB achieved 99.8% accuracy, while NB achieved 97.6% accuracy. Dimensionality reduction using PCA optimized training and testing times without loss of accuracy. The study presents a data-driven approach to e-nose-based diabetes detection, demonstrates the comparative performance of the models, and highlights the importance of sensor selection and data size optimization.

Keywords:  electronic nose; feature analysis; feature reduction; machine learning; rank analysis

DOI:  https://doi.org/10.3390/s25216607
J Hypertens. 2025 Nov 07.

Screening of the key single nucleotide polymorphisms in type 2 diabetes mellitus complicated with lower extremity arterial disease by machine learning.

Xinyi Li, Aige Yang, Xiao Liu, Shunjiang Xu, Huimin Zhou.

   OBJECTIVES: Diabetic lower extremity arterial disease (LEAD) is a manifestation of diabetic lower extremity vascular complications. This study aimed to screen the key single nucleotide polymorphism (SNP) gene signature in patients with type 2 diabetes mellitus (T2DM) and LEAD.
METHODS: A total of 147 patients with T2DM complicated by LEAD and 144 patients with T2DM without LEAD were enrolled for transcriptome sequencing. The Plink software was used to preprocess the data. Five machine learning methods were adopted to build the SNP diagnosis models. The receiver operating characteristic (ROC) curve was used to quantify the predicted probabilities of the model. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed using the cluster Profiler package. Finally, regression statistical analysis was used to correlate the key SNPs with clinical information and biochemical indicators.
RESULTS: A total of 24 SNPs were retained and 10 SNPs were risk allele genes. Nine SNPs (rs7412, rs1800629, rs699947, rs3918242, rs668, rs1800470, rs1800449, rs1800469, and rs1024611) were identified as the key SNPs sites. GO and KEGG pathway analyses revealed that these genes are mainly enriched in fluid shear stress and atherosclerosis. Finally, rs1800449 was associated with low-density lipoprotein cholesterol (LDL-C). With high density lipoprotein cholesterol (HDL-C), related site was rs1024611. The sites associated with total cholesterol (CHOL) were rs1800449 and rs7412.The site associated with apolipoprotein B (APOB) and apolipoprotein A1 (APOA1) were rs1800470 and rs1800469.
CONCLUSION: This study authenticated nine SNPs for the diagnosis of T2DM patients with LEAD, which will be of great significance in the development of diagnostic molecular biomarkers for T2DM patients.

Keywords:  APOA1; APOB; CC; CHOL; Extreme Gradient Boosting; GO; GWAS; Gene Ontology; HDL-C; KEGG; Kyoto Encyclopedia of Genes and Genomes; LDL-C; LEAD; LR; NBM; RF; ROC; Random Forest model; SNPs; SVM; Support Vector Machine; T2DM; XGBoost; apolipoprotein A1; apolipoprotein B; cellular components; cholesterol; diabetic lower extremity arterial disease; diagnosis; genome-wide association studies; high-density lipoprotein cholesterol; logistic regression; low-density lipoprotein cholesterol; machine learning; naive Bayesian model; peripheral vascular diseases; receiver operating characteristic; single nucleotide polymorphisms; single-nucleotide polymorphisms; type 2 diabetes mellitus

DOI:  https://doi.org/10.1097/HJH.0000000000004164
J Imaging Inform Med. 2025 Nov 10.

Segmentation and Clustering Analysis of Scanned Foot Models for Intelligent Diabetic Foot Ulcer Offloading Boots.

Yaser Sabzehmeidani, Andrei-Alexandru Popa.

  Diabetic foot ulcer (DFU) is a common and severe complication of diabetes that leads to amputation if not effectively managed. Intelligent offloading or rehabilitation devices or boots are used to manage or observe the ulcer and the process of treatment which requires the segmentation and clustering analysis of 3D scanned foot models. This study examines the effectiveness of two widely used clustering algorithms, K-Means and Gaussian Mixture Models (GMM), in segmenting scanned foot models to use in rehabilitation boots. The performance of K-Means and GMM was compared across 98 foot models. GMM consistently achieved higher silhouette scores (0.58 vs. 0.42 for K = 5), lower Davies-Bouldin scores (0.47 vs. 0.54 for K = 5), and more stable clustering across anatomical sections, despite requiring almost 20% more computation time. These results highlight GMM's superior ability to capture the complex nonlinear structures of diabetic feet, with implications for more precise and personalized offloading boot design. Precise segmentation of scanned foot models is a crucial step in various anatomical and medical applications, such as the design of custom orthotic devices, rehabilitation offloading boots, and the analysis of foot biomechanics-particularly useful within DFU management.

Keywords:  3D foot scanning; Clustering algorithms; Diabetic foot ulcer (DFU); Rehabilitation boot; Segmentation

DOI:  https://doi.org/10.1007/s10278-025-01746-6
J Diabetes Sci Technol. 2025 Nov 10. 19322968251388107

Exploring Relationships Between Maternal Characteristics, Continuous Glucose Monitoring Data, and Neonatal Hypoglycemia in Gestational Diabetes Pregnancies Using Probabilistic Modeling.

Giacomo Cappon, Marco Catanuso, Erica Tavazzi, Karen Elkind-Hirsch, Andrea Facchinetti.

   BACKGROUND: Gestational diabetes mellitus (GDM) is a frequent metabolic complication during pregnancy that significantly impacts both maternal and neonatal health outcomes regularly resulting in NH. Exploring the interactions between maternal characteristics, neonatal outcomes, and data collected from wearable technologies, such as continuous glucose monitoring (CGM) could potentially enable the development of predictive models and support personalized care.
METHODS: This study employed probabilistic modeling, using Bayesian networks (BNs), to analyze data from the STEADY SUGAR clinical trial (N = 118 women with GDM) with the aim of discovering interactions between maternal characteristics, CGM-derived features calculated in the 90 days preceding delivery, and neonatal outcomes, particularly NH. The final BN returns a graph and conditional probability tables between inputs and outputs, whose statistical relevance has been quantified via odds ratios (ORs).
RESULTS: Direct associations were identified between NH and maternal hypertension (OR: 2.13 [1.02, 4.46]), family history for diabetes (OR: 1.43 [0.57, 3.57]), and elevated maternal body mass index (BMI) (OR: 3.59 [1.42, 9.08] comparing lower vs higher BMI categories). Cesarean delivery also influenced NH risk (OR: 2.05 [0.98, 4.28]). Indirect associations involving medication regimens and delivery type were significant. Ethnic disparities emerged, notably higher hyperglycemia among Afro-American patients (OR: 2.91 [1.19, 7.11]), highlighting ethnicity-related variations in glycemic control. Notably, CGM-derived metrics were associated with multiple neonatal outcomes.
CONCLUSIONS: Bayesian network allowed to explore the complex interactions between variables in pregnancies affected by GDM. This framework will be extended with wider data sets to provide valuable insights for clinical decision-making able to mitigate maternal and neonatal risks.

Keywords:  Bayesian networks; continuous glucose monitoring; gestational diabetes mellitus; neonatal outcomes

DOI:  https://doi.org/10.1177/19322968251388107