bims-aukdir 2026-06-28 papers

bims-aukdir

Biomed News

on Automated knowledge discovery in diabetes research

Issue of 2026–06–28
25 papers selected by
Mott Given

An interpretable machine learning model for diabetic foot risk classification in patients with diabetes.
Diagnostic performance of multiple artificial intelligence (AI) algorithms for diabetic retinopathy screening in primary care: Evidence from real-world settings in India.
A Graph Neural Network-Based Multispectral-View Learning Model for Diabetic Macular Ischemia Detection From Color Fundus Photographs.
MultiRetNet: A Lightweight Explainable AI Approach to Diabetic Retinopathy Grading and DME Detection Using Fundus-OCT Fusion.
Privacy-aware diabetic retinopathy grading and visual lesion-focused interpretability through mixture-of-experts federated deep learning with explainable AI.
OrdPrune-KD: An Ordinal-Consistency-Based Model Compression Framework for Diabetic Retinopathy Grading.
Integrating artificial intelligence and machine learning into risk prediction for type 2 diabetes: A model for early identification of high-risk populations.
An interpretable machine learning model for predicting postoperative hypotension in type 2 diabetes mellitus undergoing non‑cardiac surgery.
Using machine learning to identify individuals at elevated risk of diabetes from low-cost complete blood count data for opportunistic screening.
Exploratory Machine Learning-Based Classification of Type 2 Diabetes Using Routine Clinical Parameters: A Single-Center Comparative Study.
An interpretable predictive model for depression risk in diabetic patients: A web-based application using NHANES data.
Empowering digital health management with on-device large language models for glucose prediction: a model development and validation study.
Explainable Lightweight AI for the Identification of Right-Sided Cardiac Dysfunction in a Saudi Arabian Diabetic Cohort.
Automated Identification and Segmentation of Diabetic Macular Edema Subtypes Using Deep Learning.
Super learner model for predicting carotid plaque regression in type 2 diabetes on PCSK9 inhibitors.
Quantifying the Economic Value of AI-Assisted Diabetic Retinopathy Screening: A Meta-Analysis Using the Incremental Net Benefit Framework.
Predictors of death in diabetes-associated Enterobacteriaceae sepsis: A Mendelian randomization and machine learning study.
Predictors of Undiagnosed Diabetes in the Bangladeshi Female Population: A Propensity Score-Weighted Machine Learning Analysis of BDHS Biomarker Data.
Foveal Density and Multi-Domain OCTA Biomarkers May Help Identify Preclinical Diabetic Microvasculopathy in Type 2 Diabetes Mellitus.
Implementation of artificial intelligence in healthcare in Aotearoa New Zealand: learnings from the diabetic retinal screening use case.
Prediction of Nocturnal Hypoglycemia Following Exercise in Type 1 Diabetes Using Temporally Structured CGM-Derived Digital Biomarkers.
Quantitative metabolomics and machine learning identify differential metabolomic biomarkers across normoglycaemia, pre-diabetes and diabetes in a Qinghai Plateau population.
Clinical Significance of Non-Invasive Skin Autofluorescence Measurement and AI Applications in Patients with Diabetic Foot Ulcers: A Scoping Review.
Assessing the Accuracy and Precision of Artificial Intelligence for Diabetes Mellitus and Hypertension Management.
[Diabetic foot care: full-lifecycle management driven by digital intelligent technologies].

Sci Rep. 2026 Jun 23.

An interpretable machine learning model for diabetic foot risk classification in patients with diabetes.

Xiaohua Hu, Huijun Xi, Zibo Fan, Leixia Zhu, Jia Shi, Shuangshuang Qian, Jing Qiu, Yingzhi Li, Hong Ding.

  Diabetic foot is a severe chronic complication of diabetes, mainly resulting from peripheral neuropathy and vasculopathy, and may progress to ulcers, infections, and amputation if not treated in a timely manner. Early identification of patients who already present a high-risk diabetic foot profile at clinical evaluation is therefore critical for guiding preventive interventions. This study aimed to develop an interpretable machine learning-based risk classification platform for classifying current diabetic foot risk status in patients with diabetes and supporting clinical risk stratification and personalized care. The model was not designed to predict future ulcer occurrence, amputation, or wound healing outcomes. In this retrospective study, data from 1938 diabetic patients at Nanfang Hospital of Southern Medical University were used for model development, and an independent external validation cohort of 695 diabetic patients from Shanghai Changhai Hospital was used to assess generalizability. Fifty clinical features covering demographic, metabolic, vascular, neurological, inflammatory, renal, and diabetes-related factors were included. Ten machine learning models were developed and compared. Recursive Feature Elimination (RFE) was applied to the top-performing models for feature selection, and SHapley Additive exPlanations (SHAP) was used to interpret model predictions and construct a diabetic foot risk classification platform. Using sixteen selected features, the CatBoost model achieved the best performance on the internal test set, with an AUC of 0.935 ± 0.016 and accuracy, precision, and recall of 0.88. In the external validation cohort, the model maintained stable performance, with an AUC of 0.922 ± 0.006 and accuracy, precision, recall, and F1-score all equal to 0.88, demonstrating good generalizability. We developed a CatBoost-based diabetic foot risk classification model incorporating sixteen clinically accessible features. The model showed stable and reliable performance across internal and external cohorts and remained robust under data noise and class imbalance, supporting its potential utility for real-world clinical risk classification and early preventive care.

Keywords:  Diabetes care; Diabetic mellitus foot; Model explanation; Risk classification model

DOI:  https://doi.org/10.1038/s41598-026-58612-w
Indian J Ophthalmol. 2026 Jun 17.

Diagnostic performance of multiple artificial intelligence (AI) algorithms for diabetic retinopathy screening in primary care: Evidence from real-world settings in India.

Anshul Chauhan, Gursimran Rana, Priyanka Verma, Mitesh Yadav, Lakshay Kumar, Gagandeep Kaur, Vishali Gupta, Geeta Menon, Piyasena Mapa Prabhath, Luke Vale, Mona Duggal.

   PURPOSE: The diagnostic performance of artificial intelligence (AI) in real-world settings remains uncertain, particularly across different fundus camera systems. This study evaluates the diagnostic accuracy of three AI algorithms for diabetic retinopathy (DR) detection using two nonmydriatic fundus cameras, assessing image gradability and DR severity.
METHODS: A prospective diagnostic accuracy study was conducted at a primary health center, Khijrabad, Punjab, India (March-July 2021). The study evaluated three commercially available artificial intelligence algorithms for DR detection using two nonmydriatic fundus cameras. Participants underwent two-field, nonmydriatic fundus imaging with both cameras. Image quality and DR presence were independently assessed by masked human graders, including optometrists and a retina specialist. Diagnostic performance was measured using sensitivity, specificity, and positive and negative predictive values.
RESULTS: A total of 272 images from 136 participants (mean age 67.7 years; 62% female) were analyzed. Human graders classified more than 97% of images as gradable, with DR detected in 47% of the images. AI-1 demonstrated the highest sensitivity (Forus: 97.5% (0.956-0.994); Intuvision: 81.7% (0.768-0.867)) but comparatively low specificity: 62.7% (58.6-66.8) and 53.8% (0.474-0.602). AI-2 displayed a balanced performance (sensitivity 80.0% and 77.0%; specificity 95.7% and 92.0%). AI-3 had a moderate sensitivity (73-80%) with the specificity ranging from 82% to 86%.
CONCLUSION: AI performance varied across camera platforms, highlighting the need for context-specific validation to ensure safe integration into primary care and guide DR screening guidelines.

Keywords:  Artificial intelligence; diabetic retinopathy screening; multiple camera; primary healthcare setting

DOI:  https://doi.org/10.4103/IJO.IJO_2774_25
Transl Vis Sci Technol. 2026 Jun 01. 15(6): 29

A Graph Neural Network-Based Multispectral-View Learning Model for Diabetic Macular Ischemia Detection From Color Fundus Photographs.

Qinghua He, Hongyang Jiang, Danqi Fang, Dawei Yang, Truong X Nguyen, Anran Ran, Clement C Tham, Simon K H Szeto, Sobha Sivaprasad, Carol Y Cheung.

Purpose: To investigate the feasibility of using color fundus photographs (CFPs) combined with artificial intelligence (AI) for detecting diabetic macular ischemia (DMI), a condition characterized by retinal capillary loss in the macula that leads to vision impairment in diabetic patients. Despite the widespread use of CFPs and AI in diagnosing other eye diseases like diabetic retinopathy (DR), their application in DMI detection remains unexplored because of skepticism regarding its viability.
Methods: A graph neural network-based multispectral-view learning (GNN-MSVL) model was developed to detect DMI from CFPs. The model uses computational multispectral imaging to reconstruct 24-wavelength pseudo-multispectral fundus images from CFPs, enhancing sensitivity to subtle reflectance changes caused by ischemic tissue. ResNeXt101 served as the backbone for multi-view feature extraction, whereas a customized GNN with jumper connections improved cross-spectral relationship learning. The study included 1078 macula-centered CFPs from 1078 eyes of 592 diabetic patients, with 530 images from 300 patients confirmed as DMI cases.
Results: The GNN-MSVL model achieved an accuracy of 84.7% and an area under the receiver operating characteristic curve of 0.900 (95% confidence interval, 0.852-0.937) at the eye level, significantly outperforming both baseline CFP-trained models and human experts (P < 0.01).
Conclusions: AI-based analysis of CFPs shows promising potential for DMI detection, offering a feasible, early, and cost-effective screening method. This approach could address the current gap in DMI diagnosis and improve clinical outcomes for diabetic patients.
Translational Relevance: This study will establish a feasible, CFP-based screening method for DMI.

DOI: https://doi.org/10.1167/tvst.15.6.29
J Imaging. 2026 May 28. pii: 236. [Epub ahead of print]12(6):

MultiRetNet: A Lightweight Explainable AI Approach to Diabetic Retinopathy Grading and DME Detection Using Fundus-OCT Fusion.

Saad Islam, Ravinesh C Deo, U Rajendra Acharya, Prabal Datta Barua, Jeffrey Soar.

  Diabetic retinopathy (DR) and diabetic macular oedema (DME) are two of the most significant preventable contributors to blindness in the adult population worldwide, yet current automated screening systems typically address each condition in isolation and rely on a single imaging modality. In this study, we propose a deep learning model that simultaneously grades DR severity and detects DME by fusing paired colour fundus and optical coherence tomography (OCT) images acquired from the same eye during the same clinical visit. Our architecture employs two parallel EfficientNet-B0 backbones pre-trained on ImageNet, one for each modality, whose 1280-dimensional feature vectors are concatenated into a 2560-dimensional joint representation. This fused representation passes through a shared fully connected block before branching into a three-class DR classification head and a binary DME detection head. We train and evaluate the model on a private dataset of 425 paired fundus and OCT eye images (850 images). The proposed architecture adopts feature-level fusion, in which modality-specific deep features are independently extracted from fundus and OCT images using separate convolutional backbones and subsequently concatenated to form a joint representation for multi-task learning. On the held-out test set (n= 85), the fusion model achieves 82.4% DR accuracy (area under the receiver operating characteristic curve [AUC] = 0.929, macro sensitivity = 0.81, macro specificity = 0.905) and 97.6% DME accuracy (AUC = 0.999, sensitivity = 0.833, specificity = 1.000). The fusion model detects 10 of 12 DME-positive eyes compared with only 7 of 12 for either the fundus-only or OCT-only baselines, representing a 43% relative improvement in DME sensitivity. Stratified five-fold cross-validation (n = 425 aggregated predictions) corroborates these findings, with the fusion model reaching 87.1% DR accuracy (AUC = 0.978) and 99.1% DME accuracy (AUC = 1.000). Gradient-weighted class activation mapping visualisations confirm that the fundus branch attends to clinically relevant macular lesions, whereas the OCT branch highlights retinal layer disruptions and subretinal fluid, providing interpretability. To the best of our knowledge, the proposed MultiRetNet is the first lightweight, task-specific multimodal architecture to jointly grade DR severity and detect DME from paired same-eye, same-visit fundus and OCT images through explicit feature-level fusion within a single end-to-end multi-task framework, distinct from recent generalist ophthalmic foundation models, supporting the value of multimodal fusion for comprehensive diabetic eye screening pending external validation.

Keywords:  Grad-CAM; deep learning; diabetic retinopathy; fundus photography; interpretability; multimodal fusion; optical coherence tomography

DOI:  https://doi.org/10.3390/jimaging12060236
Sci Rep. 2026 Jun 23.

Privacy-aware diabetic retinopathy grading and visual lesion-focused interpretability through mixture-of-experts federated deep learning with explainable AI.

Md Tanjum An Tashrif, Dipanjali Kundu, Mst Moriom Akter Bithee, Anichur Rahman, Fahmid Al Farid, Hezerul Abdul Karim, Abu Saleh Musa Miah.

  Diabetic Retinopathy (DR) is still a major cause of vision loss that can be avoided. This means that we need automated screening systems that can work across institutions without putting sensitive medical data in one place. Although Federated Learning (FL) allows for cooperative model training while preserving data locality, Non-IID data distribution, communication overhead, and unstable convergence frequently limit its effectiveness in medical imaging. This paper suggests a federated Mixture-of-Experts (FL-MoE) framework for DR classification that combines interpretable deep learning and expert specialization in order to overcome these challenges. Using the EyePACS and APTOS-2019 retinal fundus datasets, this paper evaluates multiple backbone architectures, including Convolutional Neural Networks (CNN), a hybrid CNN-LSTM model, and transformer-based Vision Transformer (ViT), within the FL-MoE framework. FL-MoE improves performance under heterogeneous client distributions for several backbone architectures, particularly CNN-LSTM, though performance varies across models. The CNN-LSTM backbone achieves 76.2% accuracy with 89.5% AUC on EyePACS while reducing communication cost by an order of magnitude compared to transformer-based models. Furthermore, CNN-LSTM exhibits more stable convergence and stronger robustness to client-level data heterogeneity. Grad-CAM based explainability analysis qualitatively shows attention maps highlighting retinal regions commonly associated with DR. To quantify localisation quality, we computed Intersection-over-Union (IoU) with IDRiD lesion masks; mean IoU values were below 0.03 for all lesion types, confirming the coarse, exploratory nature of the visualisations. Overall, the proposed FL-MoE framework with a CNN-LSTM backbone offers an effective and practical solution for scalable, privacy-aware Diabetic Retinopathy screening in federated clinical environments, outperforming both standard federated baselines and a representative personalized FL method (FedBN) under heterogeneous data conditions.

Keywords:  Diabetic retinopathy; Explainable AI; Federated learning; Mixture-of-experts; Non-IID data

DOI:  https://doi.org/10.1038/s41598-026-58292-6
Sensors (Basel). 2026 Jun 07. pii: 3636. [Epub ahead of print]26(12):

OrdPrune-KD: An Ordinal-Consistency-Based Model Compression Framework for Diabetic Retinopathy Grading.

Yuzhe Yan, Siqi Liang, Yifan Xia.

  This study proposes OrdPrune-KD, an ordinal-consistency-driven model compression framework that integrates grade-aware structured pruning with Earth Mover's Distance (EMD)-based knowledge distillation for diabetic retinopathy (DR) grading. Unlike conventional approaches that only consider ordinal relationships at the loss level, the proposed method incorporates ordinal priors into both model compression and knowledge transfer stages. Extensive experiments on APTOS 2019, Messidor-2, and IDRiD demonstrate that the proposed framework achieves a favorable balance between model compactness and predictive performance. In particular, under a 77% parameter reduction, the student model achieves competitive performance relative to the teacher model in terms of QWK while maintaining strong high-risk sensitivity. Additional ablation studies and fairness-controlled comparisons confirm that the performance gains are primarily attributed to the proposed ordinal-aware design rather than output formulation differences. These results indicate that OrdPrune-KD provides an effective and deployable solution for lightweight DR grading systems.

Keywords:  Earth Mover’s Distance (EMD); clinical decision support; deep learning; diabetic retinopathy grading; knowledge distillation; lightweight neural networks; medical image analysis; model compression; ordinal learning; structured pruning

DOI:  https://doi.org/10.3390/s26123636
Diabetes Res Clin Pract. 2026 Jun 22. pii: S0168-8227(26)00309-8. [Epub ahead of print] 113389

Integrating artificial intelligence and machine learning into risk prediction for type 2 diabetes: A model for early identification of high-risk populations.

Kiran Anya Chilu Kuri, Viola Viola, Paolo Pozzilli.

Family history, body mass index (BMI), and ethnicity are three key, well-established determinants of susceptibility to type 2 diabetes mellitus (T2DM), reflecting genetic predisposition, modifiable metabolic risk, and biological as well as social influences, respectively. These factors interact in complex, non-linear patterns that are not fully captured by conventional risk prediction models. This review examines how artificial intelligence (AI) and machine learning approaches can integrate these variables to improve risk stratification and early identification of individuals at high risk of T2DM. By leveraging large-scale, longitudinal datasets, data-driven models facilitate the capture of population-level heterogeneity and identify risk patterns that extend beyond static thresholds. Incorporating AI-enhanced prediction tools into clinical and public health settings could enable more timely, targeted, and equitable interventions. Ultimately, integrating advances in AI with a deeper understanding of the interplay between BMI, ethnicity, and genetic predisposition may support more personalised prevention strategies and risk-stratified care pathways for T2DM.

DOI: https://doi.org/10.1016/j.diabres.2026.113389
Cardiovasc Diabetol. 2026 Jun 22.

An interpretable machine learning model for predicting postoperative hypotension in type 2 diabetes mellitus undergoing non‑cardiac surgery.

Yu Gao, Guojiang Yin, Zheng Qi, Xiaoyang Song, Xiang Zhou, Kun Li.

   BACKGROUND: Postoperative hypotension (POH) is a common and serious complication in patients with type 2 diabetes mellitus (T2DM) undergoing non‑cardiac surgery, yet predictive tools tailored to this high‑risk population remain scarce.
METHODS: This single‑center cohort study developed and validated a machine learning (ML) model to predict the risk of postoperative hypotension (POH) occurring during the post‑anaesthesia care unit (PACU) stay, defined as systolic blood pressure < 90 mmHg after leaving the operating theatre and before transfer to the general ward, consistent with the Perioperative Quality Initiative (POQI) consensus. Data from 34,012 retrospective (2012-2022) and 10,528 prospective (2023-2025) T2DM patients undergoing non‑cardiac surgery were used. Following rigorous preprocessing and a four‑step feature selection, 13 predictors were retained. Fourteen ML models were trained and evaluated using area under the curve (AUC), sensitivity, specificity, and calibration. Model interpretability was enhanced using SHapley Additive exPlanations (SHAP).
RESULTS: Random Forest achieved the best overall performance, with AUCs of 0.843 (95% CI 0.837-0.849) on training, 0.854 (95% CI 0.848-0.860) on internal validation, and 0.847 (95% CI 0.840-0.854) on prospective validation. External validation on an independent hospital cohort (n = 2156) yielded an AUC of 0.822 (95% CI 0.805-0.839), confirming generalisability. It demonstrated high sensitivity (0.932) and reliable calibration. SHAP analysis identified intraoperative blood loss, age, heart failure, obstructive sleep apnoea, and body mass index as the top predictors, providing transparent global and local explanations for individual risk.
CONCLUSION: An interpretable ML model based on routinely collected clinical data accurately predicts POH risk in T2DM patients after non‑cardiac surgery. The model combines strong discriminative performance with clinical explainability, suggesting its potential as a practical tool for preoperative risk stratification and personalized postoperative monitoring in T2DM patients within similar clinical settings.

Keywords:  Explainable artificial intelligence; Machine learning; Postoperative hypotension; Random forest; Type 2 diabetes mellitus

DOI:  https://doi.org/10.1186/s12933-026-03256-3
Sci Rep. 2026 Jun 26.

Using machine learning to identify individuals at elevated risk of diabetes from low-cost complete blood count data for opportunistic screening.

Xiaolong Guo, Naiwen Zhang, Bingyi Li, XiaXia Yu, Jing Guo, Guo Dan.

Type 2 diabetes and prediabetes remain substantially underrecognized, highlighting the need for practical screening approaches based on routinely available data. This study aimed to develop and temporally validate a low-cost complete blood count (CBC)-based machine learning model for identifying individuals at elevated risk of diabetes in a routine health examination setting. We conducted a retrospective single-center study using electronic health record data from Shenzhen University General Hospital. A development cohort of 70,725 individuals examined between 2018 and 2023 was used for model development with stratified 5-fold cross-validation, and an independent temporal validation cohort of 26,650 individuals examined in 2024 was used for final evaluation. The outcome was defined as a binary diabetes-risk label based on HbA1c, with HbA1c < 5.7% classified as non-risk and HbA1c ≥ 5.7% classified as diabetes-risk, including both prediabetes and HbA1c-defined diabetes. Candidate models included logistic regression, LightGBM, multilayer perceptron, Tabular ResNet, and an attention-based tabular neural network. In the development cohort, LightGBM achieved the best overall performance, with a mean AUROC of 0.821 ± 0.006 and a mean AUPRC of 0.628 ± 0.010. In the independent 2024 temporal validation cohort, LightGBM again performed best, with an AUROC of 0.791 and an AUPRC of 0.699. Calibration analysis showed preserved risk ranking but some overestimation of absolute risk, and subgroup analysis revealed performance heterogeneity across sex, age, and BMI strata. These findings suggest that CBC-based machine learning, particularly LightGBM, may provide a practical approach for opportunistic identification of individuals at elevated risk of diabetes and support further confirmatory testing in routine health examination settings.

DOI: https://doi.org/10.1038/s41598-026-59213-3
Healthcare (Basel). 2026 Jun 15. pii: 1710. [Epub ahead of print]14(12):

Exploratory Machine Learning-Based Classification of Type 2 Diabetes Using Routine Clinical Parameters: A Single-Center Comparative Study.

Neşe Bülbül, Rukiye Çiftçi, İpek Atik, Özgür Eken, Nuriye Efe Ertürk, Monira I Aldhahi.

  Background/Objectives: Type 2 diabetes mellitus (T2DM) is a prevalent metabolic disorder associated with substantial long-term morbidity and mortality. Routinely collected anthropometric, biochemical, and hematological variables may contain useful discriminatory information for data-driven classification. This study aimed to compare the apparent classification performance of multiple machine learning algorithms for distinguishing individuals with and without T2DM using routinely obtained clinical parameters in a single-center dataset. Methods: This single-center observational study included 160 adults (95 females, 65 males) evaluated at the Endocrinology Outpatient Clinic of Gaziantep Islam Science and Technology University, Faculty of Medicine, Ersin Arslan Training and Research Hospital. The dataset comprised anthropometric measurements, biochemical markers, and complete blood count parameters. SMOTE was applied only within the training folds to address class imbalance and to avoid information leakage. Following fold-internal data preprocessing, which included imputing missing values and feature standardization where appropriate, the dataset was evaluated using stratified 5-fold cross-validation. SHAP analysis was performed to interpret the model predictions. A calibration curve was used to assess the model's reliability. Eight supervised machine learning models were evaluated with and without HbA1c: Logistic Regression, Linear Discriminant Analysis, Quadratic Discriminant Analysis, Decision Tree, Random Forest, Extra Trees, Gaussian Naive Bayes, and k-Nearest Neighbors. Model performance was evaluated using accuracy, sensitivity, specificity, and F1 score, and ROC curves were used as a diagnostic tool. Results: The models were evaluated in two different ways: with and without HbA1c. Random Forest demonstrated the best classification performance in the cross-validated evaluation; without HbA1c, it achieved 92.2% accuracy, 93.9% sensitivity, 97.9% specificity, and a 95.9% F1 score. When HbA1c was included, it achieved 98.0% accuracy, 97.9% sensitivity, 98.8% specificity, and a 99.0% F1 score. Decision Tree and Extra Trees demonstrated strong performance with accuracy rates of 87.6% and 92.8%, respectively, without HbA1c, and 90% and 93.5% when HbA1c was included; in contrast, KNN yielded the lowest accuracy rate (70.6%). Overall, tree-based models performed better than linear classifiers on this dataset. Conclusions: Machine learning models based on routine clinical and anthropometric variables demonstrated promising performance for T2DM classification in this single-center dataset; tree-based approaches yielded the most promising results. Including HbA1c improved the models' ability to classify individuals with and without T2DM. However, since HbA1c was included both as a predictor and as part of the operational definition of the diabetes group, the findings should be interpreted with caution due to the risk of target leakage. Therefore, these results should be considered exploratory rather than evidence of clinically applicable predictive performance, and an independent external validation study should be conducted prior to clinical application.

Keywords:  HbA1c; artificial intelligence; complete blood count; decision tree; machine learning; random forest; type 2 diabetes mellitus

DOI:  https://doi.org/10.3390/healthcare14121710
Medicine (Baltimore). 2026 Jun 26. 105(26): e49402

An interpretable predictive model for depression risk in diabetic patients: A web-based application using NHANES data.

Yishi Li, Tong Ren, Guanghong Zhou, Zhi Li, Chunyan Hu, Junfeng Zhao, Tianlin Guo, Yongqing Jiao, Chuanguang Zhou, Xun Wang.

  Depression is a common comorbidity in individuals with diabetes and is associated with adverse clinical outcomes. Early identification of high-risk individuals remains challenging due to the multifactorial and nonlinear nature of depression risk. Machine learning (ML) may enhance risk prediction but requires appropriate handling of class imbalance and sufficient interpretability for clinical application. The present study aimed to develop and rigorously evaluate an interpretable, class-imbalance-aware ML model for predicting depression risk among adults with diabetes, using nationally representative data from the National Health and Nutrition Examination Survey (NHANES). We analyzed cross-sectional data from 1140 adults with diabetes in the US (NHANES, 2007-2018). Depression was defined as a Patient Health Questionnaire-9 score ≥10. Predictors included demographic, clinical, lifestyle, and socioeconomic factors. Class imbalance was addressed using combined sampling and cost-sensitive learning. Seven ML models (logistic regression, support vector machine [SVM], random forest, adaptive boosting, decision tree, extreme gradient boosting, and categorical boosting [CatBoost]) were trained and evaluated. Model performance was assessed using the area under the receiver operating characteristic curve, with screening-oriented metrics optimized via threshold tuning. Model interpretability was examined using SHapley Additive exPlanations (SHAP), and clinical utility was evaluated using decision curve analysis. The SVM model demonstrated the most balanced performance after threshold optimization, achieving superior sensitivity and positive-class F1-score for depression detection. Key predictors identified by SHAP included chest pain, poverty-income ratio, sleep duration, sex, body mass index, physical activity, triglyceride levels, and diet quality (Healthy Eating Index-2020). Decision curve analysis indicated favorable net benefit for screening, particularly at lower risk thresholds. An interactive web-based application was developed to provide individualized risk predictions and explanations. An interpretable, imbalance-aware SVM model effectively predicts depression risk among adults with diabetes and supports individualized risk stratification, offering a potential tool for precision screening and early intervention.

Keywords:  NHANES; depression; diabetes; machine learning; predictive model

DOI:  https://doi.org/10.1097/MD.0000000000049402
EBioMedicine. 2026 Jun 25. pii: S2352-3964(26)00226-4. [Epub ahead of print]129 106343

Empowering digital health management with on-device large language models for glucose prediction: a model development and validation study.

Taiyu Zhu, Joanna Howson, Alejo Nevado-Holgado.

   BACKGROUND: Long-term management of chronic diseases such as diabetes is increasingly based on wearable technologies, particularly continuous glucose monitoring (CGM), integrated with smartphone-based digital health systems. When combined with artificial intelligence, especially deep learning, these systems offer highly personalised decision support, including glucose prediction. Although large language models (LLMs) have demonstrated strong performance across various healthcare tasks, their integration into day-to-day digital health remains limited, primarily due to privacy concerns associated with transmitting sensitive data to remote servers. Recent advances in lightweight LLMs create new opportunities for secure and local deployment.
METHODS: In this study, we first evaluated the zero-shot glucose prediction performance of eight pretrained lightweight LLMs across multiple model families. None achieved clinically viable outputs, highlighting the need for domain-specific adaptation. To address this, we propose GluLLM, a multimodal adaptor-based framework that enhances pretrained LLMs for on-device glucose forecasting. GluLLM integrates CGM data, daily activity logs, and electronic health records using customised encoder and decoder modules while preserving the foundational capabilities of pretrained LLMs. We trained and evaluated GluLLM on the REPLACE-BG dataset, which includes 226 individuals with type 1 diabetes, and validated it on an external cohort comprising 207 individuals with type 2 diabetes or without diabetes.
FINDINGS: Compared with 15 state-of-the-art deep learning baselines for time-series prediction, GluLLM (LLaMA 3.2 1B backbone) demonstrated superior performance, with significantly lower 30-min root mean square error than the strongest baseline (Crossformer) on REPLACE-BG and Móstoles (20.6 ± 3.5 and 9.6 ± 2.9 mg/dL; p < 0.001), and improved hypoglycaemia prediction (glucose <70 mg/dL; AUROC: 0.79 and 0.84; AUPRC: 0.55 and 0.60), respectively. Furthermore, deployment of GluLLM on two smartphone platforms demonstrated feasible computational requirements, with acceptable CPU and memory usage and low inference latency.
INTERPRETATION: GluLLM demonstrates that LLMs can support the next generation of smartphone-based digital health systems, delivering real-time, privacy-preserving clinical decision support.
FUNDING: Novo Nordisk Postdoctoral Fellowship run in partnership with the University of Oxford.

Keywords:  Continuous glucose monitoring; Deep learning; Digital health; Glucose prediction; Large language models; On-device inference

DOI:  https://doi.org/10.1016/j.ebiom.2026.106343
J Clin Med. 2026 Jun 17. pii: 4719. [Epub ahead of print]15(12):

Explainable Lightweight AI for the Identification of Right-Sided Cardiac Dysfunction in a Saudi Arabian Diabetic Cohort.

Umar Hasan, Metab Algeffari, Haifa F Alhasson, Muhammad Ali Nayeem.

  Background: Tricuspid valve dysfunction is historically underdiagnosed, and right-sided cardiac abnormalities are clinically important in high-risk diabetic populations. Diabetes may promote right-sided dysfunction through cardiometabolic remodeling, diastolic dysfunction, and elevated pulmonary pressures. This study introduces an explainable, lightweight artificial intelligence framework to infer extreme right heart phenotypes from left-sided echocardiographic and systemic clinical markers. Methods: We retrospectively analyzed an existing clinical dataset of approximately 370 Saudi Arabian individuals with diabetes mellitus. Seven baseline machine learning classifiers were evaluated using leak-aware preprocessing. To reduce optimism bias, models were validated with a True Nested Cross-Validation protocol. Probabilistic calibration, parameter reduction, and computational efficiency were prioritized for clinical triage, and SHapley Additive exPlanations (SHAP) supported transparent decision-making. Results: Random Forest was selected for its balance of discrimination and calibration. Under True Nested Cross-Validation, it achieved an AUC-ROC of 0.8602, AUC-PR of 0.8343, accuracy of 0.8075, sensitivity of 0.7733, specificity of 0.8239, Brier Score of 0.1281, Calibration Slope of 1.1720, and Observed/Expected Ratio of 0.8522. Feature ablation indicated a holistic cardiometabolic severity signal, with left atrial dimensions, diastolic dysfunction grade, and diuretic use as primary predictors. The model required 205.00 kilobytes of storage and 14.1956 milliseconds for inference. Conclusions: Left-sided cardiac markers combined with systemic indicators can flag concurrent severe right-sided abnormalities. This lightweight framework is a promising triage-oriented screening prototype for prioritizing echocardiographic assessment in high-risk diabetic patients during routine clinical visits without requiring specialized hardware; however, until prospective external validation is completed, this study must strictly be viewed as hypothesis-generating.

Keywords:  Saudi Arabia; artificial intelligence; clinical triage; diabetes mellitus; digital health; echocardiography; explainable AI; machine learning; risk stratification; tricuspid valve dysfunction

DOI:  https://doi.org/10.3390/jcm15124719
Transl Vis Sci Technol. 2026 Jun 01. 15(6): 32

Automated Identification and Segmentation of Diabetic Macular Edema Subtypes Using Deep Learning.

Ming Yan, Xianggui Zhang, Ruilong Li, Qin Ding, Ya Ye, Zhen Huang, Cong Chen, Wenjing Zhang, Lulu Tang, Yanping Song.

Purpose: The purpose of this study was to develop a deep learning algorithm capable of accurately classifying diabetic macular edema (DME) subtypes and segmenting the lesions in patients with diabetic retinopathy (DR) using structural optical coherence tomography (OCT) images.
Methods: We retrospectively collected 3120 DME OCT B-scan images from 823 eyes of patients with DME, acquired from 4 different spectral-domain and swept-source OCT devices (Topcon 3D-2000, Topcon Triton, BK400K UWF SS-OCT, and VG200 SS-OCT) to enhance device diversity and evaluate cross-device generalizability. An annotation team, consisting of two mid-career ophthalmologists and one senior retinal specialist, performed meticulous multi-label annotations, including DME subtype categories, detection bounding boxes, and pixel-level segmentation masks, to build the DME-Seg dataset. Based on this dataset, we fine-tuned the YOLO11x-Seg model for the detection and segmentation tasks.
Results: The fine-tuned model achieved promising performance on the DME-Seg dataset. For lesion detection, it attained an average mAP50(B) of 0.82, mAP50-95(B) of 0.56, and Dice coefficients of 0.82 ± 0.20 (95% confidence interval [CI] = 0.81-0.83). For segmentation, it achieved an mAP50(M) of 0.84, mAP50-95(M) of 0.54, and Dice coefficients of 0.79 ± 0.18 (95% CI = 0.78-0.80).
Conclusions: The constructed DME-Seg dataset and the validated model demonstrate promising performance in the automated detection and segmentation of DME subtypes, with encouraging cross-device generalization capability. This resource provides a foundation for advancing artificial intelligence (AI)-assisted diagnosis and personalized treatment planning for DME.
Translational Relevance: This automated quantification tool bridges the gap between AI research and clinical utility by assisting ophthalmologists in the precise diagnosis and treatment of DME.

DOI: https://doi.org/10.1167/tvst.15.6.32
Acta Diabetol. 2026 Jun 22.

Super learner model for predicting carotid plaque regression in type 2 diabetes on PCSK9 inhibitors.

Jingyao Chen, Qinglong Yang, Gaoming Hou, Jianqiao Bi, Qi Xu, Haiting Guo, Jia Sun.

   OBJECTIVES: Evidence regarding the effects of proprotein convertase subtilisin/kexin type 9 (PCSK9) inhibitors on carotid plaque regression in patients with type 2 diabetes mellitus (T2DM) and subclinical atherosclerosis remains limited; therefore, this study aimed to develop and validate a machine learning-based model for predicting carotid plaque regression in this population.
METHODS: This retrospective study included a development cohort of 204 patients with T2DM and subclinical atherosclerosis receiving combined statin and evolocumab therapy, with external validation performed in an independent cohort. The primary outcome was the change in mean carotid plaque thickness. Thirteen predictors were selected using the Boruta algorithm to construct a Super Learner model, which was validated through repeated five-fold cross-validation. Model interpretability was assessed using SHapley Additive exPlanations analysis.
RESULTS: After 24 weeks of treatment, carotid plaque regression was observed in 52.9% of patients. The model demonstrated strong discriminatory performance in the internal validation set (area under the curve [AUC] = 0.958; sensitivity = 0.881; specificity = 0.908) and moderate performance in the external validation cohort (AUC = 0.755; sensitivity = 0.875; specificity = 0.560). Low-density lipoprotein cholesterol, total cholesterol, high-sensitivity C-reactive protein, high-density lipoprotein cholesterol, and alanine aminotransferase were identified as the most influential predictors.
CONCLUSIONS: This study demonstrates the feasibility of applying machine learning algorithms to predict responses to evolocumab in patients with T2DM and subclinical atherosclerosis. Machine learning may support individualized risk stratification by identifying individuals most likely to benefit from therapy, thereby supporting personalized strategies for secondary cardiovascular disease prevention.

Keywords:  ASCVD; Machine learning; PCSK9 inhibitors; Prediction; Type 2 diabetes

DOI:  https://doi.org/10.1007/s00592-026-02729-x
Value Health. 2026 Jun 23. pii: S1098-3015(26)02522-2. [Epub ahead of print]

Quantifying the Economic Value of AI-Assisted Diabetic Retinopathy Screening: A Meta-Analysis Using the Incremental Net Benefit Framework.

Chen Wei, Liujin Zhang, Chen Chen, Yuan Le, Qing Feng, Zhihao Li, Zhiwei Leng.

   OBJECTIVE: To quantitatively synthesize the economic value of artificial intelligence (AI)-assisted screening for diabetic retinopathy (DR), a leading cause of preventable blindness.
METHODS: We conducted a systematic review and meta-analysis of model-based economic evaluations comparing AI-assisted DR screening with traditional (human grader-based) screening or no screening. Eight databases were searched for studies published between January 1, 2015, and August 1, 2025. The incremental net monetary benefit (INMB) of AI-assisted screening versus each comparator was pooled using a random-effects model. Heterogeneity was explored through subgroup analysis by analytic perspective (healthcare system vs. societal).
RESULTS: From 4,130 records, 14 studies were included in the systematic review, which 11 provided sufficient data for meta-analysis. Narrative synthesis indicated that most studies found AI-assisted screening to be cost-effective or cost-saving. Meta-analysis showed that AI-assisted screening was significantly more cost-effective than human grader-based screening, with a pooled INMB of $2,179.39 (95% CI: 1,165.13 to 3,193.65) per individual. Compared with no screening, AI-assisted screening yielded a pooled INMB of $3,606.10 (95% CI: 3,240.41 to 3,971.80) from a societal perspective.
CONCLUSIONS: AI-assisted DR screening is a cost-effective strategy, particularly for expanding screening services in resource-limited settings. Its adoption in established programs should be informed by local factors such as ophthalmologist costs and program scale. Future evaluations should incorporate real-world evidence and adhere to standardized reporting guidelines.

Keywords:  Artificial Intelligence; Cost-effectiveness; Diabetic Retinopathy; Incremental Net Benefit; Meta-Analysis; Systematic Review

DOI:  https://doi.org/10.1016/j.jval.2026.06.005
Medicine (Baltimore). 2026 Jun 19. 105(25): e49373

Predictors of death in diabetes-associated Enterobacteriaceae sepsis: A Mendelian randomization and machine learning study.

Hengtong Lou, Kaiyu Han, Yunpeng Wang.

  Diabetes mellitus is a common comorbidity that may increase susceptibility to severe infections and worse outcomes. We aimed to identify clinical predictors of in-hospital mortality in diabetic patients with Gram-negative Enterobacteriaceae sepsis using a data-driven feature-selection and machine learning (ML) approach, and to examine the causal relationship between diabetes and sepsis using Mendelian randomization (MR). We performed a retrospective cohort study of patients admitted to The Second Affiliated Hospital of Harbin Medical University between January 2018 and August 2025 with diabetes and Gram-negative Enterobacteriaceae sepsis. The Boruta algorithm was used to identify variables important for mortality prediction; selected features were entered into ML classifiers and evaluated using receiver operating characteristic analysis. Separately, MR analysis was conducted with diabetes as the exposure and sepsis as the outcome. The inverse-variance weighted method was used as the primary estimator, and MR-Egger, weighted median, simple mode, and weighted mode approaches were applied as sensitivity analyses to assess robustness and potential directional pleiotropy. Boruta selection retained 10 variables associated with in-hospital mortality: intensive care unit admission, concomitant shock, respiratory failure, liver abscess, coma, anemia, decreased platelet count, reduced red blood cell count, infection with extended-spectrum β-lactamase-producing organisms, and carbapenem-resistant Enterobacteriaceae infection. Among ML models, a neural network classifier achieved the highest discriminative performance on the validation set (area under the curve = 0.957; 95% confidence interval: 0.905-1.000). MR indicated a modest but statistically significant association of genetically predicted diabetes with risk of sepsis (inverse-variance weighted P < .05; odds ratio = 1.08, 95% confidence interval: 1.02-1.13). In this single-center cohort of diabetic patients with Gram-negative Enterobacteriaceae sepsis, intensive care unit admission, shock, respiratory failure, hepatic abscess, impaired consciousness, cytopenias, and infections due to extended-spectrum β-lactamase- or carbapenem-resistant Enterobacteriaceae-producing organisms were key predictors of mortality. A neural network model demonstrated excellent discrimination for in-hospital death. MR indicates a causal effect of diabetes on sepsis risk.

Keywords:  Boruta algorithm; Enterobacteriaceae; Mendelian randomization; diabetes; machine learning; sepsis

DOI:  https://doi.org/10.1097/MD.0000000000049373
J Diabetes Res. 2026 ;2026(1): e2162121

Predictors of Undiagnosed Diabetes in the Bangladeshi Female Population: A Propensity Score-Weighted Machine Learning Analysis of BDHS Biomarker Data.

Nahid Sultana, Kanis Fatama Ferdushi.

   BACKGROUND: In Bangladesh, diabetes mellitus has become a substantial public health burden, imposing a strain on the population both economically and clinically. Robust methods for identifying high-risk populations using national data are urgently needed due to the increasing prevalence of metabolic noncommunicable illnesses throughout South Asia. Using data from the most recent wave of a reliable national survey, this study was aimed at creating and testing a diabetes prediction model for adult female Bangladeshis.
METHODS: The 2022 Bangladesh Demographic and Health Survey (BDHS) biomarker data were examined, where key factors of undiagnosed diabetes were identified using stacked ensemble machine learning (ML) algorithms together with explainable AI (XAI) approaches after adjusting for selection bias using a propensity score model. A nomogram, a user-friendly clinical tool for risk assessment, was created. The area under the receiver operating characteristic (AUROC) curve was used to assess the model's performance.
RESULTS: Of the 18,547 weighted female participants (unweighted n = 7833), the prevalence of undiagnosed diabetes was 6.26% (95% CI: 5.52%-7.00%). Significant predictors identified included age (mean 42 ± 16 years vs. 38 ± 16 years; p < 0.001), BMI (mean 24.7 ± 4.6 kg/m2; p < 0.001), and hypertension (mean 120/78 mmHg; p = 0.001). ML models demonstrated high predictive accuracy (AUROC > 0.80), and a simplified clinical nomogram was developed to provide personalized risk scores. XAI (SHAP) analysis emphasized nonlinear influences, particularly in urban residents, who accounted for 40% of undiagnosed cases, and the richest quintile, which bore a disproportionate burden of 33% (p < 0.001) compared to just 13% in the poorest quintile.
CONCLUSION: For early diabetes screening in Bangladesh, the established predictive model and nomogram provide an evidence-based method. These technologies can help healthcare providers and policymakers tailor interventions toward high-risk groups by using nationally representative data. This could potentially lessen the sustained health and financial effects of the diabetes epidemic.

Keywords:  2022 BDHS; diabetes mellitus; explainable AI (XAI); nomogram; propensity score; public health

DOI:  https://doi.org/10.1155/jdr/2162121
Medicina (Kaunas). 2026 Jun 13. pii: 1153. [Epub ahead of print]62(6):

Foveal Density and Multi-Domain OCTA Biomarkers May Help Identify Preclinical Diabetic Microvasculopathy in Type 2 Diabetes Mellitus.

Marko Zlatanović, Maja L J Živković, Nevena Zlatanović, Mladen Brzaković, Mihailo Jovanović.

  Background and Objectives: Type 2 diabetes mellitus (T2DM) causes retinal microvascular changes that precede clinically apparent diabetic retinopathy (DR). We aimed to identify which optical coherence tomography angiography (OCTA) biomarkers best distinguish eyes with T2DM without clinical DR from healthy controls and to evaluate machine learning classifiers trained on a comprehensive 68-parameter OCTA panel. Materials and Methods: In this prospective case-control study, 80 patients with T2DM without clinical DR and 33 controls underwent 3 × 3 mm macular OCTA using an Optovue RTVue Avanti System. After outlier screening, 221 eyes (155 T2DM, 66 controls) were analyzed. Sixty-eight OCTA parameters were extracted, covering FAZ morphometry (including foveal density FD-300), SCP and DCP vessel density and layer thickness, outer-retina and choriocapillaris flow, and a full retinal-thickness map. Between-group comparisons used the Mann-Whitney U test with Benjamini-Hochberg FDR correction. Logistic regression, random forest, and XGBoost classifiers were evaluated with patient-grouped 10-fold cross-validation; feature importance was quantified via SHAP. Results: Forty-two of 68 parameters reached FDR significance (q < 0.05). Deep capillary plexus vessel density was the most discriminative family (whole image rb = -0.66, q = 2.5 × 10-13; parafovea rb = -0.64). FD-300 was reduced in T2DM (median 47.55% vs. 51.86%; rb = -0.57; q = 1.0 × 10-10) and emerged as the top SHAP feature (mean |SHAP| = 0.81). FAZ circularity decreased without FAZ-area enlargement, and outer-retina flow was paradoxically elevated (rb = +0.39), consistent with a projection artifact. XGBoost using all 68 features achieved a patient-grouped cross-validated AUC of approximately 0.91, compared with 0.85 for conventional SCP + DCP whole-image density. No parameter correlated with current HbA1c in T2DM (all q > 0.98), and the well-controlled (<7%) and poorly controlled (≥7%) subgroups were indistinguishable across five of six principal biomarkers, consistent with metabolic memory. FD-300 remained independent after adjustment for hypertension, hyperlipidemia, and age (OR = 0.76; 95% CI 0.69-0.84; p < 0.001). Conclusions: A multi-compartment OCTA panel outperforms conventional two-layer vessel-density metrics in detecting preclinical diabetic microvasculopathy, although external validation is required before clinical use. FD-300 is the single most informative biomarker, while choriocapillaris and retinal thickness measures provide complementary, compartment-specific signals. Because the OCTA signature is decoupled from the current HbA1c, screening should not be deferred in well-controlled T2DM.

Keywords:  FD-300; OCT angiography; SHAP; deep capillary plexus; diabetic retinopathy; foveal density; machine learning; metabolic memory; preclinical DR

DOI:  https://doi.org/10.3390/medicina62061153
N Z Med J. 2026 Jun 26. 139(1637): 131-136

Implementation of artificial intelligence in healthcare in Aotearoa New Zealand: learnings from the diabetic retinal screening use case.

Rosie Dobson, Cheng Kai Jin, Robyn Whittaker.

Artificial intelligence (AI) tools in diabetic retinal screening (DRS) are currently in use overseas within public health systems, with growing evidence of effectiveness. A proof of concept aimed at piloting a model of care for AI-integrated DRS to increase access to timely screening for Pacific peoples in Aotearoa New Zealand was undertaken but faced a range of challenges. What appeared to many as a straightforward AI use case proved not to be so in many ways. Challenges arose from issues related to the digital systems, challenges with adjusting models of care, variable clinician readiness and the appropriateness of the AI tools. These challenges are not unique and need to be overcome to realise the benefits of AI across many use cases. Careful planning, adequate resourcing and organisational buy-in and support are all required for any AI implementation projects to succeed and for the benefits to the population and health system to be realised.

DOI: https://doi.org/10.26635/6965.7342
Sensors (Basel). 2026 Jun 17. pii: 3842. [Epub ahead of print]26(12):

Prediction of Nocturnal Hypoglycemia Following Exercise in Type 1 Diabetes Using Temporally Structured CGM-Derived Digital Biomarkers.

Agnese Piersanti, Gaia Maria Manes, Libera Lucia Del Giudice, Laura Burattini, Christian Göbl, Andrea Tura, Micaela Morettini.

  Nocturnal hypoglycemia (NH) following exercise represents a critical challenge in the management of type 1 diabetes (T1D), particularly in pediatric populations, where its occurrence is associated with severe adverse outcomes and increased caregiver burden. This study aimed to identify an interpretable early signature based on CGM-derived digital biomarkers of post-exercise NH risk in children and adolescents with T1D. CGM data from 49 pediatric subjects (DirecNet cohort) were used to extract several CGM metrics across two temporal configurations: (i) Exercise + Cumulative, where features were computed over the exercise window and over an extended window spanning from exercise onset through recovery (16:00-17:00 and 16:00-22:00); and (ii) Exercise + Post-exercise, where features were computed separately over two non-overlapping intervals, capturing the exercise phase and the subsequent recovery phase (16:00-17:00 and 17:00-22:00). A Random Forest classifier was trained within a Leave-One-Out Cross Validation framework, incorporating variance inflation factor (VIF)-based multicollinearity filtering, minimum redundancy-maximum relevance (mRMR) feature selection, and SMOTE-based class balancing. The Exercise + Post-exercise configuration achieved superior performance: balanced accuracy (BA) = 76.9%, F1-score = 0.71, Area Under Receiver Operating Characteristic Curve (ROC-AUC) = 0.75, outperforming the Exercise + Cumulative configuration; this result was achieved using only five features: CONGA-15_EX (short-term glucose variability during exercise) emerged as the most robust predictor, alongside below_54 and above_250 (time spent in hypoglycemic and hyperglycemic ranges), MAG (mean absolute glucose change), and GRADE_hypo (hypoglycemia risk score). The generalizability of the temporal framework was further supported by independent validation on the OhioT1DM free-living cohort, where the Exercise + Post-exercise configuration (BA = 76.3%, ROC-AUC = 0.804) again outperformed the cumulative approach. These results suggest that a small set of interpretable CGM-derived features, extracted from the exercise and recovery windows, can effectively discriminate pediatric T1D subjects at risk of NH, supporting the development of lightweight CGM-only decision support tools for safer exercise management.

Keywords:  SMOTE; continuous glucose monitoring; exercise; feature engineering; machine learning; nocturnal hypoglycemia; type 1 diabetes

DOI:  https://doi.org/10.3390/s26123842
BMJ Public Health. 2026 ;4(2): e004416

Quantitative metabolomics and machine learning identify differential metabolomic biomarkers across normoglycaemia, pre-diabetes and diabetes in a Qinghai Plateau population.

Tiemei Li, Bin Zhang, Qingxia Huang, Ruijie Xu, Haobo Gao, Lin Shi, Haijing Wang, Lei Zhao, Huiru Tang, Youfa Wang, Wen Peng.

   Objective: To identify biomarkers for pre-diabetes mellitus (PreDM) and diabetes mellitus (DM) in high-altitude Tibetan populations through integrated analysis of body fat composition (BFC), haematologic parameters, and metabolomics.
Methods: 904 Tibetan participants were included. T-tests and Wilcoxon tests assessed differences in 21 blood parameters, 15 BFC measures, 7 anthropometrics, 332 metabolites among Control, PreDM and DM. Key biomarkers were identified using least absolute shrinkage and selection operator (LASSO), random forest (RF) and extreme gradient boosting (XGBoost) models. Diagnostic performance was evaluated using logistic regression, reporting precision, F1-score, sensitivity, specificity and the area under the receiver operating characteristic curve (AUROC).
Results: Preliminary analysis showed significant differences in 302 metabolites, 6 anthropometric, 14 biochemical and 15 BFC indicators between disease and control groups (p<0.05). Based on LASSO, RF and XGBoost methods, the biomarkers including body mass index, waist, visceral fat level (VFL), blood cholesterol, cholesteryl esters-to-total lipids ratio in low-density lipoprotein 6 (LDL6), phospholipid-to-total lipids ratio in intermediate-density lipoprotein (IDL) (IDPLp), polyunsaturated fatty acid/total fatty acid (PUFAp), triglycerides in LDL1, free cholesterol to total lipids ratio in LDL6, hypersensitive C reactive protein (hsCRP), red blood cell distribution width and LDL1 particle number showed differences between Control and PreDM; body mass index, waist, systolic blood pressure (BP), diastolic BP, hsCRP, VFL and particle number in IDL (IDPN) showed differences between Control and DM; IDPN, triglycerides-to-total lipids ratio in LDL3, cholesteryl esters-to-total lipids ratio in LDL1, and 3-hydroxybutyrate showed differences between PreDM and DM. Selected biomarkers demonstrated high diagnostic performance. AUROCs were 0.919 (0.883 to 0.954) for PreDM versus Control, 0.980 (0.943 to 1.000) for DM versus Control, 0.694 (0.597 to 0.783) for DM versus PreDM.
Conclusions: This study identified metabolic biomarkers (eg, IDPLp, PUFAp, hsCRP, VFL) that distinguish normoglycaemia from dysglycaemia in high-altitude Tibetans, but not preDM from DM. These findings provide insights into high-altitude metabolic dysregulation and lay a foundation for future altitude-specific screening.

Keywords:  Biometry; Diabetes Mellitus; Public Health

DOI:  https://doi.org/10.1136/bmjph-2025-004416
J Pers Med. 2026 May 26. pii: 285. [Epub ahead of print]16(6):

Clinical Significance of Non-Invasive Skin Autofluorescence Measurement and AI Applications in Patients with Diabetic Foot Ulcers: A Scoping Review.

Cosimo Aliani, Piergiorgio Francia, Cosimo Nardi, Alessandra De Bellis, Roberto Anichini, Leonardo Bocchi.

  Emerging optical technologies may offer new opportunities for the non-invasive assessment of diabetic foot ulcers (DFUs), but the role of artificial intelligence (AI)-assisted autofluorescence-based approaches remains unclear. This scoping review aimed to map and summarise the published evidence on AI-assisted analysis of autofluorescence/fluorescence-based signals for DFU assessment and management. We searched Scopus, Web of Science, Embase, PubMed, CINAHL, Google Scholar, and the SPIE Digital Library, and also considered conference proceedings. We included English-language studies published between 2010 and October 2025. Of 197 records identified through database searching, 22 full-text articles were assessed for eligibility, and 5 studies met the inclusion criteria. Four studies focused on infection-related applications, specifically bacterial burden detection and Gram-type classification, whereas one study investigated tissue oxygenation estimation using a related optical imaging approach. All included studies were published between 2022 and 2025, were conducted in India, and four of the five evaluated the same device family or related variants. Overall, the evidence base was limited, geographically restricted, and technologically narrow. In addition, reporting of participant characteristics and AI methodology was often incomplete, with several studies relying on embedded proprietary or insufficiently described algorithmic components. Taken together, the available literature supports early proof-of-feasibility in restricted and largely device-specific evaluation settings rather than robust evidence of broad clinical validity, implementation readiness, or routine-care utility. Larger, more diverse, and independently validated studies with standardised acquisition procedures and more transparent AI reporting are needed before these approaches can be meaningfully evaluated for routine DFU care.

Keywords:  artificial intelligence; autofluorescence imaging; bacterial infection; diabetic foot ulcer; machine learning; multispectral imaging; wound autofluorescence

DOI:  https://doi.org/10.3390/jpm16060285
J Clin Med. 2026 Jun 07. pii: 4419. [Epub ahead of print]15(12):

Assessing the Accuracy and Precision of Artificial Intelligence for Diabetes Mellitus and Hypertension Management.

Abdullah Al Hamid, Reema Alshahrani, Maryam Alghareeb, Hadeel Hawsawi, Renad Alghareeb, Monther Abdolmohsin Alsultan.

  Background/Objectives: Diabetes mellitus and hypertension are major chronic conditions that markedly affect patients' health and quality of life worldwide. With the rapid development of technology, there has been a growing interest in exploring the potential role of artificial intelligence (AI) in the management of such diseases. This study aims to assess the accuracy and reliability of artificial intelligence tools in providing information for diabetes mellitus and hypertension management. Methods: This study assessed the accuracy and reliability of the information provided by major AI tools such as ChatGPT, Gemini, POE, Claude, Consensus, and Perplexity. Twenty questions that are essential for the management of diabetes mellitus and hypertension were constructed based on the chapters of the respective guidelines and were fed to the AI tools. The outcomes were compared with evidence-based treatment guidelines, such as those from the American Diabetes Association (ADA), the American Heart Association (AHA), the European Society of Cardiology (ESC), and the National Institute for Health and Care Excellence (NICE). Answers were classified into "accurate ", "inaccurate", and "accurate with missing information". Three rounds of six-week intervals were conducted to assess accuracy and reliability. In addition, they were conducted to evaluate data updates by comparing answers across the rounds. Results: In round one of the evaluations, ChatGPT and Poe showed the highest accuracy, both at 65% (95% CI: 41.0-83.7), followed by Claude at 60% (95% CI: 41.0-83.7). ChatGPT had the lowest inaccuracy rate at 5% (95% CI: 1.75-33.1), while Claude demonstrated the smallest percentage of responses with missing information at only 6%. (95% CI: 12.8-54.3). In round 2, Claude markedly outperformed all other tools, achieving an accuracy rate of 95% (95% CI: 73.0-99.7) and no responses with missing information (0%). In round 3, ChatGPT came second with 70% (95% CI: 45.70-87.2) accuracy and maintained the lowest inaccuracy rate of 5% (95% CI: 0.26-26.9). Consensus had the largest inaccuracy rate at 40% (95% CI: 20.0-63.6) and the lowest accuracy rate at 40% (95% CI: 20.0-63.6). Overall, statistically significant pairwise comparisons showed that Cloud in the second round has the highest accuracy compared to Poe (p = 0.0154), Gemini (p = 0.0421), Consensus (p = 0.0035), and Perplexity (p = 0.0302). In the assessment of performance shift from round 1 to round 2, Claude achieved the greatest improvement in accuracy at 40%. In the assessment of performance shift from round 2 to round 3, Poe improved the most with an accuracy increase of 25%, while ChatGPT followed with 20%. When evaluating the unprompted and guideline-prompted questions for all AI tools using McNemar's test, it did not reveal a statistically significant distinction in the proportion of accurate responses (p > 0.05). Conclusions: Throughout the three rounds, ChatGPT maintained the best performance, with the fewest missing data. Claude and Poe followed, showing high accuracy with relatively low inaccuracy rates. On the other hand, Perplexity and Gemini performed moderately, while Consensus had the lowest accuracy.

Keywords:  artificial intelligence (AI) tools; diabetes mellitus; guidelines; hypertension

DOI:  https://doi.org/10.3390/jcm15124419
Zhongguo Xiu Fu Chong Jian Wai Ke Za Zhi. 2026 Jun 15. 40(6): 893-898

[Diabetic foot care: full-lifecycle management driven by digital intelligent technologies].

Lihong Chen, Xingwu Ran.

   Objective: To conduct a comprehensive review of the latest diagnostic and therapeutic advancements in early warning, regenerative medicine, and digital management for the diabetic foot (DF).
Methods: Through a systematic review of recent global literature, the paradigm shifts in diabetic foot definitions, artificial intelligence (AI)-driven clinical applications, regenerative strategies, and full-lifecycle management models were evaluated.
Results: The emergence of "diabetes-related foot disease" and the "remission" status has refocused the clinical attention on proactive prevention. AI has demonstrated high objectivity and consistency in risk stratification, early warning, automated wound quantification, and decision support. Meanwhile, regenerative therapies, particularly mRNA-based treatments and bio-active scaffolds, have emerged as a premier research frontier and a pivotal breakthrough in the repair of refractory wounds. The implementation of full-lifecycle management has effectively achieved a closed-loop continuum of care, spanning from early intervention to clinical remission.
Conclusion: The field of diabetic foot care is evolving into a new era characterized by the integration of digital intelligence and precision regeneration. Future initiatives should focus on the multidisciplinary-led clinical translation and standardization of emerging technologies to establish a comprehensive, precision management framework aimed at achieving "long-term remission".

Keywords:  Diabetic foot; artificial intelligence; diabetes-related foot disease; full-lifecycle management; regenerative medicine

DOI:  https://doi.org/10.7507/1002-1892.202603041