bims-aukdir 2026-07-26 papers

bims-aukdir

Biomed News

on Automated knowledge discovery in diabetes research

Issue of 2026–07–26
twenty-one papers selected by
Mott Given

An Attention-Based Transformer Algorithm Integrated with YOLO Framework for Lesion Localization and Detection Diabetic Retinopathy.
Multimodel Interpretable Machine Learning for Osteoporosis Prediction in Diabetes Using a Large Critical Care Electronic Health Record Database.
A bibliometric analysis of global trends in AI-driven digital health technologies for diabetes management.
Enhancing the performance of diabetic retinopathy grading models using synthetic images.
Ultra-widefield color fundus images and artificial intelligence for diagnosis of diabetic retinopathy: A systematic review and meta-analysis.
Fusion of ConvNeXt-Tiny and Swin-Tiny backbones: a comparative analysis for diabetic retinopathy classification.
GENERALIZABILITY OF DEEP LEARNING MODELS FOR REFERABLE DIABETIC RETINOPATHY DETECTION: A CROSS-POPULATION STUDY.
A systematic review on deep learning techniques for diabetic retinopathy classification in retinal fundus images.
A lightweight DeepME model based on improved YOLOv11 architecture for macular edema detection and treatment monitoring.
Prediction models for progression from prediabetes to diabetes: a systematic review and meta-analysis.
Enhanced prediction of coronary heart disease risk in diabetic patients via Machine learning incorporating multiple inflammatory and metabolic indices: A study with Dual-Cohort validation.
Noninvasive diabetes mellitus classification based on HbA1c using the mel-frequency cepstral coefficients from the photoplethysmographic signal.
Classifying Diabetic and Healthy β-Cells in Type 2 Diabetes Using Machine Learning on Single-Cell RNA Sequencing Data.
Bridging Audience Segmentation and Message Intervention: AI-Generated HBM-Based Messages for Diabetes Prevention.
Machine learning-driven risk prediction for post-hospitalization diabetes case management: Integrating clinical and social determinants of health.
ANGPTL4 in gestational diabetes: a diagnostic biomarker linked to placental senescence.
Development and temporal validation of an interpretable XGBoost model for predicting cardiovascular death in patients with diabetic foot.
Exploring key targets for psychodiabetological interventions in adults with type 1 diabetes mellitus: an exploratory network and intervention modeling study using polish MyDiaMate data.
Retraction: Application of IRSA-BP neural network in diagnosing diabetes.
Evaluation of the accuracy and consistency of DeepSeek and ChatGPT in addressing type 1 diabetes mellitus-related queries in adults.
SM-GAT: a safety-aware multi-task graph attention network for multi-target anti-diabetic lead discovery from natural products.

Photodiagnosis Photodyn Ther. 2026 Jul 20. pii: S1572-1000(26)00248-6. [Epub ahead of print] 105581

An Attention-Based Transformer Algorithm Integrated with YOLO Framework for Lesion Localization and Detection Diabetic Retinopathy.

Garidepalli Revathi, Shanker Chandre, Rajchandar K.

  Diabetic Retinopathy (DR) is one of the major causes of vision loss among diabetic patients across the globe, and it is a major challenge to the public health. It is important to detect the retinal abnormalities early to avoid the progression of the disease and permanent loss of vision. In this regard, the framework proposed provides an effective and automated method of DR screening with retinal fundus images. The model combines the merits of attention-based and transformer-based models and the You Only Look Once (YOLO) detection model to improve the level of feature representation and detection quality. It is intended to do multi-task learning such as lesion detection, accurate segmentation, and severity of DR stages. The framework is tested using well known publicly available databases like APTOS, EyePACS, and Messidor which are robust and reliable. The experimental findings suggest that Attention Transformer-YOLO for Diabetic Retinopathy Detection (AT-YOLO-DR) is more accurate especially with respect to the detection of small and subtle lesions, and it is also characterized by better generalization than the current deep learning models. The system offers a scalable, reliable and practical solution to real life clinical screening applications.

Keywords:  Automatic learning rate Optimization; Deep learning; Eye fundus images; Image augmentation

DOI:  https://doi.org/10.1016/j.pdpdt.2026.105581
J Vis Exp. 2026 Jun 30.

Multimodel Interpretable Machine Learning for Osteoporosis Prediction in Diabetes Using a Large Critical Care Electronic Health Record Database.

Dongdong Cheng, Wei Zhang, Jingyu Chen, Xinhui Zhu.

Diabetes mellitus is associated with increased skeletal fragility and elevated fracture risk; however, early identification of osteoporosis (OP) in diabetic populations remains challenging. This study aimed to develop and compare multiple machine-learning (ML) models for predicting OP among patients with diabetes, while evaluating their clinical utility and enhancing model transparency using SHapley Additive exPlanations (SHAP). Using routinely collected demographic and laboratory data from the MIMIC-IV database, multiple ML algorithms, including tree ensembles, gradient boosting, and linear baselines, were trained and validated. Model performance was comprehensively evaluated using discrimination metrics, including the area under the receiver operating characteristic curve (AUC) and precision-recall analysis, as well as probability calibration curves and decision curve analysis (DCA) to assess net clinical benefit. The best-performing model was further interpreted using SHAP to quantify and visualize feature contributions at both global and individual levels. Among all evaluated models, ensemble tree-based methods showed improved performance. The ExtraTrees classifier achieved the highest validation performance (AUC = 0.862; average precision = 0.866). Furthermore, the selected model exhibited excellent probability calibration (Brier score = 0.154) and demonstrated substantial net clinical benefit across a wide range of risk thresholds in the DCA. SHAP analysis identified gender and age as the most influential predictors, followed by routine hematologic and metabolic laboratory indicators. Local explanations provided clinically interpretable insights into individual predictions. An interpretable ensemble tree-based model effectively predicts OP risk in patients with diabetes using routinely available clinical variables. The integration of SHAP improves clinical transparency, and strong calibration coupled with positive net clinical benefit supports its potential implementation as a reliable decision-support tool for early OP risk stratification in diabetic populations.

DOI: https://doi.org/10.3791/71386
Medicine (Baltimore). 2026 Jul 24. 105(30): e49815

A bibliometric analysis of global trends in AI-driven digital health technologies for diabetes management.

Jingwen Song, Wenli Tao, Di Zhou, Xuanbin Li, Sijing Peng, Chunyan Li, Juan Yuan.

   BACKGROUND: Digital health technologies are increasingly applied in diabetes care, enabling continuous monitoring, personalized support and remote interventions. Meanwhile, artificial intelligence (AI) is enhancing the precision and effectiveness of these tools. This study aims to map global research trends and thematic developments in AI-driven digital health technologies for diabetes management and to explore their future directions.
METHODS: We collected data from the Web of Science Core Collection, including articles and reviews published up to July 12, 2025, using CiteSpace, VOSviewer, and Microsoft Excel to analyze countries/regions, institutions, journals, references, authors, and keywords.
RESULTS: A total of 673 publications were included in the analysis. Global publications on AI-driven digital health technologies for diabetes increased steadily, with the USA leading in output. The University of London ranked as the most productive institution. Sensors and diabetes care were the most frequently published and cited journals in this field. Herrero P was among the most prolific authors. The most cited article was "Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs." "diabetes" was the most frequently occurring keyword. Keyword cluster analysis identified 3 primary research hotspots: AI-enabled monitoring, digital health interventions, and AI-based diabetic retinopathy screening.
CONCLUSIONS: This study summarizes the evolution of AI-driven digital health technologies in diabetes care. Although challenges remain in data security, standardization and validation, these technologies hold increasing potential for accurate diagnosis, real-time monitoring and personalized care.

Keywords:  artificial intelligence; bibliometric; development trends; diabetes management; digital health technology; visual analysis

DOI:  https://doi.org/10.1097/MD.0000000000049815
Retina. 2026 Jul 21.

Enhancing the performance of diabetic retinopathy grading models using synthetic images.

Jiashi Zhao, Yiting Zhang, Yu Miao, Weili Shi, Jianhua Liu, Zhengang Jiang.

   PURPOSE: To determine whether strictly curated synthetic images can improve grading performance in diabetic retinopathy (DR) diagnosis within limited data regimes, and to compare the utility of representative Generative Adversarial Network (GAN) and diffusion architectures.
METHOD: StyleGAN3 and Medfusion models were pre-trained on an auxiliary dataset. To prevent mode collapse and severity downshift, synthetic images were curated via a domain-matched Oracle network using Latent Space Rejection Sampling and Class-Conditioned SDEdit Escalation. Efficacy was evaluated on an isolated downstream DR classification task (IDRiD dataset) across three architectures (ConvNeXt, ResNet50, VGG16) using five random seeds.
RESULTS: Curated StyleGAN3 dataset achieved better Fréchet Inception Distance (24.25) and Padded SSIM (0.9037) compared to Medfusion (58.70 and 0.8852), though Medfusion yielded a higher Inception Score (2.35). In the isolated downstream classification task, baseline ConvNeXt models trained solely on real IDRiD data achieved a Quadratic Weighted Kappa (QWK) of 0.6350 ± 0.0135. Augmentation with curated StyleGAN3 improved the QWK to 0.7185 ± 0.0309, while Medfusion augmentation achieved 0.7695 ± 0.0254 (an absolute improvement of 0.1345). Crucially, Medfusion augmentation successfully doubled the sensitivity (Recall) for Proliferative DR from 0.2615 ± 0.1595 to 0.5385 ± 0.0942.
CONCLUSION: Data augmentation using strictly curated synthetic images successfully addresses data imbalance and scarcity, significantly improving the ordinal accuracy and minority-class sensitivity of DR diagnosis models. Furthermore, standard global image quality metrics like FID may not fully capture localized pathological utility; diffusion models with higher FID scores ultimately provided superior structural diversity and downstream clinical performance.

Keywords:  Generative models; deep learning; diabetic retinopathy; synthetic images

DOI:  https://doi.org/10.1097/IAE.0000000000004919
Retina. 2026 Jul 24.

Ultra-widefield color fundus images and artificial intelligence for diagnosis of diabetic retinopathy: A systematic review and meta-analysis.

Kunihiko Hirosawa, Takenori Inomata, Ken Nagino, Jaemyoung Sung, Akie Midorikawa-Inomata, Atsuko Eguchi, Kenta Fujio, Hiroyuki Kobayashi, Shintaro Nakao.

   PURPOSE: Ultra-widefield (UWF) fundus cameras capture a larger retinal area without pupil dilation. We summarized evidence and diagnostic performance of artificial intelligence (AI)-driven diabetic retinopathy (DR) assessments using UWF images (UWFIs).
METHODS: We searched PubMed, Scopus, the Cochrane Library, and Web of Science to February 9, 2025, for studies evaluating DR using UWFIs and AI analyses. We followed the PRISMA guidelines and assessed study quality using the Joanna Briggs Institute Critical Appraisal Checklist for diagnostic accuracy. Forest plots and summary receiver operating characteristic curves estimated sensitivity and specificity of AI-driven DR screening using UWFI.
RESULTS: Of 527 records identified, 17 studies were included in the systematic review and four in the meta-analysis. All studies used Optos software. Among the included studies, two evaluated DR discrimination from multiple retinal disorders, six addressed DR stage classification based on guideline scales, six evaluated DR screening, and one each focused on referable retinal pathology detection, DR progression prediction, and detection of clinical DR features. The summary sensitivity, specificity, and area under receiver operating characteristic curve for AI-driven DR screening using UWFI were 85.0%, 72.5%, and 0.870 (0.838-0.897), respectively. UWFI-based AI screening showed acceptable sensitivity but limited specificity.
CONCLUSION: AI-driven DR screening using UWFIs may achieve acceptable sensitivity but with limited specificity; however, these findings should be interpreted with caution, given the limited number of available studies. Further studies using larger, diverse datasets with rigorous external validation across UWF platforms are needed to improve generalizability and clarify the clinical role of UWFI-based AI.

Keywords:  artificial intelligence; color fundus photography; deep learning; diabetic retinopathy; diagnostic accuracy; meta-analysis; systematic review; telemedicine; ultra-widefield color fundus images; ultra-widefield imaging

DOI:  https://doi.org/10.1097/IAE.0000000000004939
Front Artif Intell. 2026 ;9 1812599

Fusion of ConvNeXt-Tiny and Swin-Tiny backbones: a comparative analysis for diabetic retinopathy classification.

J Paranthaman, Sathya Pichandi, Aparna Mohanty.

  Diabetic retinopathy (DR) is a leading cause of preventable blindness, which has motivated the development of reliable automated grading systems on retinal fundus images. In this study, we perform a controlled comparative evaluation of ConvNeXt-Tiny, Swin-Tiny and their feature fusion for DR classification using the Asia Pacific Tele-Ophthalmology Society (APTOS) 2019 dataset. All models were initialized with weights pre-trained on ImageNet-1K and evaluated with two transfer learning strategies: direct fine-tuning on APTOS 2019, and EyePACS-based domain adaptation with task-specific fine-tuning. Systematic ablation experiments were carried out to evaluate the contribution of Contrast Limited Adaptive Histogram Equalization (CLAHE) preprocessing and channel-spatial attention modules (CSAM). We carried out experiments on the APTOS 2019 dataset with fixed train, validation and test splits and evaluated model stability across three runs with different random seeds by reporting mean ± standard deviation of performance metrics, while performance varied widely across architectures and training settings. After domain adaptation, the fusion-based models achieved more balanced results, while the standalone Swin-Tiny showed weaker adaptation to the retinal imaging domain, and was less sensitive to subtle lesion patterns under the EyePACS-based transfer learning. Adding CLAHE preprocessing and CSAM integration did not consistently improve class-balanced metrics. The best fusion configuration achieved a mean test accuracy of 88.34% ± 1.09 and a macro F1-score of 0.7376 ± 0.0183 on the APTOS 2019 dataset across repeated runs. These results suggest that domain-specific adaptation and architectural complementarity are more beneficial in boosting DR classification performance than auxiliary preprocessing or attention enhancement. The study also emphasizes the importance of controlled comparative evaluation, stability analysis, and configuration-specific evaluation in the research of medical image classification.

Keywords:  APTOS 2019 dataset; Contrast Limited Adaptive Histogram Equalization (CLAHE); ConvNeXt-Tiny; EyePACS Combined dataset; ImageNet-1k dataset; Swin-Tiny transformer; channel-spatial attention module (CSAM); diabetic retinopathy

DOI:  https://doi.org/10.3389/frai.2026.1812599
Retina. 2026 Jul 21.

GENERALIZABILITY OF DEEP LEARNING MODELS FOR REFERABLE DIABETIC RETINOPATHY DETECTION: A CROSS-POPULATION STUDY.

Ceren Durmaz Engin, Mustafa Alper Selver, Seher Köksaldı, Mustafa Kayabaşı.

   PURPOSE: To evaluate cross-population performance of deep learning models for referable diabetic retinopathy (DR) detection and assess whether inclusion of local data improves robustness in a Turkish external validation setting.
METHODS: Global datasets (DDR, IDRiD, Messidor) and a Turkish real-world clinical dataset were used. Images were re-graded by three ophthalmologists and binarized as referable/non-referable DR. Five architectures were evaluated: EfficientNet-B0, EfficientNet-V2, ResNet50, ConvNeXt-V2-Tiny, and Swin V2 CR Small. Models were tested under three settings: Global Train-Global Test, Global Train-Local Test, and Mixed Train-Mixed Test. Performance metrics included accuracy, sensitivity, specificity, NPV, ROC-AUC, and PR-AUC; Grad-CAM was used for explainability.
RESULTS: The global training set included 9,881 images; two external test sets (2,042 images each) were used. Inter-grader agreement was high (ICC[3,1]=0.84; ICC[3,k]=0.96). Swin V2 CR Small achieved the highest performance in Global Train-Global Test (accuracy 0.938; ROC-AUC 0.985). Performance declined in Global Train-Local Test, where ConvNeXt-V2-Tiny performed best (accuracy 0.823; sensitivity 0.940; ROC-AUC 0.913). In Mixed Train-Mixed Test, ConvNeXt-V2-Tiny remained the most balanced model (accuracy 0.823; sensitivity 0.939; NPV 0.948; ROC-AUC 0.875).
CONCLUSION: Models performed strongly under matched conditions but degraded across populations. ConvNeXt-V2-Tiny showed the most stable screening performance, supporting the need for local validation before deployment.

Keywords:  deep learning; diabetic retinopathy; domain shift; external validation; fundus photography

DOI:  https://doi.org/10.1097/IAE.0000000000004924
iScience. 2026 Jul 17. 29(7): 116565

A systematic review on deep learning techniques for diabetic retinopathy classification in retinal fundus images.

José Araque-Gallardo, Eugenia Arrieta-Rodríguez, Lenis Rueda-Gómez, Oscar Teherán-Forero, Emiro De-La-Hoz-Franco, Margarita Gamarra, Javier Sierra-Carrillo, José Escorcia-Gutierrez.

  Diabetic retinopathy (DR) is the leading cause of preventable blindness worldwide, particularly in low- and middle-income countries. Although expert analysis of color fundus images (CFIs) enables reliable classification, this process is time-consuming. In recent years, deep learning (DL) techniques have demonstrated remarkable performance in analyzing CFI for the detection (binary classification) and multi-class classification of DR. To synthesize recent advances in this field, this study presents a systematic review, analyzing 146 peer-reviewed studies published between 2019 and 2025. This review examines lesion detection and segmentation, as well as DR detection and classification in CFI using two approaches: direct full-image and lesion-based classification. In addition, a brief scientometric analysis was performed to identify publication trends, leading journals, and countries contributing to this domain. The insights provided by this review aim to support researchers in selecting effective strategies and advancing the development of DL-based systems for the detection and classification of DR.

Keywords:  Health sciences; Medical specialty; Medicine; Ophthalmology

DOI:  https://doi.org/10.1016/j.isci.2026.116565
Front Endocrinol (Lausanne). 2026 ;17 1864912

A lightweight DeepME model based on improved YOLOv11 architecture for macular edema detection and treatment monitoring.

Xue Bai, Ming Yi, Tianye Chen, Naijing Feng, Quan Shi, Tingxue Li, Rui Hua.

   Objective: We developed and validated an improved YOLOv11-based deep learning algorithm for accurate macular edema detection in optical coherence tomography (OCT) images, and built DeepME-a lightweight system for diagnosis and treatment recommendations.
Methods: We compiled a comprehensive dataset combining hospital clinical data and public OCT resources, covering macular edema and other retinal diseases. External validation used an anti-vascular endothelial growth factor (anti-VEGF) cohort of 336 eyes from 300 patients with diabetic retinopathy or retinal vein occlusion. The improved YOLOv11n integrated the Convolutional Block Attention Module (CBAM) to enhance feature extraction. DeepME combined this detector with an optimized DeepSeek model, current clinical guidelines, and expert knowledge.
Results: DeepME achieved better performance over standard YOLOv11: accuracy 0.980, specificity 0.990, sensitivity 0.970, precision 0.990, F1-score 0.980, and AUC 0.9993. Grad-CAM visualizations confirmed precise localization of cystoid macular edema within anatomically correct retinal layers. In the anti-VEGF cohort, central foveal thickness decreased significantly at one-month follow-up (p < 0.001), and DeepME showed substantial agreement with manual grading (p < 0.001), enabling rapid, accurate diagnosis and treatment guidance.
Conclusion: This study introduces DeepME, a novel clinical decision support system that integrates an improved YOLOv11 detection architecture for comprehensive macular edema management. DeepME delivers high accuracy in evaluating anti-VEGF treatment response and shows strong potential for real-world clinical decision support.

Keywords:  YOLOv11; anti-vegf; central foveal thickness; deep learning; diabetic macular edema; diabetic retinopathy; optical coherence tomography; swept-source OCT angiography

DOI:  https://doi.org/10.3389/fendo.2026.1864912
Front Endocrinol (Lausanne). 2026 ;17 1888466

Prediction models for progression from prediabetes to diabetes: a systematic review and meta-analysis.

Yanxian Wang, Cuili Wang, Jianxiu Wang.

   Purpose: Prediabetes increases the risk of type 2 diabetes mellitus (T2DM). Accurate prediction is crucial for early prevention, but evidence on prediction models has not been comprehensively synthesized. This study systematically evaluated the accuracy of such models in predicting prediabetes-to-T2DM progression.
Methods: Databases including Cochrane Library, Embase, PubMed, and Web of Science were searched up to June 2, 2025. PROBAST was applied to evaluate the risk of bias. STATA 15.0 was employed to analyze the pooled concordance index (C-index) with 95% CI, to conduct subgroup and sensitivity analyses, and to assess publication bias.
Results: Sixteen studies were included, covering 1,368,130 prediabetic individuals with 187,225 progressing to T2DM. Pooled incidence was 42.3‰ (95% CI: 27.2‰-60.4‰). Pooled C-indices of the training and validation sets were 0.76 (0.71-0.80) and 0.84 (0.82-0.86), respectively. Logistic regression and random forest yielded C-indices of 0.81 and 0.86, respectively.
Conclusions: Prediction models show promising accuracy for predicting progression from prediabetes to T2DM, although the evidence remains limited, particularly due to the lack of external validation. Future research should strengthen model development, external validation, and reporting quality to improve the robustness and clinical applicability of prediction models for the progression of prediabetes.
Systematic Review Registration: https://www.crd.york.ac.uk/PROSPERO/, identifier CRD420251104222.

Keywords:  machine learning; meta-analysis; prediabetes; prediction model; systematic review; type 2 diabetes mellitus

DOI:  https://doi.org/10.3389/fendo.2026.1888466
Int J Cardiol Heart Vasc. 2026 Aug;65 101933

Enhanced prediction of coronary heart disease risk in diabetic patients via Machine learning incorporating multiple inflammatory and metabolic indices: A study with Dual-Cohort validation.

Pinpeng Xie, Xiangzhen Ran, Weichun Zhu, Yunfang Xie, Youran Dong, Zhiyuan Wang, Chungang Zhai, Lei Qiao, Jianmin Yang, Wenqiang Chen.

   Background: Coronary heart disease (CHD) remains a leading cause of mortality worldwide, with individuals with diabetes mellitus (DM) facing markedly elevated risk due to complex inflammatory and metabolic disturbances. Emerging composite inflammatory and metabolic indices have demonstrated promise in enhancing cardiovascular risk stratification, yet research quantifying and comparing their respective predictive performance and assessing their relative contributions remains limited. This study aimed to develop a clinically applicable model for early CHD risk prediction in diabetic patients using novel composite inflammatory and metabolic indices.
Methods: This study utilized data from 3379 diabetic participants in the NHANES 1999-2018 survey cycles. Additionally, an independent external set of 902 patients from Qilu Hospital of Shandong University served as the validation cohort. Novel inflammatory and metabolic indices were calculated. Feature selection was performed via LASSO regression, the univariate logistic regression, and the Boruta algorithm. Nine machine learning (ML) models were developed using selected predictors. Model performance was evaluated using the receiver operating characteristic (ROC) curve, calibration curves, and decision curve analysis. SHapley Additive exPlanations (SHAP) values were used to interpret model predictions and identify key contributing features.
Results: Five composite indices UHR (uric acid-to-HDL ratio), MHR (monocyte-to-HDL ratio), NPAR (neutrophil-to-albumin ratio), NLR (neutrophil-to-lymphocyte ratio), and AIP (atherogenic index of plasma) were identified as robust predictors of CHD. The random forest (RF) algorithm achieved the highest performance, with an AUC of 0.852 in the internal validation set and 0.713 in the external cohort. Calibration plots, Brier scores, and decision curve analysis further confirmed the RF model's predictive reliability and clinical utility. SHapley Additive exPlanations (SHAP) value analysis revealed that UHR, MHR, NLR, age, and hypertension were the key features driving CHD prediction.
Conclusion: We developed and externally validated an ML model incorporating five composite inflammatory and metabolic indices (UHR, MHR, NLR, NPAR, AIP), which demonstrated promising performance in predicting CHD risk in diabetic patients.

Keywords:  Coronary heart disease; Diabetes mellitus; Dual-Cohort Validation; Inflammatory and metabolic indices; Machine learning

DOI:  https://doi.org/10.1016/j.ijcha.2026.101933
Physiol Meas. 2026 Jul 24.

Noninvasive diabetes mellitus classification based on HbA1c using the mel-frequency cepstral coefficients from the photoplethysmographic signal.

Surendra Kumar Saini, Sakshi Tiwari, Anil Sharma, Amit Mahesh Joshi, Ravi Maddila.

  Diabetes mellitus (DM) is a metabolic condition with improper regulation of blood sugarlevels and is one of the leading global causes of death. Glycated hemoglobin (HbA1c)serves as a crucial indicator for managing diabetes. This study proposes a noninvasiveapproach to classify glycemic status based on HbA1c levels. This work uses novelMel-frequency cepstral coefficient features of finger photoplethysmographic (PPG) signalsand physiological parameters, enabling straightforward detection of DM. A finger PPGdataset (in reflective mode) comprising 180 subjects with diabetes, prediabetes, andnormal HbA1c levels is curated and used to validate the proposed method. The datasetcomprises 93 normal (HbA1c < 5.7%), 57 prediabetic (HbA1c 5.7% - 6.4%), and 30diabetic (HbA1c ≥ 6.5%) individuals. Furthermore, a hybrid feature (HyF) selectionmethod is employed for feature reduction. The HyF Selection-based gradient boostingmodel achieved effective accuracies of 93.89% for binary classification and 91.67% formulticlass classification. The results are compared with the gold standard HbA1c test.Both the binary and multiclass classifications show improved overall performance. Thesefindings indicate that PPG signals are a feasible substitute for noninvasive HbA1cdetection and have potential for wearable HbA1c monitoring.

Keywords:  Diabetes mellitus; HbA1c; MFCC; Machine learning; Photoplethysmography

DOI:  https://doi.org/10.1088/1361-6579/ae9037
Comput Struct Biotechnol J. 2026 ;35(1): 0114

Classifying Diabetic and Healthy β-Cells in Type 2 Diabetes Using Machine Learning on Single-Cell RNA Sequencing Data.

Daniel F O Onah, María De La Luz Lomboy Toledo.

Diabetes is a chronic metabolic disorder characterized by elevated blood glucose levels due to impaired insulin production or function. Two main forms are recognized: type 1 diabetes, which involves the autoimmune destruction of insulin-producing β-cells, and type 2 diabetes (T2D), which arises from insulin resistance and progressive β-cell dysfunction. Understanding the molecular mechanisms underlying these diseases is essential for the development of improved therapeutic strategies, particularly those targeting β-cell dysfunction. To investigate these mechanisms in a controlled and biologically interpretable setting, mouse models have played a central role in diabetes research. Owing to their genetic and physiological similarity to humans, together with the ability to precisely manipulate their genome, mice enable detailed investigation of disease progression and gene function. In particular, mouse models have provided critical insights into β-cell development, cellular heterogeneity, and functional failure under diabetic conditions. Building on these experimental advances, this study applies machine learning methods to single-cell transcriptomic data from mouse pancreatic islets. Specifically, we evaluate 2 supervised approaches identified in the literature, extra trees classifier and partial least squares discriminant analysis, to assess their ability to identify T2D-associated gene expression signatures at a single-cell resolution. Model performance is evaluated using standard classification metrics, with an emphasis on interpretability and biological relevance.

DOI: https://doi.org/10.34133/csbj.0114
J Health Commun. 2026 Jul 23. 1-15

Bridging Audience Segmentation and Message Intervention: AI-Generated HBM-Based Messages for Diabetes Prevention.

Qinxuan Yin, Xin Ma.

  Health communication often faces a segmentation-intervention gap: psychographic segmentation identifies meaningful audience profiles, but these profiles are rarely translated into experimentally tested segment-specific messages. This study proposes and tests a segmentation-to-intervention pipeline that integrates Health Belief Model (HBM) diagnostics with generative artificial intelligence (GAI) for type 2 diabetes prevention. In Study 1 (N = 993), a two-step cluster analysis identified three HBM-based psychographic segments among at-risk adults. In Study 2 (N = 737), a randomized experiment tested AI-generated messages produced by Gemini 2.5 Pro to match each segment's HBM profile. Compared with a general message, AI-generated segment-specific messages increased preventive intentions at the aggregate level and within all three segments, suggesting broad persuasive benefits across psychologically distinct groups. Mediation analysis showed that AI personalization was associated with greater message elaboration, which in turn was linked to cognitive trust, affective trust, and preventive intention. Exploratory path results further suggested a direct association between elaboration and affective trust, extending existing models of human-AI trust. These findings demonstrate the feasibility of a scalable, theory-driven framework for precision health communication.

Keywords:  AI-generated health communication; Audience segmentation; Type 2 diabetes; generative artificial intelligence; health belief model; health promotion; message tailoring

DOI:  https://doi.org/10.1080/10810730.2026.2706638
Int J Med Inform. 2026 Jul 11. pii: S1386-5056(26)00344-8. [Epub ahead of print]220 106604

Machine learning-driven risk prediction for post-hospitalization diabetes case management: Integrating clinical and social determinants of health.

Seung-Yup Lee, Mohammad Saleem, Andrew M Land, Erin W DeLaney, Alison R Garretson, Mahee Patel, Allyson G Hall, Salisa Westrick, Fernando Ovalle, Andrea L Cherrington, Jane C Banaszak-Holl.

   OBJECTIVE: Patients hospitalized for diabetes-related conditions face elevated risks of emergency department (ED) visits post-discharge, driven by both clinical factors and social determinants of health (SDoH). This study aimed to develop and validate predictive models integrating clinical and SDoH data to identify high-risk patients for post-hospitalization diabetes case management.
METHODS: We conducted a retrospective cohort study using electronic health record data from the University of Alabama at Birmingham Medical Center, including 162,063 inpatient encounters (January 2020-June 2024) for training and testing and 16,164 encounters (January-May 2025) for temporal validation. Patients were identified by diabetes-related ICD-10 codes or HbA1c ≥ 6.5 %, reflecting the scope of the institution's diabetes case management program. Predictors included demographics, diabetes-related comorbidities, surgical procedures, laboratory values, medications, and both area-level and individual-level SDoH. Logistic regression, decision trees, and XGBoost models were developed to predict diabetes-related ED visits within 3 months post-hospitalization. Hyperparameters for decision tree and XGBoost models were tuned via 10-fold cross-validation, and calibration was assessed using Brier scores and calibration plots.
RESULTS: Among 162,063 hospitalizations, 6.2 % resulted in a diabetes-related ED visit. XGBoost achieved the best performance (area under the curve [AUC] 0.846, precision 0.420, sensitivity 0.296, specificity 0.972), maintained on temporal validation (AUC 0.842). Key predictors included past ED visit frequency, insulin prescriptions, age, and area-level SDoH indices. Individual-level SDoH factors, including home safety issues and work disability, also contributed to prediction. Targeting the top 20 % of predicted risk captured 64.1 % of all ED visits. Model discrimination was consistent across racial subgroups (AUC range: 0.843-0.851). Calibration was clinically acceptable across datasets.
CONCLUSIONS: Integration of clinical and SDoH data achieved effective prediction of post-hospitalization ED visits. XGBoost provided excellent discrimination with temporal stability. Decision trees offered greater interpretability. A pilot implementation delivering daily risk-stratified patient lists to the diabetes case manager is underway, demonstrating a practical pathway from model development to clinical decision support.

Keywords:  Artificial intelligence; Case management; Diabetes; Emergency department; Machine learning; Social determinants of health

DOI:  https://doi.org/10.1016/j.ijmedinf.2026.106604
Funct Integr Genomics. 2026 Jul 23. pii: 201. [Epub ahead of print]26(1):

ANGPTL4 in gestational diabetes: a diagnostic biomarker linked to placental senescence.

Yu-Feng He, Hui Chen, Chuan-Meng Zhang, Su-Su Yang, Xiao-Long Yang, Li-Shuai Jiang, Fan Wang, Shu-Guang Zhou.

  Current diagnostic methods for Gestational Diabetes Mellitus (GDM) inadequately reflect early placental pathophysiology. Accelerated cellular senescence in trophoblasts has been reported as a pathological feature associated with the GDM placenta. However, the potential of senescence-related genes (SRGs) as candidate biomarkers or molecular indicators for GDM has not been systematically investigated. Placental transcriptomic data from GDM and control samples were obtained from the Gene Expression Omnibus (GEO) database, and differentially expressed senescence-related genes (DE-SRGs) were identified. A classification model was built using an integrative machine learning framework, and RT-qPCR validated model gene expression in placental tissues. In parallel, a prospective nested case-control study (30 GDM cases and 30 controls) was conducted, with peripheral blood collected in the first trimester, second trimester, and at delivery. Plasma Angiopoietin-like 4 (ANGPTL4) levels were measured by ELISA to determine the key gene for downstream investigation. Single-cell RNA sequencing (scRNA-seq) was then used to resolve the cellular expression patterns and cell-cell communication networks of the key gene. (Chinese Clinical Trial Registration No. ChiCTR2400091955; Registration Date: 2024-11-06). The classification model integrating ANGPTL4 and DST demonstrated favorable retrospective discriminatory performance (training set AUC = 0.95; validation set AUC = 0.87). This model served as a discovery tool rather than a clinical predictor. RT-qPCR indicated upregulation of ANGPTL4 in GDM placentas (P < 0.05), ANGPTL4 was significantly elevated in the first trimester of women who later developed GDM (P_adj = 0.038); this difference was no longer significant in the second trimester. Accordingly, ANGPTL4 was prioritized for further study, ANGPTL4 was predominantly expressed in extravillous trophoblasts (EVTs), where its expression positively correlated with cellular senescence. Cell communication analysis predicted potential ligand-receptor interactions between ANGPTL4 and ITGA5/ITGB1/SDC4 in EVTs. DE-SRGs in GDM placentas were highly enriched in lipid metabolism pathways, aligning with ANGPTL4's known function. Integrating machine learning with single-cell transcriptomics, we identified placental senescence-related gene candidates associated with GDM. ANGPTL4 is elevated in first‑trimester plasma and enriched in extravillous trophoblasts, where its expression correlates with lipid metabolism and cellular senescence signatures. These correlative findings generate testable hypotheses linking ANGPTL4‑mediated signaling to placental senescence. ANGPTL4 emerges as a candidate biomarker whose clinical potential warrants further evaluation in prospective cohorts.

Keywords:  ANGPTL4; Gestational diabetes mellitus; Machine learning; Senescence-related markers; Single-cell transcriptomics

DOI:  https://doi.org/10.1007/s10142-026-01978-x
Biomol Biomed. 2026 Jul 17.

Development and temporal validation of an interpretable XGBoost model for predicting cardiovascular death in patients with diabetic foot.

Xiaoling Wan, Ting Shi, Qiao Liu, Ying She, Jing Yu.

Patients with diabetic foot (DF) have a high risk of cardiovascular (CV) death, yet dedicated risk-prediction tools for this population are lacking. We developed and temporally validated an interpretable machine learning (ML) model for predicting CV death in patients with DF. This single-center retrospective cohort study included 2,835 patients admitted between February 2017 and May 2025. The development cohort comprised 2,325 patients, including 748 CV deaths, and was divided into training, internal validation, and held-out test sets; an independent temporal validation cohort included 510 patients, including 220 CV deaths. Nine supervised ML algorithms were compared using the area under the receiver operating characteristic curve (AUC). Extreme gradient boosting (XGBoost) showed the best overall performance. The optimal model, incorporating demographic and diabetes-related characteristics, routine laboratory parameters, and DF-specific features, achieved AUCs of 0.829 (95% confidence interval [CI]: 0.752-0.905) in internal validation, 0.844 (95% CI: 0.806-0.881) in the held-out test set, and 0.828 (95% CI: 0.789-0.868) in temporal validation. The model demonstrated good calibration and favorable net benefit on decision curve analysis. SHapley Additive exPlanations (SHAP) identified age, serum creatinine, glycated hemoglobin, triglycerides, and body mass index as the most influential predictors of increased model-predicted risk. This interpretable XGBoost model may support early identification and individualized risk stratification of patients with DF at high risk of CV death; however, prospective multicenter validation is required before clinical implementation.

DOI: https://doi.org/10.17305/bb.2026.14337
Diabetes Res Clin Pract. 2026 Jul 21. pii: S0168-8227(26)00375-X. [Epub ahead of print] 113455

Exploring key targets for psychodiabetological interventions in adults with type 1 diabetes mellitus: an exploratory network and intervention modeling study using polish MyDiaMate data.

Katarzyna Cyranka, Tomasz Klupa, Bartłomiej Matejko, Andrzej Juryk, Katarzyna Klasa, Frank Snoek, Dominika Dudek, Maciej T Małecki, Jerzy A Sobański.

   AIMS: To examine the symptom-level structure of diabetes-related psychological burden in adults with Type 1 diabetes mellitus using selected cross-sectional data from Polish MyDiaMate users, and to identify model-implied intervention targets.
METHODS: We analyzed anonymized app-derived records from 101 active users with complete data on 15 binary reflective items using bootstrapped undirected network modeling, Bayesian directed modeling, and intervention simulation.
RESULTS: The most frequently endorsed items concerned future complications (85.1%), guilt or worry when diabetes was out of control (79.2%), diabetes burnout (72.3%), and feeling that diabetes controls one's life (72.3%). Network modeling revealed a densely connected psychological-burden domain and a more compact fatigue-somatic domain. Directed modeling suggested a pathway from diabetes-related anxiety to perceived loss of control, burnout, and guilt, with subsequent links to reduced perceived coping with complications and worry about future complications. Anxiety showed the largest modeled downstream influence, whereas perceived loss of agency appeared to be a more local intervention target.
CONCLUSIONS: Diabetes-related anxiety and perceived loss of agency emerged as potentially relevant, model-implied intervention targets. These exploratory findings require validation in longitudinal and interventional studies.

Keywords:  Bayesian networks; MyDiaMate; Type 1 diabetes mellitus; diabetes distress; network analysis; psychodiabetology

DOI:  https://doi.org/10.1016/j.diabres.2026.113455
PLoS One. 2026 ;21(7): e0354316

Retraction: Application of IRSA-BP neural network in diagnosing diabetes.

PLOS One Editors

DOI: https://doi.org/10.1371/journal.pone.0354316
Sci Rep. 2026 Jul 21.

Evaluation of the accuracy and consistency of DeepSeek and ChatGPT in addressing type 1 diabetes mellitus-related queries in adults.

Junyun Feng, Xiao Fei, Tingyi Qian, Huilan Luo, Suying Wang, Yihuan Cai, Yaobin Ouyang, Foqiang Liao, Lingyan Zhu.

  Adult type 1 diabetes mellitus (T1DM) involves complex diagnosis, treatment, and long-term self-management, creating a need for accurate and accessible health education. Large language models (LLMs) are increasingly used for medical information seeking, yet their accuracy and consistency in adult T1DM-related queries remain insufficiently evaluated. A guideline-based comparative evaluation assessed DeepSeek-V3.2 and ChatGPT-5.0 using 22 English-language prompts derived from the 2021 ADA/EASD consensus report, covering basic knowledge, diagnosis and differential diagnosis, treatment, and complications. The prompts were submitted to both models twice, two weeks apart. Responses were independently evaluated by two blinded endocrinology specialists using a predefined four-point scoring rubric, with disagreements adjudicated by a third senior endocrinologist. Short-term consistency was assessed by expert judgment and TF-IDF cosine similarity. Inter-rater agreement was good (Cohen's κ = 0.71). Expert-judged consistency was 95.45% (21/22) for both models; TF-IDF cosine similarity was 0.52 ± 0.09 for DeepSeek and 0.54 ± 0.09 for ChatGPT. Overall accuracy scores were 3.59 ± 0.59 and 3.77 ± 0.43, respectively, with no statistically significant difference (p = 0.102). Comprehensive ratings accounted for 63.64% and 77.27%, respectively, and mixed correct and incorrect or outdated information accounted for 4.55% and 0.00%. Both models may support adult T1DM-related health education, but outputs should be interpreted as supplementary educational material under professional guidance rather than as diagnostic or therapeutic advice.

Keywords:  ChatGPT; DSMES; DeepSeek; Large language models; Type 1 diabetes mellitus

DOI:  https://doi.org/10.1038/s41598-026-62671-4
Front Mol Biosci. 2026 ;13 1854162

SM-GAT: a safety-aware multi-task graph attention network for multi-target anti-diabetic lead discovery from natural products.

Jianxin Zhang, Hongyi Liu, Shengnan Guo.

  Traditional Chinese Medicine constitutes a chemically diverse and pharmacologically rich reservoir of bioactive compounds, often exhibiting multi-target pharmacological properties that are potentially valuable for complex metabolic disorders such as type 2 diabetes mellitus (T2DM). However, systematic prioritization of active and safe constituents remains challenging, as therapeutic efficacy must be optimized concurrently with toxicity risk. Here, we present a safety-aware Multi-Task Graph Attention Network (SM-GAT) framework that jointly models anti-diabetic efficacy and toxicity liabilities of TCM-derived compounds within a unified architecture. Four tasks are simultaneously optimized: inhibition of dipeptidyl peptidase-4 (DPP4), inhibition of α-glucosidase, acute oral toxicity, and clinically relevant toxicity. By enabling shared molecular representation learning across heterogeneous yet biologically related endpoints, SM-GAT facilitates knowledge transfer between efficacy and safety domains. Across all prediction tasks, SM-GAT achieved competitive or superior performance compared with single-task graph neural networks and other baseline models, achieving ROC-AUC values up to 0.892 for α-glucosidase inhibition. Notably, multi-task learning yields pronounced improvements in data-limited settings, highlighting effective cross-task regularization. Large-scale virtual screening of the TCMBank library demonstrates practical applicability, enabling efficient prioritization of structurally diverse candidates with favorable predicted efficacy-safety balance. Several structurally diverse lead compounds, including ellagic acid derivatives, are identified with favorable predicted efficacy-safety balance. Furthermore, atom-level attention analysis highlighted chemically interpretable substructures associated with predicted efficacy and toxicity-related molecular representations. Collectively, this study establishes an interpretable multi-objective framework for safety-aware lead discovery, providing a computational framework for integrating traditional botanical resources into anti-diabetic lead discovery.

Keywords:  anti-diabetic activities; graph attention network; multi-task learning; natural products; virtual screeening

DOI:  https://doi.org/10.3389/fmolb.2026.1854162