bims-aukdir 2026-04-19 papers

bims-aukdir

Biomed News

on Automated knowledge discovery in diabetes research

Issue of 2026–04–19
23 papers selected by
Mott Given

Early Type 2 diabetes risk prediction using explainable machine learning in a two-stage approach.
Comparative assessment of diagnostic agreement between artificial intelligence and general practitioners in diabetic retinopathy screening using non-mydriatic fundus photography.
Robust Deep Learning Framework for Early Diabetic Retinopathy Detection Using Preprocessed Fundus Images and Optimized CNNs.
Biomarkers discovery of diabetic retinopathy through serum metabolomics - Retinal image association analysis with machine learning.
Multimodal Machine Learning for Diabetic Peripheral Neuropathy Diagnosis: Methodological Advances and Future Directions.
Gated Backbone Fusion with Transformer Encoder for Diabetic Foot Osteomyelitis Screening and Localization in Radiographs.
Fully Automatic Diabetic Wound Segmentation Using Lightweight Deep Convolutional Neural Networks.
Machine Learning-Based Risk Stratification for Metabolic Dysfunction Severity Among Diabetic Patients With Established MASLD: Distinguishing Normal-Weight From Overweight Individuals.
Retraction: DiabetesXpertNet: An innovative attention-based CNN for accurate type 2 diabetes prediction.
Referral uptake after diabetic retinopathy screening with artificial intelligence-assisted care pathways: a systematic review and meta-analysis.
Comparative machine learning analysis of saliva and plaque microbiomes in Kuwaitis with type 1 diabetes.
Dynamic Surveillance of the Cardiovascular Risk and Identification of Treatment Responders in Type 2 Diabetes Using A Machine Learning-Based Model.
Accuracy of deep learning for risk prediction and screening of diabetic foot Ulcers: A systematic review and meta-analysis.
Artificial intelligence in mapping nursing diagnoses, interventions, and outcomes for diabetes management.
Predicting 10-Year Diabetes Risk Through Physiological Acceleration: A Longitudinal Deep Learning Ensemble Approach.
Use of Continuous Glucose Monitoring With Machine Learning to Identify Metabolic Subphenotypes and Inform Precision Lifestyle Changes.
Prediction of maturity-onset diabetes of the young subtypes using machine learning.
An Explainable Artificial Intelligence Tool (FibroX) for Detecting Advanced Liver Fibrosis in Type 2 Diabetes: Protocol for a Pilot Crossover Trial.
Lipidomics and machine learning revealing dysregulation of specific triacylglycerol and phosphatidylglycerol as hub lipids associated with fetal growth in gestational diabetes mellitus.
Robust benchmarking of DeepLabv3 hybrid models for multiclass fundus screening and referable-disease triage on the FIVES dataset.
Enhancing Type 1 Diabetes Polygenic Risk Prediction Through Neural Networks and Entropy-Derived Insights.
DPA-Net: A dual-path attention neural network for estimating glycemic metrics from self-monitored blood glucose data.
Challenges of using generative artificial intelligence for diabetes patient education: a cross-platform analysis of the quality, readability, and actionability of text generated by large language models.

Front Digit Health. 2026 ;8 1743619

Early Type 2 diabetes risk prediction using explainable machine learning in a two-stage approach.

Silas Majyambere, Tony Lindgren, Celestin Twizere, Isaac Ntakirutimana.

   Background: Diabetes is a chronic disease characterized by elevated blood glucose levels. Without early detection and proper management, it can lead to serious complications and increase healthcare costs. Its global prevalence is rising, with many cases remaining undiagnosed. In this study, we developed an explainable machine learning model using a two-stage approach for predicting diabetes.
Methods: Five machine learning (ML) models, including Multi-Layer Perceptron, Support Vector Machine, K-Nearest Neighbor, Extreme Gradient Boosting (XGBoost), and Naïve Bayes, were trained and evaluated using a two-stage approach. In Stage one, a public dataset containing 520 samples was used, and Shapley Additive exPlanations (SHAP) and MLP weights were applied for feature selection. In Stage two, the same models were trained and evaluated using a dataset of 270,943 samples collected from Rwanda. SHAP was further employed to explain the model output.
Results: In Stage one, the Multi-Layer Perceptron model achieved the best performance on a public dataset, with an accuracy of 95.19%. Feature selection techniques identified the top 10 influential predictors associated with diabetes risk, including those recommended by diabetes care providers in Rwanda. In Stage two, the XGB model outperformed other models, achieving an accuracy of 97.14%.
Conclusion: This study presents a two-stage, explainable machine learning framework for systematic screening for type 2 diabetes. The first stage evaluates risk based on reported symptoms, while the second stage incorporates demographic, anthropometric, and vital sign data for refined risk assessment. Integration of these models into the mUzima mobile application can enhance community health workers' capacity to identify and refer high-risk individuals. By enabling early and accurate detection, the proposed approach has the potential to reduce undiagnosed diabetes and support improved disease management.

Keywords:  Multi-Layer Perceptron; diabetes management; diabetes prediction; explainable machine learning; interpretability; shap

DOI:  https://doi.org/10.3389/fdgth.2026.1743619
Prim Care Diabetes. 2026 Apr 16. pii: S1751-9918(26)00079-3. [Epub ahead of print]

Comparative assessment of diagnostic agreement between artificial intelligence and general practitioners in diabetic retinopathy screening using non-mydriatic fundus photography.

María Alcalá-Millán, Juan Carlos Romero-Vigara, Eva Trillo-Calvo.

   AIMS: To assess the diagnostic agreement between an artificial intelligence (AI) system and general practitioners (GPs) interpreting fundus photographs for diabetic retinopathy (DR) screening, using the ophthalmologists' assessment as the reference standard.
METHODS: We performed a cross-sectional study of 500 primary care patients with type 2 diabetes (T2DM). Each underwent two 45° non-mydriatic fundus photos per eye. Images were independently evaluated by three trained GPs, a deep learning-based AI system (EyeArt v3.0.0 (Eyenuk Inc.)), and an ophthalmologist (reference standard). Diagnostic agreement was measured with Cohen's kappa and Cramer's V, and sensitivity, specificity, predictive values, likelihood ratios, and overall accuracy were calculated.
RESULTS: Mean age was 64 years, and 59% were men. DR prevalence was 11%. AI showed near-perfect agreement with ophthalmology (κ = 0.91) and outperformed GPs, with 98.2% sensitivity, 98.9% specificity, and 98.8% accuracy. GPs showed substantial agreement (κ = 0.84), with lower sensitivity (86.4%) but similarly high specificity (98.1%). In multinomial analysis, AI achieved 87.6% sensitivity for mild DR and 100% for moderate-to-severe DR, missing no clinically relevant cases.
CONCLUSIONS: In this cohort restricted to gradable images, AI demonstrated higher sensitivity, specificity, and diagnostic agreement than GPs when compared with a single ophthalmologist reference standard. Supervised use in primary care could strengthen population-based screening, but large-scale adoption will require multicentre studies, formal cost-effectiveness analyses, and validation in unselected screening populations.

Keywords:  Artificial intelligence; Diabetic retinopathy; Non-mydriatic fundus photography; Population screening; Primary care; Type 2 diabetes mellitus

DOI:  https://doi.org/10.1016/j.pcd.2026.04.007
J Vis Exp. 2026 Mar 24.

Robust Deep Learning Framework for Early Diabetic Retinopathy Detection Using Preprocessed Fundus Images and Optimized CNNs.

Zhina Zhu, Dandan Lin, Qiuyu Wang.

Diabetic retinopathy (DR) is a leading cause of vision loss globally, particularly among patients with poorly controlled diabetes. Early detection through automated image analysis has emerged as a scalable solution to support timely intervention. Deep learning-based models, particularly convolutional neural networks (CNNs), have shown significant promise in DR detection from retinal fundus images. This study aimed to (i) develop an ensemble deep learning framework using EfficientNetB0 and DenseNet121 for five-stage DR classification, (ii) systematically evaluate the impact of fundus image preprocessing on diagnostic performance, (iii) incorporate Grad-CAM-based visual explanations for lesion localization, and (iv) achieve high diagnostic accuracy with computational efficiency suitable for scalable screening. A dataset comprising 53,412 fundus images from EyePACS, APTOS 2019, and a tertiary-care centre in China was curated. Preprocessing steps included contrast-limited adaptive histogram equalization (CLAHE), artifact removal, and normalization. Transfer learning was applied using EfficientNetB0 and DenseNet121 backbones, followed by hybrid assembling. The models were evaluated using accuracy, macro-AUC, sensitivity, specificity, and F1-score. Grad-CAM was used to visualize lesion localization. The hybrid ensemble model achieved 91.2% accuracy, 0.961 macro-AUC, 92.1% sensitivity, and an F1-score of 0.913. Preprocessing improved performance by 3-4%, and the ensemble approach outperformed standalone CNNs. Grad-CAM overlays confirmed accurate lesion localization. Model performance was evaluated for both five-stage DR grading and binary referable DR detection to reflect clinical screening requirements. This study presents a clinically viable, explainable deep learning model for DR detection. Future work will include external validation on independent datasets, prospective real-world evaluation, and model optimization (e.g., pruning and quantization) for mobile and point-of-care screening applications.

DOI: https://doi.org/10.3791/69901
Exp Eye Res. 2026 Apr 12. pii: S0014-4835(26)00163-6. [Epub ahead of print]268 111007

Biomarkers discovery of diabetic retinopathy through serum metabolomics - Retinal image association analysis with machine learning.

Deliang Lv, Haozhuo Liu, Xianguo Shi, Muchun Zhu, Huazhen Huang, Wang Li, Xiaobing Wu, Fengzhu Xie, Wei Xie, Qinggang Shang, Xiaoyan Wu, Youxi Luo, Fan Hu, Peng Yin, Nan Li, Zhiguang Zhao.

The aim of this study is to develop an innovative method of machine learning combining metabolomic and radiomic analyses for identifying biomarkers to distinguish diabetic retinopathy (DR) patients, non-retinopathy diabetic (NDR) patients and healthy individuals. The serum metabolic profiling of 94 DR patients, 95 NDR patients, and 95 healthy individuals was acquired through the Shimadzu LC-40D X3 and AB Sciex zenoTOF 7600 tandem system. We first conducted a differential analysis on metabolomic data, identifying distinct metabolites and metabolic pathways. Then, we employed Artificial Intelligence (AI) model for learning phenomics (retinal fundus image) and identifying the candidate metabolic biomarkers. Resnet50 was used as the backbone network for DR test. Finally, we performed a correlation analysis between image data and metabolome data. We unveiled the serum metabolic profiling of 94 DR patients, 95 NDR patients and 95 healthy individuals. DR test shows that the AUC (Area Under the Curve) of 0.89 on the independent test set of DR and NDR retinal images. The result of correlation analysis demonstrated 20 significantly metabolites, with three potential biomakers validated through evidenced-based filtering. Using machine learning algorithms, we achieved remarkable accuracy (AUC = 0.99) in distinguishing non-retinopathy diabetic individuals from DR patients using L-carnitine, beta-hydroxymyristic acid, and 5-Methyl-H4SPT. This study synergizes metabolomic and radiomic methodologies to identify biomarkers that can effectively distinguish NDR patients from DR patients. Furthermore, we have proven the feasibility of DR diagnosis through a doctor-free AI model using phenomics-metabolomics analysis. Additionally, exploring the metabolome data may provide new insights into the mechanism of DR.

DOI: https://doi.org/10.1016/j.exer.2026.111007
J Ultrasound Med. 2026 Apr 15.

Multimodal Machine Learning for Diabetic Peripheral Neuropathy Diagnosis: Methodological Advances and Future Directions.

Jiaxin Cai.

DOI: https://doi.org/10.1002/jum.70257
J Imaging Inform Med. 2026 Apr 14.

Gated Backbone Fusion with Transformer Encoder for Diabetic Foot Osteomyelitis Screening and Localization in Radiographs.

S Qasim Abbas, Sajib Saha, Jason Dowling, Maryam Mehdizadeh, Janardhan Vignarajan, Ashu Gupta, Shaun Frost, Aaron Nicolson, Hang Min.

  Diabetic foot osteomyelitis (DFO) is a leading cause of lower-extremity complications in individuals with diabetes, and timely, accurate screening is critical to prevent severe outcomes such as limb amputation. Although conventional radiography remains the most accessible imaging modality, the subtle and heterogeneous appearance of DFO often results in delayed or missed detection. Despite the rich morphological information encoded in foot radiographs, current deep learning methods tend to underperform in capturing localized pathological patterns due to architectural limitations. In this work, we propose Dual Backbone with Gated Fusion and Transformer encoder (DualBack-GFT), a deep learning framework for automated detection and localization of DFO in plain radiographs. The model leverages two complementary backbones, EfficientNet-B6 and ResNet-50, fused via a gated mechanism that adaptively combines image-specific features. These fused representations are further refined using transformer encoders, which effectively model long-range dependencies. The architecture operates in two stages: binary classification followed by confidence-weighted bounding-box localization. We evaluate DualBack-GFT on a curated, expert-annotated baseline dataset of diabetic foot X-rays with both diagnostic and bounding-box labels. The model achieves an AUC of 0.9683 and an average ground truth coverage of 62.71%, outperforming established baselines. These results underscore the potential of dual-stage, attention-enhanced models for interpretable and robust DFO assessment in clinical radiographs.

Keywords:  Deep learning; Diabetic foot osteomyelitis; Dual-backbone networks; Lesion localization; Radiographic screening

DOI:  https://doi.org/10.1007/s10278-026-01936-w
J Imaging Inform Med. 2026 Apr 14.

Fully Automatic Diabetic Wound Segmentation Using Lightweight Deep Convolutional Neural Networks.

Sajib Saha, Janardhan Vignarajan, Cesar Munoz, Carmen Gloria Bezanilla Collell, Inti Paredes, Cristián Salas Del Campo, Elvira Isabel Villegas Barrios, Christian Larsen Mella, Rajiv Jayasena.

  Diabetic foot ulcers, resulting from neuropathic and/or vascular complications in patients with diabetes mellitus, pose a major global health challenge. Early detection and consistent monitoring of wound progression are essential for timely intervention, effective treatment, and the prevention of severe complications such as amputation. In modern diabetic foot care, images captured using digital cameras and mobile phones are increasingly employed for remote wound assessment. In this context, automated segmentation of these wounds from such images plays a vital role by enabling objective and quantitative evaluation of wound areas-crucial for tracking the progression of healing over time. Recent years have witnessed growing interest in deep learning-based wound segmentation techniques, with a particular focus on models that are both computationally efficient and suitable for deployment on resource-constrained devices, including smartphones and point-of-care platforms. In this study, we propose a lightweight convolutional neural network (CNN) for diabetic foot wound segmentation that augments the U-Net architecture with ghost feature generation and Convolutional Block Attention Modules (CBAM) to improve computational efficiency and feature representation. The model was evaluated on a privately annotated dataset of 3450 diabetic foot wound images and compared against state-of-the-art architectures, including SegNet, U-Net, MobileNetV2, Mask R-CNN, and the domain-specific approach of Wang et al. We further investigated a fully automated two-step pipeline for wound segmentation incorporating a prior foot segmentation-based ROI detection. Using ROI detection, the proposed CNN achieved a precision of 85.13%, recall of 91.84%, Dice coefficient of 86.95%, and IoU of 77.23%. These results demonstrate competitive performance relative to high-capacity models while maintaining substantially reduced computational complexity, highlighting its suitability for real-time clinical deployment in low-resource environments.

Keywords:  CNN; Convolutional neural network; Diabetic foot ulcer; Diabetic wounds; Segmentation

DOI:  https://doi.org/10.1007/s10278-026-01941-z
J Diabetes Res. 2026 ;2026(1): e4207980

Machine Learning-Based Risk Stratification for Metabolic Dysfunction Severity Among Diabetic Patients With Established MASLD: Distinguishing Normal-Weight From Overweight Individuals.

Yuelan Yin, Sa Ke, Yilin Liu, Xiaoling Wang, Qing Xue, Weining Sun.

   OBJECTIVE: To rank biochemical and clinical features that distinguish normal-weight (BMI 18-24 kg/m2) from overweight diabetics already carrying a discharge diagnosis of metabolic dysfunction-associated steatotic liver disease (MASLD), and to build a parsimonious model that can flag lean individuals at highest risk of advanced metabolic complications.
METHODS: This study collected a total of 3524 samples from hospitalized patients with diabetes and nonalcoholic fatty liver disease (NAFLD), with 54 NAFLD features serving as the original dataset. After data preprocessing, 2624 samples and 52 NAFLD features were screened from the original dataset to form the final dataset for model input. Among these, 1848 patients were labeled as Class 0 (BMI > 24 kg/m2), and 776 patients were labeled as Class 1 (BMI between 18 and 24 kg/m2). Data visualization and exploratory data analysis were performed using t-SNE and heat maps. A five-fold cross-validation with 10 repetitions was employed for model optimization. The predictive model was evaluated using a confusion matrix. Three optimal predictive models with the smallest error were established: random forest, logistic regression, and gradient boosting. The top 30 feature variables were ultimately selected. An independent dataset containing 699 cases was used for external validation.
RESULTS: Uric acid, vitamin D, hemoglobin, and creatine kinase are the most significant features in normal-weight diabetes patients with MASLD. Gradient boosting was considered the best model; the average area under the ROC curve (AUC) was 0.733 (95% CI: 0.7089-0.7578).
CONCLUSION: Gradient boosting is the optimal predictive model, which can assist healthcare professionals in risk assessment and management for diabetic MASLD patients with normal BMI.

Keywords:  fatty liver; lean MASLD; machine learning; metabolic dysfunction-associated steatotic liver disease (MASLD); predictive models

DOI:  https://doi.org/10.1155/jdr/4207980
PLoS One. 2026 ;21(4): e0345989

Retraction: DiabetesXpertNet: An innovative attention-based CNN for accurate type 2 diabetes prediction.

PLOS One Editors

DOI: https://doi.org/10.1371/journal.pone.0345989
NPJ Digit Med. 2026 Apr 17.

Referral uptake after diabetic retinopathy screening with artificial intelligence-assisted care pathways: a systematic review and meta-analysis.

James A Leigh, Alex Sherrington, Angus R J Barber, Angus W Turner, Michael Kidd, John Powell, Catherine Pope.

Artificial intelligence (AI) is an accurate screening tool for diabetic retinopathy (DR), the leading cause of blindness among working-aged adults. However, its impact on referral uptake is uncertain. We searched Embase, MEDLINE, Scopus, Web of Science and Cochrane Library databases from year 2000 to February 17, 2025. Randomised and non-randomised studies comparing referral uptake after AI-assisted DR screening versus standard of care were included. 2644 articles were identified, and six included for analysis. The relative risk of DR referral uptake with AI-assisted screening compared with the status quo was 1.89 (95% CI, 1.18, 3.03, I2 = 91.9%). Settings which underwent referral pathway transformation from routine to targeted referrals for DR demonstrated the greatest effect size. Most (n = 4) studies also utilised behavioural change interventions enabled by immediate results acquisition of AI to enhance health-seeking behaviour. Our findings suggest the effectiveness of DR screening is derived not only from diagnostic technology, but from AI-enabled care pathway redesign encompassing both health system transformation and coordinated patient-facing interventions which improve referral uptake.

DOI: https://doi.org/10.1038/s41746-026-02616-3
Front Microbiol. 2026 ;17 1735375

Comparative machine learning analysis of saliva and plaque microbiomes in Kuwaitis with type 1 diabetes.

Hend Alqaderi, Rebecca Batorsky, George Azar, Md Zubbair Malik, Rasheeba Nizam, Khaled Altabtbaei, Sriraman Devarajan, Rasheed Ahmad, Dominique S Michaud, Naisi Zhao, Athanasios Zavras, Fahd Al-Mulla.

   Background: Type 1 diabetes (T1D) is associated with microbial dysbiosis. While most research has focused on the gut microbiome, limited data addresses the role of the oral microbiome in T1D. The oral and gut microbiomes overlap substantially, and the oral cavity may influence gut microbial composition. Saliva and dental plaque represent two distinct oral niches with unique microbial communities, but it remains unclear which is better associated to systemic disease states like T1D. This study compares the performance of salivary and plaque microbiomes in classifying pediatric T1D status.
Methods: Paired saliva and plaque samples were collected from 46 children (23 with T1D, 23 healthy controls). Microbial DNA was extracted and sequenced targeting the 16S rRNA gene. Data were processed using QIIME 2 for taxonomic classification and centered log-ratio transformation. Alpha diversity, microbial abundance, and clustering analyses were performed to compare the oral microbiome between T1D and control groups. Random forest classifiers were used to evaluate and compare the predictive accuracy of saliva- and plaque-based models, both with and without clinical metadata.
Results: Saliva samples exhibited lower alpha diversity than plaque but had significantly higher bacterial load and total microbial abundance. Saliva-based models outperformed plaque-based models, achieving a classification accuracy of 94.2% with or without clinical metadata, compared to 73.3% accuracy for plaque-based models. ROC curve analysis further supported this difference, with saliva models reaching an AUC of approximately 0.94, versus 0.75 for plaque, indicating superior discriminative performance. UMAP clustering revealed more distinct separation of T1D and control groups in salivary profiles than in plaque. Feature importance analysis identified both unique and shared taxa predictive of T1D in each niche. Incorporating clinical and demographic metadata did not enhance model performance, underscoring the robustness and predictive strength of microbiome data alone.
Conclusion: The salivary microbiome is a more effective biospecimen than dental plaque for characterizing T1D-associated microbial profiles in children. It offers superior classification accuracy and greater sensitivity in distinguishing T1D status, supporting saliva's potential as a non-invasive, scalable medium for future microbiome-based monitoring.

Keywords:  biofilm; diabetes; microbiome; oral; plaque; saliva; type 1 diabetes

DOI:  https://doi.org/10.3389/fmicb.2026.1735375
Eur J Prev Cardiol. 2026 Apr 09. pii: zwag196. [Epub ahead of print]

Dynamic Surveillance of the Cardiovascular Risk and Identification of Treatment Responders in Type 2 Diabetes Using A Machine Learning-Based Model.

Qi Huang, Xiantong Zou, Edward J Boyko, Zhouhui Lian, Xianghai Zhou, Xueyao Han, Linong Ji.

   AIMS: To establish a model to predict cardiovascular risk and identify treatment response for patients with type 2 diabetes.
METHODS: Data from 11677 patients without prior cardiovascular disease across four clinical trials: ACCORD and CANVAS (80% for training and 20% for holdout) and CANVAS-R and CREDENCE (for external testing) were used to develop and validate the machine learning (ML)-cardiovascular disease (CVD) model. The model utilized baseline and 1-year changes of interim factors for the dynamic prediction of the primary endpoint, including cardiovascular death, nonfatal myocardial infarction, nonfatal stroke.
RESULTS: The ML-CVD Primary model demonstrated strong predictive performance with Harrell's C-index of 0.66-0.71 in the holdout and external testing datasets, outperforming traditional scores in primary-prevention subgroup. Intensive interventions (intensive glycemic control and canagliflozin) significantly mitigated the progression of ML-CVD Primary risk scores during the 1-year observation period compared to control treatments. Each standard deviation decrease in the ML-CVD Primary score was significantly associated with a reduced risk of primary cardiovascular outcomes. We stratified patients in the canagliflozin arms based on their score changes: 'Responders' were defined as individuals with a decrease in the ML-CVD Primary score, whereas 'Non-Responders' showed no change or an increase in the score. 'Responders' exhibited a 45% lower risk of primary cardiovascular outcomes compared to 'Non-Responders'.
CONCLUSION: The ML-CVD Primary model enables dynamic prediction of cardiovascular events, facilitating ongoing risk surveillance and identification of individual drug responses. This approach holds promise for guiding personalized cardiovascular protective therapies in patients with type 2 diabetes.

Keywords:  cardiovascular risk; responders; type 2 diabetes

DOI:  https://doi.org/10.1093/eurjpc/zwag196
Diabetes Res Clin Pract. 2026 Apr 13. pii: S0168-8227(26)00181-6. [Epub ahead of print] 113262

Accuracy of deep learning for risk prediction and screening of diabetic foot Ulcers: A systematic review and meta-analysis.

Tongling Zhao, Jing Yu, Ruixue Deng, Lu Ding, Xiangyan Li, Jia Mi.

   BACKGROUND: This research sought to systematically estimate the accuracy of deep learning (DL) in diagnosing diabetic foot ulcers (DFUs), thereby providing novel insights for the development and updating of artificial intelligence (AI)-assisted tools.
METHODS: This study searched PubMed, Cochrane, Embase, and Web of Science up to October 2025 for original studies that used image-based DL to detect DFUs. The risk of bias of the included studies was estimated utilizing QUADAS-AI. In the meta-analysis, we constructed diagnostic 2 × 2 tables and applied a bivariate mixed-effects model.
RESULTS: In total, 55 studies were included. Of these, 32 were included in the meta-analysis, involving 87 diagnostic 2 × 2 tables for validating DL models. The pooled results exhibited that for the overall validation sets, the sensitivity and specificity were 0.96 (95% CI: 0.94-0.98) and 0.97 (95% CI: 0.94-0.98), respectively. In multicenter datasets, the sensitivity, specificity, 0.87 (95% CI: 0.68-0.96), 0.92 (95% CI: 0.82-0.97).
CONCLUSIONS: This research discloses that image-based DL for detecting DFUs might be a promising approach. Future studies should further ascertain the generalizability of models derived from public databases to individual-level data across diverse geographic regions, thereby providing evidence for the evaluation and development of AI-assisted tools.

Keywords:  Deep learning; Diabetic foot ulcer; Imaging

DOI:  https://doi.org/10.1016/j.diabres.2026.113262
Invest Educ Enferm. 2026 Mar;44(1):

Artificial intelligence in mapping nursing diagnoses, interventions, and outcomes for diabetes management.

Beatriz Barboza Fernandes, Rodrigo de Araujo Marques, Rosane Barreto Cardoso, Rejane Prado Dos Santos, Thaíssa Felix Affonso.

   Objective: To map nursing diagnoses, nursing outcomes, and nursing interventions based on the clinical indicators described in the Type 2 Diabetes Diagnosis and Management Manual, using artificial intelligence (AI) (GPT-4®).
Methods: Descriptive study with adapted cross-mapping. GPT-4® was applied with a structured prompt to identify clinical indicators in the manual and correlate them with nursing classifications.
Results: AI identified 43 clinical indicators, and after manual review, 30 were confirmed, with 23 overlapping between methods. From these, 30 nursing diagnoses, 30 expected outcomes, and 30 interventions were mapped. Manual mapping consolidated 15 nursing diagnoses, 15 outcomes, and 15 interventions.
Conclusion: AI proved effective in expediting and standardizing cross-mapping in nursing. However, human clinical judgment was indispensable to validate and adjust inconsistencies, capturing nuances not identified by AI. The integration of AI with clinical reasoning can strengthen care systematization, support evidence-based protocols, and improve outcomes in patients with diabetes.

Keywords:  artificial intelligence; cross-mapping; diabetes mellitus type 2; nursing care; nursing diagnosis; nursing process; standardized nursing terminology; treatment outcome

DOI:  https://doi.org/10.17533/udea.iee.v44n1e07
Diagnostics (Basel). 2026 Mar 25. pii: 992. [Epub ahead of print]16(7):

Predicting 10-Year Diabetes Risk Through Physiological Acceleration: A Longitudinal Deep Learning Ensemble Approach.

Sangsoo Kim, Seonghee Park, Jinmi Kim, Ha Jin Park, Soree Ryang, Myungsoo Im, Doohwa Kim, Kyeongjun Lee.

  Background/Objectives: Type 2 diabetes (T2D) develops gradually over many years through a prolonged preclinical phase, yet traditional static risk scores often fail to capture these dynamic metabolic trajectories. We propose a longitudinal deep learning framework to predict the 10-year risk of Type 2 diabetes onset defined by comprehensive ADA criteria by modeling the physiological acceleration of routine clinical biomarkers. Methods: Utilizing an 18-year longitudinal dataset from the community-based Korean Genome and Epidemiology Study (KoGES) cohort, we selected N=4354 participants with complete follow-up records, ensuring high data integrity without requiring synthetic data augmentation. We constructed a 3-dimensional tensor of 21 non-invasive clinical variables spanning a 6-year observation window. To resolve the inherent precision-recall trade-offs of individual models, we developed a stacking ensemble that integrates Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures via a logistic regression meta-learner. To evaluate the added value of longitudinal modeling, we compared this dynamic framework against a static XGBoost baseline that only saw the most recent data. Results: Evaluated on an independent test set (n=874), the ensemble significantly outperformed baseline models, achieving an overall accuracy of 0.90 (95% CI: 0.88-0.92) and an AUROC of 0.94 (95% CI: 0.93-0.95). By harmonizing LSTM's sensitivity and GRU's precision, the model yielded an exceptional Positive Predictive Value (PPV) of 0.97, a sensitivity of 0.80, and a specificity of 0.98. Conclusions: This framework provides a highly accurate, resource-efficient triage instrument for T2D screening, thereby reducing unnecessary clinical alerts and improving screening efficiency.

Keywords:  Type 2 Diabetes Mellitus; deep learning; ensemble learning; longitudinal trajectory; precision medicine

DOI:  https://doi.org/10.3390/diagnostics16070992
J Diabetes Sci Technol. 2026 Apr 14. 19322968261431860

Use of Continuous Glucose Monitoring With Machine Learning to Identify Metabolic Subphenotypes and Inform Precision Lifestyle Changes.

Ahmed A Metwally, Heyjun Park, Yue Wu, Tracey McLaughlin, Michael P Snyder.

  The classification of diabetes and prediabetes by static glucose thresholds obscures the pathophysiological dysglycemia heterogeneity, primarily driven by insulin resistance (IR), β-cell dysfunction, and incretin deficiency. This review demonstrates that continuous glucose monitoring (CGM) and wearable technologies enable a paradigm shift toward non-invasive, dynamic metabolic phenotyping. We show evidence that machine learning models can leverage high-resolution glucose data from at-home, CGM-enabled oral glucose tolerance tests to accurately predict gold-standard measures of muscle IR and β-cell function. This personalized characterization extends to real-world nutrition, where an individual's unique postprandial glycemic response (PPGR) to standardized meals, such as the relative glucose spike to potatoes versus grapes, could serve as a biomarker for their metabolic subtype. Moreover, integrating wearable data reveals that habitual diet, sleep, and physical activity patterns, particularly their timing, are uniquely associated with specific metabolic dysfunctions, informing precision lifestyle interventions. The efficacy of dietary mitigators in attenuating PPGR is also shown to be phenotype-dependent. Collectively, this evidence demonstrates that CGM can deconstruct the complexity of early dysglycemia into distinct, actionable subphenotypes. This approach moves beyond simple glycemic control, paving the way for targeted nutritional, behavioral, and pharmacological strategies tailored to an individual's core metabolic defects, thereby paving the way for a new era of precision diabetes prevention.

Keywords:  CGM; artificial intelligence; diabetes; wearables

DOI:  https://doi.org/10.1177/19322968261431860
Front Digit Health. 2026 ;8 1656161

Prediction of maturity-onset diabetes of the young subtypes using machine learning.

Israel Figueroa, Ricardo Flores, Andrea Millán, Alejandro de Dios, Gustavo Daniel Frechtel, Ariel Pablo López, Daniela Mennickent.

   Introduction: Maturity-onset diabetes of the young (MODY) is a monogenic type of diabetes caused by different pathogenic genetic variants in glucose metabolism-related genes, with GCK-MODY and HFN1A-MODY subtypes being the most frequent. Diagnosing the specific MODY subtype is essential for correct treatment and follow-up, but it requires gene sequencing, a time-consuming and costly process that depends on highly skilled professionals. Therefore, it is mandatory to develop tools that allow to correctly determine in which order to study the involved genes, reducing the number of sequencing procedures to find the causal variant and making the diagnostic process more efficient. This proof-of-concept study evaluates machine learning as a complement to clinical characterization and genetic testing, by optimizing binary classification models for explainable prediction of MODY subtypes, with a focus on GCK-MODY and HFN1A-MODY.
Methods: To meet this aim, we analyzed medical data from a diabetes cohort from Buenos Aires, Argentina. By employing imputation and oversampling techniques we created 10 datasets for each subtype to feed a pipeline that trained, optimized and evaluated 10 machine learning techniques.
Results: Gaussian Naive Bayes achieved the best predictive power for GCK-MODY with a ROC AUC score of 0.724, meanwhile Random Forest yielded 0.712 for HNF1A-MODY. SHAP analysis provided insights into feature importance, highlighting the explainability of our approach.
Discussion and conclusion: This novel study demonstrates for the first time the viability of machine learning as a supplementary tool prior to MODY genetic testing, by providing cost-effective and explainable models able to assist health professionals in the diagnosis of MODY subtypes.

Keywords:  MODY; classification; diabetes; explainable AI; machine learning; subtypes

DOI:  https://doi.org/10.3389/fdgth.2026.1656161
JMIR Res Protoc. 2026 Apr 13.

An Explainable Artificial Intelligence Tool (FibroX) for Detecting Advanced Liver Fibrosis in Type 2 Diabetes: Protocol for a Pilot Crossover Trial.

Basile Njei, Ulrick Sidney Kanmounye.

BACKGROUND: Metabolic dysfunction-associated steatotic liver disease is highly prevalent in adults with type 2 diabetes, and advanced fibrosis is its strongest prognostic marker. However, existing noninvasive tools may underperform in diabetes care and are inconsistently used in practice.
OBJECTIVE: To evaluate the feasibility, usability, and preliminary diagnostic effectiveness of FibroX, an explainable artificial intelligence tool for identifying metabolic dysfunction-associated steatotic liver disease-associated advanced fibrosis in adults with type 2 diabetes.
METHODS: This 12-month provider-level randomized crossover pilot trial will enroll at least 36 primary care clinicians managing adults with type 2 diabetes. Participants will complete 2 simulated care periods, one with FibroX-enabled care and one with usual care, separated by a 1-week washout. FibroX generates individualized advanced fibrosis risk estimates from routine clinical data, provides guideline-aligned triage categories, and displays case-level explanatory outputs using Shapley Additive Explanations. FibroX is an investigational tool and is not currently used in routine clinical practice. Primary outcomes are feasibility and usability, while secondary and exploratory outcomes include workflow efficiency, preliminary diagnostic performance, and implementation measures.
RESULTS: Institutional Review Board approval has been obtained. The protocol was first posted on ClinicalTrials.gov on December 1, 2025 (NCT07305324). At the time of submission, recruitment had not yet started. The estimated study start date is June 15, 2026, with primary completion anticipated on May 15, 2027, and overall study completion on June 15, 2027.
CONCLUSIONS: This pilot study will provide preliminary evidence on the feasibility, usability, and diagnostic performance of an explainable artificial intelligence tool for fibrosis risk stratification in diabetes care and will inform the design of a future larger trial.
CLINICALTRIAL: ClinicalTrials.gov NCT07305324.

DOI: https://doi.org/10.2196/90456
PeerJ. 2026 ;14 e20841

Lipidomics and machine learning revealing dysregulation of specific triacylglycerol and phosphatidylglycerol as hub lipids associated with fetal growth in gestational diabetes mellitus.

Zhenguo Lin, Lei Li, Yuhang Fan, Hao Chen, Yanhua Cai, Rongrong Yang, Mei Chen, Guolin Hong.

   Objective: Emerging evidence links lipid metabolism to the pathogenesis of gestational diabetes mellitus (GDM). This study aimed to identify lipidomic biomarkers and explore their clinical significance for GDM and related fetal growth and development through serum lipid profiling.
Methods: Lipidomic profiles of pregnant women with and without GDM were analyzed using Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA), Uniform Manifold Approximation and Projection (UMAP), volcano plots, and heatmaps. Carbon chain length and unsaturation effects on fold change (FC) were evaluated. Pathway analysis was performed via the Lipid Ontology (LION) platform, while lipid networks were constructed using Debiased Sparse Partial Correlation (DSPC). Hub lipids were identified through topological analysis and visualized with UpsetR. A GDM detection model was developed using Boruta and LogitBoost algorithms, assessed by receiver operating characteristic (ROC) curve analysis, and interpreted via Local Interpretable Model-agnostic Explanations (LIME).
Results: Twelve serum lipid metabolites were significantly associated with GDM risk. Phosphatidylglycerol (PG)(O-27:1) and triacylglycerol (TG)(35.5) were identified as hub lipids. The GDM detection model, incorporating TG(35:5), PG(O-27:1), total protein (TP), and red blood cell distribution width (RDW), achieved high accuracy.
Conclusion: This study preliminarily characterized lipid metabolic pathway disturbances in patients with GDM, highlighting the potential of integrating lipidomics with interpretable machine learning techniques for biomarker discovery and mechanistic insight.

Keywords:  Biomarkers; Fetal growth; Gestational diabetes mellitus; Lipidomics; LogitBoost

DOI:  https://doi.org/10.7717/peerj.20841
Front Med (Lausanne). 2026 ;13 1771083

Robust benchmarking of DeepLabv3 hybrid models for multiclass fundus screening and referable-disease triage on the FIVES dataset.

Igor Garcia-Atutxa, Laura Martinez-Baliu, Francisca Villanueva-Flores.

   Background: Automated analysis of color fundus photographs can support scalable screening for diabetic retinopathy (DR), age-related macular degeneration (AMD), and glaucoma, but single-run reporting and accuracy-only summaries can mask clinically relevant instabilities and failure modes.
Methods: Using the public FIVES dataset, we benchmarked six deep learning configurations for four-class fundus screening (AMD, DR, glaucoma, normal): three DeepLabv3-backbone hybrids (ResNet50, DenseNet121, EfficientNet-B0) and three backbone-only classifiers. All experiments were evaluated using five independent stratified splits generated with different random seeds (n = 5 runs), each defining a distinct 20% held-out test set. Models were trained on the remaining 80% (training/validation), and all reported metrics are computed on the 20% test set of each run and summarized as mean ± SD across runs. Performance was summarized with accuracy, sensitivity, specificity, and one-vs.-rest AUC; we further characterized clinical behavior via row-normalized confusion matrices, per-class precision/recall/F1, and a screening-style binary triage setting (referable = AMD ∪ DR ∪ glaucoma vs. normal).
Results: Hybrid models consistently achieved higher discrimination than simple classifiers (AUC 0.969-0.979 vs. 0.908-0.920), despite similar accuracies (0.924-0.941). The selected model, DeepLabv3-DenseNet121, reached the highest AUC (0.979 ± 0.009). Class-wise analysis revealed strong performance for Normal (F1 0.970 ± 0.014) and Glaucoma (F1 0.896 ± 0.048), while DR was the main bottleneck (sensitivity 0.738 ± 0.117), with most DR errors redistributed to AMD (13.6%) and Glaucoma (12.0%) and minimal confusion with Normal (0.5%). In binary triage, the model achieved sensitivity 0.993 ± 0.011 and specificity 0.963 ± 0.034, with PPV 0.987 ± 0.013 and NPV 0.980 ± 0.032, and a stable referral rate (∼0.73-0.77) across runs.
Conclusion: DeepLabv3-based hybrids provide a robust advantage in AUC for multiclass fundus screening on FIVES. The residual risk concentrates in the DR-AMD-Glaucoma decision boundary, suggesting that deployment-oriented policies should prioritize conservative handling of DR-adjacent cases while leveraging the stability of Normal predictions for screening workflows.

Keywords:  DeepLabv3; age-related macular degeneration; diabetic retinopathy; fundus photography; glaucoma; referable-disease triage

DOI:  https://doi.org/10.3389/fmed.2026.1771083
Int J Mol Sci. 2026 Mar 25. pii: 2966. [Epub ahead of print]27(7):

Enhancing Type 1 Diabetes Polygenic Risk Prediction Through Neural Networks and Entropy-Derived Insights.

Antonio Nadal-Martínez, Guillermo Pérez-Solero, Sandra Ferreiro López, Jorge Blom-Dahl, Eduard Montanya, Marta Alonso-Bernáldez, Moises Shabot, Christian Binsch, Lukasz Szczerbinski, Adam Kretowski, Julián Nevado, Pablo Lapunzina, Robert Wagner, Jair Tenorio-Castano.

  Type 1 diabetes (T1D) is an autoimmune disease with a strong genetic component (~70% heritability). Early identification of individuals at risk is crucial for early intervention or risk assessment. Although polygenic risk scores (PRS) have shown promise in risk assessment, most current approaches remain constrained by linear assumptions and limited generalizability. We aimed to develop a neural network-driven classifier using T1D-associated single nucleotide polymorphisms (SNPs). In addition, we explored the inclusion of an entropy-derived feature as a complementary variable, representing the degree of genetic variability within an individual's genotype profile across the 67 T1D-associated SNPs, to evaluate its potential additive contribution to the model performance. We analyzed genotype data from 11,909 individuals in the UK BioBank (546 T1D cases and 11,363 controls). Sixty-seven well-known SNPs associated with T1D were utilized as inputs to the model, using two distinct allele-encoding strategies. A feed-forward neural network was evaluated under varying case-control ratios through five-fold cross-validation. Performance was assessed using the area under the receiver operating characteristic curve (AUC) on a held-out test set and on an external European cohort as a validation cohort. Across five-fold cross-validation, the best configuration achieved a median AUC of 0.903. On the held-out UK Biobank test set, the model generalized well, with an AUC of 0.8889 (95% CI: 0.8516-0.9262). A probability-based risk framework, constructed using five risk groups ("very low", "low", "intermediate", "high", and "very high" risk), yielded a negative predictive value (NPV) of 98.9% for the "very low" risk group and a Positive Predicted Value (PPV) of 61.9% with a specificity of 97.3% for the "very high" risk group, assuming a 10% T1D prevalence. External validation in the German Diabetes Study reproduced clear case-control separation; for individuals with recent onset diabetes and glutamic acid decarboxylase antibodies (GADA+) vs. controls, specificity reached 91.9% in the "high" risk group (PPV of 94.3%) and 97.6% in the "very high" risk group (PPV of 95.7%). The proposed neural network reliably predicts T1D genetic risk using a compact SNP panel of 67 SNPs and maintains accuracy in both internal and external European cohorts. Its probabilistic output enables clinically interpretable risk thresholds, while entropy features contributed modestly to performance. These results demonstrate that a neural network-based approach achieves discriminative performance that is comparable to established T1D genetic risk models, while offering flexible probability-based risk stratification and architectural extensibility for future integration of additional features.

Keywords:  PRS; UK Biobank; entropy; genomic medicine; machine learning; neural network; newborn screening; polygenic risk score; risk stratification; type 1 diabetes

DOI:  https://doi.org/10.3390/ijms27072966
Comput Biol Med. 2026 Apr 11. pii: S0010-4825(26)00237-4. [Epub ahead of print]208 111673

DPA-Net: A dual-path attention neural network for estimating glycemic metrics from self-monitored blood glucose data.

Canyu Lei, Benjamin Lobo, Jianxin Xie.

  Continuous glucose monitoring (CGM) provides dense and dynamic glucose profiles that enable reliable estimation of glycemic metrics, such as time-above-range (TAR), time-in-range (TIR), and time-below-range (TBR). However, the cost and limited accessibility of CGM restrict its widespread adoption, particularly in low- and middle-income countries. In contrast, self-monitoring of blood glucose (SMBG) is inexpensive and widely available, but produces sparse and irregular measurements that are typically event-driven: patients often check glucose levels when feeling unwell (e.g., dizziness, fatigue, or discomfort). Such behaviorally triggered sampling leads to biased estimates of TAR, TIR, and TBR. To address this challenge, we propose a Dual-Path Attention Neural Network (DPA-Net) that generates unbiased time-in-ranges estimates from SMBG data by leveraging generalizable knowledge learned from large collections of paired SMBG-CGM data. DPA-Net integrates two complementary paths: (1) a spatial-channel attention path that reconstructs a CGM-like continuous glucose trajectory from sparse SMBG inputs, and (2) a multi-scale residual network path that directly predicts glycemic metrics. An inter-path alignment mechanism enforces consistency between the reconstructed trajectory and the predicted metrics, thereby reducing bias and mitigating overfitting. Furthermore, to overcome the scarcity of real-world paired SMBG-CGM datasets, we develop an Active Point Selector (APS) that models behavioral patterns underlying SMBG measurements. Utilizing large-scale CGM recordings, APS identifies the most probable temporal instances at which users would self-monitor their glucose levels and formulates a synthetic SMBG-CGM paired dataset. Experimental results demonstrate that DPA-Net achieves robust accuracy with low estimation errors and minimal systematic bias. To the best of our knowledge, this is the first machine learning framework that utilizes the knowledge of a vast amount of CGM data designed to infer key glycemic metrics from SMBG data, offering a practical framework to enhance SMBG-based glycemic assessment in settings where CGM is unavailable or unaffordable.

Keywords:  Deep learning; Diabetes management; Glycemic control; Self-monitored blood glucose (SMBG); Time in range (TIR)

DOI:  https://doi.org/10.1016/j.compbiomed.2026.111673
Front Public Health. 2026 ;14 1804524

Challenges of using generative artificial intelligence for diabetes patient education: a cross-platform analysis of the quality, readability, and actionability of text generated by large language models.

Zhiqiang Wang, Xiaoya Li, Xianglan Tao, Jie Li, Li Zhang, Xiaorong He, Jing Yang.

   Objective: To compare, across large language model (LLM) platforms, the quality, readability, and completeness of action-oriented instructions in diabetes self-management education texts, and to quantify the associations among these domains to inform model selection and risk mitigation.
Methods: Ten LLM platforms were used to generate diabetes education texts (total n = 200), stratified by topic. Outcomes included the Global Quality Score (GQS), the Patient Education Materials Assessment Tool for Printable Materials (PEMAT-P), and EQIP-36 (Ensuring Quality Information for Patients, 36-item version). Text characteristics, including word count, sentence count, and syllable count, were recorded. Readability was assessed using the Automated Readability Index (ARI), Coleman-Liau Index (CLI), Flesch-Kincaid Grade Level (FKGL), Flesch Reading Ease Score (FRES), Gunning Fog Index (GFOG), Linsear Write (LW), and the Simple Measure of Gobbledygook (SMOG). Between-platform differences were evaluated using one-way ANOVA or the Kruskal-Wallis test, as appropriate. Associations between readability indices and GQS, PEMAT-P, and EQIP-36 were examined using correlation heat maps and exploratory stepwise multiple linear regression. Because the readability indices were highly intercorrelated, these regression analyses were considered exploratory and were used to identify candidate readability-related correlates rather than definitive independent predictors.
Results: GQS and PEMAT-P differed significantly across platforms (both p < 0.001), whereas EQIP-36 did not (p = 0.062). Text length and readability also varied by platform (most p < 0.001). After stratification by topic, PEMAT-P understandability, PEMAT-P total score, and GQS no longer differed significantly across topics (p = 0.356, p = 0.247, and p = 0.182, respectively), whereas PEMAT-P actionability (p < 0.001), EQIP-36 (p < 0.001), and several readability metrics remained significantly different. Difficulty indices were strongly intercorrelated, and FRES was inversely associated with multiple difficulty indices. Exploratory regression analyses suggested that greater reading burden tended to co-occur with lower GQS, PEMAT-P, and EQIP-36 scores.
Conclusion: LLM-generated diabetes education texts exhibit marked cross-platform heterogeneity, and exploratory analyses suggest a potential trade-off between readability and both information quality and the completeness of action-oriented instructions. Clinical implementation should therefore combine careful platform selection, structured prompting with templates, human-AI review, and continuous quality monitoring to support safe, readable, and actionable patient education.

Keywords:  cross-platform evaluation; diabetes mellitus; diabetes self-management education and support (DSMES); large language model (LLM); readability; text quality

DOI:  https://doi.org/10.3389/fpubh.2026.1804524