bims-aukdir 2026-03-08 papers

bims-aukdir

Biomed News

on Automated knowledge discovery in diabetes research

Issue of 2026–03–08
sixteen papers selected by
Mott Given

Modeling strategies for CGM data: A scoping review of mechanistic, machine learning, and hybrid approaches in diabetes management.
Development and validation of a machine learning model to predict comorbid hypertension in patients with type 2 diabetes.
Research on a Predictive Model for Microalbuminuria in Type 2 Diabetes Based on Machine Learning and SHAP Analysis.
Future-aware blood glucose forecasting using knowledge distillation with transformer-based sequence-to-sequence models.
Exploring predictive factors of physiological, biochemical indicators, and lifestyle for macrovascular complications in type 2 diabetes: a synthesis of machine learning models.
A copula based supervised filter for feature selection in machine learning driven diabetes risk prediction.
Dynamic C-reactive protein trajectories predict prolonged healing time in diabetic wounds: a machine learning model based on a single-center cohort with standardized wound size.
Patient-specific deep offline artificial pancreas for blood glucose regulation in type 1 diabetes.
Application of AI and digital health tools in public health management of T2DM: from mechanism prediction to personalized treatment.
Autonomous AI-assisted diabetic retinopathy screening at primary care is associated with increased presentation to eye care by at risk patients.
MVTT-GMamba: A Multimodal Graph Reasoning Framework for Anti-VEGF Efficacy Prediction in Diabetic Macular Edema.
Grading Diabetic Retinopathy Using Comparative Assessment: A Pilot Study Comparing Paired Image Comparisons With Direct Grading.
A review of optimization strategies for deep and machine learning in diabetic macular edema.
Screening, validation, and transcriptional regulation analysis of oxidative stress-related biomarkers in gestational diabetes mellitus: SH3BP5, ITGAM, PRRG1, and MIS12.
An integrative analysis of cardiac autonomic neuropathy and nephropathy risk assessed with SUDOSCAN in individuals with type 2 diabetes.
Evaluating Accuracy of ChatGPT-4o in Automated Carbohydrate Estimation From Images as a Self-Management Tool for Adolescents With Type 1 Diabetes.

Comput Biol Med. 2026 Mar 03. pii: S0010-4825(26)00164-2. [Epub ahead of print]205 111601

Modeling strategies for CGM data: A scoping review of mechanistic, machine learning, and hybrid approaches in diabetes management.

Nilde Fera, Anna Procopio, Paolo Zaffino, Antonio Cutruzzolà, Concetta Irace, Carlo Cosentino.

   BACKGROUND: The increasing availability of continuous glucose monitoring (CGM) data has opened new avenues for modeling glucose dynamics in diabetes management.
OBJECTIVE: This scoping literature review uniquely explores the full methodological spectrum applied to CGM data analysis, ranging from classical Mechanistic Models (MM) and statistical time series approaches, to modern Artificial Intelligence (AI) techniques, and emerging hybrid frameworks that combine the two paradigms. Unlike prior reviews focused solely on Type 1 Diabetes Mellitus (T1DM), the present work includes modeling efforts across different classes of populations, including Type 2 Diabetes Mellitus (T2DM), Gestational Diabetes Mellitus (GDM), and other forms of diabetes.
METHODS: Literature was systematically retrieved from Elsevier Scopus®, Clarivate Web of Science™, and PubMed®, providing a comprehensive and comparative assessment of state-of-the-art strategies for CGM-based analysis in diverse clinical contexts.
RESULTS: The reviewed studies indicate a clear methodological shift toward data-driven and hybrid frameworks. Overall, both mechanistic and AI-based models achieve satisfactory predictive performance; however, their strengths differ across tasks. Machine learning methods are particularly effective for event detection and feature extraction, whereas deep learning models excel in forecasting CGM-derived glucose trajectories. Hybrid approaches further enhance predictive accuracy while preserving physiological interpretability, especially over longer prediction horizons.
CONCLUSION: Challenges such as data heterogeneity, the limited availability of high-quality datasets beyond T1DM, and reduced cross-cohort generalizability persist, underscoring the need for standardized validation procedures and physiologically informed modeling strategies.

Keywords:  Continuous glucose monitoring; Diabetes; Physiological models

DOI:  https://doi.org/10.1016/j.compbiomed.2026.111601
Front Med (Lausanne). 2026 ;13 1754916

Development and validation of a machine learning model to predict comorbid hypertension in patients with type 2 diabetes.

Hailu Yang, Changfeng Fan, Chunyan Liu.

   Background: Hypertension is a critical comorbidity in patients with type 2 diabetes mellitus that significantly increases cardiovascular risk. Although several predictive models have been developed using conventional logistic regression or basic machine learning algorithms, these approaches often face significant limitations. Many existing models suffer from a lack of external validation which limits their generalizability, or they operate as black boxes without providing interpretable clinical insights. Furthermore, most prior studies have focused exclusively on biological indicators while overlooking the potential impact of socioeconomic determinants and lifestyle factors on disease progression.
Objective: To address these gaps, this study aimed to develop a high-performance Random Forest model for predicting hypertension risk in diabetic patients by integrating multidimensional data, including clinical metrics, lifestyle habits, and socioeconomic status. The study further sought to validate the model's robustness using an independent external cohort and assess its clinical utility through SHAP analysis, providing transparent interpretations of risk factors to guide personalized medical decision-making.
Methods: A multicenter retrospective cohort study was conducted using electronic medical records from two tertiary hospitals. Eligible adults with type 2 diabetes and no prior hypertension were included. A total of 900 eligible patients were included, with 420, 180, and 300 participants in the training, testing, and external validation cohorts, respectively. Feature selection combined Boruta and LASSO methods, yielding seven predictors. Seven algorithms were tested, and model performance was assessed through cross-validation, independent testing, and external validation. The random forest model was explained using SHAP analysis.
Results: Among 900 participants, the random forest model achieved the best discrimination, with AUCs of 0.89 in internal testing and 0.83 in external validation. Calibration and decision curve analyses confirmed stability and clinical utility. Key predictors included alcohol consumption, triglycerides, diabetes duration, health insurance type, fasting blood glucose, estimated glomerular filtration rate, and exercise frequency.
Conclusion: The validated random forest model effectively predicts hypertension in type 2 diabetes patients, integrating metabolic, behavioral, and socioeconomic factors. Its interpretability and robust performance support its potential use for early identification and personalized prevention of hypertension in clinical practice.

Keywords:  hypertension risk; machine learning; predictive modeling; random forest; type 2 diabetes mellitus

DOI:  https://doi.org/10.3389/fmed.2026.1754916
Int J Endocrinol. 2026 ;2026 6356560

Research on a Predictive Model for Microalbuminuria in Type 2 Diabetes Based on Machine Learning and SHAP Analysis.

Zixuan Liu, Zhuolin Zhou, Yu Sun, Xiaotian Du, Haixia Zhang, Cheng Ji.

   Background: Type 2 diabetes mellitus (T2DM) is associated with kidney damage, with microalbuminuria (MAU) serving as an early marker indicating the risk of progression to severe renal and cardiovascular complications, and there is an urgent need for effective prediction tools to identify MAU risk in T2DM patients and prevent adverse outcomes. This study aims to develop a machine learning-based model to enhance the early identification of high-risk individuals and facilitate timely, personalized interventions.
Methods: The electronic medical records of 4170 patients were retrospectively extracted from the diabetes special database of Nanjing Drum Tower Hospital (Ethics approval number: 2021-403-02). The data were divided into training and testing sets (8:2 ratio), and random forest-based recursive feature elimination method was employed to identify the most pertinent input variables for the predictive model. Five machine learning models were applied to predict the progression to MAU. The Shapley additive explanations (SHAP) values were applied for model interpretation to assess feature contributions. Ten features were selected for the construction of a prediction model.
Results: For predicting the progression to MAU, the Light GBM model demonstrated the best performance (AUC 0.85, 95% CI 0.82-0.88). By analyzing the Shapley values of the model outputs, we identified the key risk factors for predicting the diagnosis of MAU at both the cohort and individual levels.
Conclusions: This study developed an interpretable machine learning model to predict MAU in T2DM patients, enabling effective risk stratification and identification of high-risk individuals based on baseline data to guide personalized clinical interventions and optimization of treatment.

Keywords:  albuminuria; algorithms; electronic medical records; machine learning; type 2 diabetes mellitus

DOI:  https://doi.org/10.1155/ije/6356560
Sci Rep. 2026 Feb 28.

Future-aware blood glucose forecasting using knowledge distillation with transformer-based sequence-to-sequence models.

Xiaoyu Sun, Hongru Li, Xia Yu.

Accurate blood glucose level (BGL) forecasting is critical for diabetes self-management and clinical decision-making. Although deep learning models based on continuous glucose monitoring (CGM) data have achieved encouraging results, most approaches rely exclusively on historical observations and cannot explicitly account for future disturbances, such as insulin delivery and meal intake, that are unavailable at deployment. To address this limitation, we propose a future-aware learning framework for multi-step BGL prediction that leverages privileged information during training while preserving deployability at inference. A Transformer-based teacher model is trained offline using both historical CGM data and future disturbance information to learn disturbance-aware temporal representations. A student model with a similar sequence-to-sequence structure is then trained using knowledge distillation to approximate the teacher's representations based solely on historical inputs, enabling real-time forecasting without access to future data. The proposed framework is evaluated on the publicly available OhioT1DM and AZT1D datasets for prediction horizons ranging from 30 to 120 minutes and compared with several established methods. The results show consistent reductions in root mean squared error and mean absolute error, together with improved clinical reliability as assessed by Clarke error grid analysis, with over 90% of predictions falling within clinically acceptable regions. These findings demonstrate the potential of future-aware training strategies to enhance glucose forecasting performance under realistic deployment constraints.

DOI: https://doi.org/10.1038/s41598-026-41787-7
Front Endocrinol (Lausanne). 2026 ;17 1696240

Exploring predictive factors of physiological, biochemical indicators, and lifestyle for macrovascular complications in type 2 diabetes: a synthesis of machine learning models.

Bo Shang, Yaoqin Lu, Chengjing Wei, Yunlong Li, Qingyue Yang, Yukai Li, Sheng Jiang, Yinxia Su.

   Background: Traditional risk models for macrovascular complications in type 2 diabetes (T2DM) rely on physiological and biochemical indicators, which may lack long-term follow-up data and thus potentially overlook key variables.
Methods: A retrospective cohort study was conducted on 4,186 T2DM patients from the Diabetes Health Management Platform in Hotan, Xinjiang, covering the period from 2015 to 2023. Eight machine learning (ML) algorithms were used, with an 8:2 random split into training (n=3,348) and validation (n=838) sets. Performance was evaluated using the area under the receiver operating characteristic curve (AUC), and feature contributions were analyzed using SHAP values. The clinical applicability was verified through decision curve analysis.
Results: The T2DM with macrovascular complications group had significantly higher waist circumference, oropharyngeal abnormalities, and absent lung crackles (P < 0.05). The T2DM with macrovascular complications group also had significantly higher BMI, body temperature, and ALT (P < 0.05), but lower fasting blood glucose, with borderline abnormalities in blood urea and AST. The T2DM with macrovascular complications group had higher smoking, alcohol consumption, and exercise frequency (P < 0.05), but a reverse trend in self-reported "poor" health status (P < 0.05). Among all machine learning models (training AUC 0.68-0.85), XGBoost performed best (training AUC = 0.830, validation AUC = 0.850), with superior clinical net benefit compared to traditional strategies. SHAP analysis revealed that BMI (contribution +0.1116), body temperature (+0.0923), and LDL-C (+0.0821) were key predictive factors, with elevated body temperature potentially indicating subclinical inflammation activation.
Conclusions: Among patients with vascular complications, the disconnect between health behavior risks and subjective health perception is more pronounced. Elevated body temperature, high blood pressure, triglycerides, and fasting glucose indicate inflammation, increasing cardiovascular risk; moderate regular exercise provides protection.

Keywords:  XGBoost; machine learning; macrovascular complications; predictive modeling; type 2 diabetes mellitus

DOI:  https://doi.org/10.3389/fendo.2026.1696240
Sci Rep. 2026 Mar 05.

A copula based supervised filter for feature selection in machine learning driven diabetes risk prediction.

Agnideep Aich, Md Monzur Murshed, Sameera Hewage, Amanda Mayeaux.

  Effective feature selection is critical for building robust and interpretable predictive models, particularly in medical applications where identifying risk factors in the most extreme patient strata is essential. Traditional methods often focus on average associations, potentially overlooking predictors whose importance is concentrated in the tails of the data distribution. In this study, we introduce a novel, computationally efficient supervised filter that leverages a Gumbel copula implied upper-tail concordance score ([Formula: see text], a monotone transformation of Kendall's τ) to rank features by their tendency to be simultaneously extreme with the positive class. We evaluated this method against four standard baselines (Mutual Information, mRMR, ReliefF, and L1/Elastic-Net) across four classifiers on two diabetes datasets: a large-scale public health survey (CDC, [Formula: see text]) and a classic clinical benchmark (PIMA, [Formula: see text]). Our analysis included comprehensive statistical tests, permutation importance, and robustness checks. On the CDC dataset, our method was the fastest selector and reduced the feature space by ≈52%. While this resulted in a minimal but statistically significant performance trade-off compared to using all 21 features, our filter significantly outperformed standard filters (Mutual Information, mRMR) and was statistically indistinguishable from the strong ReliefF baseline. On the PIMA dataset (8 predictors), our method's ranking produced the numerically highest ROC-AUC, despite paired DeLong tests showing no statistically significant differences versus strong baselines. PIMA thus serves as a ranking-only sanity check that our upper-tail criterion behaves sensibly in a low-dimensional clinical setting. Across both datasets, the Gumbel-[Formula: see text] selector consistently identified clinically coherent and impactful predictors. We conclude that feature selection via upper-tail dependence is an efficient and interpretable screening approach that can complement standard feature-selection baselines in public health and clinical risk prediction.

Keywords:  Copula-based feature selection; Diabetes; Gumbel copula; Machine learning; Public health; Risk prediction; Supervised feature selection; Tail dependence

DOI:  https://doi.org/10.1038/s41598-026-41874-9
Front Med (Lausanne). 2026 ;13 1778003

Dynamic C-reactive protein trajectories predict prolonged healing time in diabetic wounds: a machine learning model based on a single-center cohort with standardized wound size.

Sichao Jiang, Qixuan Song, Junhuan Wang, Xiaohong Sun, Zhen Zhang, Shouyu Wang, Junwei Zong.

   Objective: To develop a machine learning (ML) model for predicting prolonged healing (>8 weeks) in diabetic wounds, focusing on dynamic C-reactive protein (CRP) trajectories.
Methods: This was a retrospective single-center cohort study. We included 465 patients with type 2 diabetes, standardized wound sizes (5-8 cm2), and debridement alone (2021-2024: training set, n = 325; 2025: temporal validation set, n = 140). Serial CRP was measured at admission (CRP), post-antibiotic preoperatively (CRP_2nd), and postoperatively at discharge (CRP_3rd). Therapeutic response variables (therapeutic_response_1/2/all) were calculated as percentage changes in serial CRP levels across treatment phases, reflecting anti-inflammatory/antimicrobial efficacy. LASSO regression selected features, 12 ML models were constructed, and performance was evaluated via AUC, sensitivity, and specificity. SHAP analysis interpreted predictions.
Results: The GradientBoosting model exhibited superior performance (validation set: accuracy = 0.9357, sensitivity = 0.8689, specificity = 0.9873). LASSO regression identified 15 key variables [including CRP_2nd, CRP_3rd, albumin (ALB)]. SHAP analysis revealed CRP_2nd as the most influential predictor (mean absolute SHAP value = 0.460), with elevated CRP_2nd/CRP_3rd associated with prolonged healing and higher ALB/favorable therapeutic responses as protective factors.
Conclusion: Dynamic CRP trajectories, particularly CRP_2nd, are critical for predicting prolonged diabetic wound healing. The GradientBoosting model provides a clinically actionable tool for risk stratification.

Keywords:  CRP dynamics; cohort analysis; diabetes mellitus; machine learning model; wound prognosis

DOI:  https://doi.org/10.3389/fmed.2026.1778003
Smart Health (Amst). 2026 Mar;pii: 100633. [Epub ahead of print]39

Patient-specific deep offline artificial pancreas for blood glucose regulation in type 1 diabetes.

Yixiang Deng, Kevin Arao, Christos S Mantzoros, George Em Karniadakis.

  Due to insufficient insulin secretion, patients with type 1 diabetes mellitus (T1DM) are prone to blood glucose fluctuations ranging from hypoglycemia to hyperglycemia. While dangerous hypoglycemia may lead to coma immediately, chronic hyperglycemia increases patients' risks for cardiorenal and vascular diseases in the long run. In principle, an artificial pancreas - a closed-loop insulin delivery system requiring patients to manually input insulin dosage according to the upcoming meals - could supply exogenous insulin to control the glucose levels and hence reduce the risks from hyperglycemia. However, insulin overdosing in some type 1 diabetic patients, who are physically active, can lead to unexpected hypoglycemia beyond the control of the common artificial pancreas. Therefore, it is important to take into account the glucose decrease due to physical exercise when designing the next-generation artificial pancreas. In this work, we develop a framework integrating systems biology-informed neural networks (SBINN), deep reinforcement learning (RL) algorithms, and T1DM data collected from wearable devices, to automate insulin dosing for patients. In particular, we build patient-specific computational models using SBINN to mimic the glucose-insulin dynamics for a few patients from the dataset, by simultaneously considering patient-specific carbohydrate intake and physical exercise intensity. Our patient-specific artificial pancreas, based on two deep RL algorithms, provided better insulin dosage, leading to safer glucose levels compared to those in the original dataset.

Keywords:  Artificial pancreas; Digital twin; Offline reinforcement learning; Physical exercise; Type 1 diabetes; Wearable devices

DOI:  https://doi.org/10.1016/j.smhl.2026.100633
Front Public Health. 2026 ;14 1756755

Application of AI and digital health tools in public health management of T2DM: from mechanism prediction to personalized treatment.

Chonger Yu.

  Type 2 diabetes mellitus (T2DM) poses a significant global public health challenge, with its prevalence escalating continuously and disproportionately affecting low- and middle-income countries (LMICs), imposing a substantial burden on healthcare systems. Traditional management models have limitations in disease prediction, personalized treatment, and public health intervention. Artificial intelligence (AI) and digital health technologies provide novel insights for precise prediction and intelligent management of T2DM. This review systematically summarizes research progress in AI's role in deciphering T2DM pathogenesis, personalized treatment, and public health management. By integrating multi-omics and environmental data, AI reveals key mechanisms including gene-environment (G × E) interactions, β-cell dysfunction, and inflammatory pathways, significantly enhancing early screening and risk prediction. In clinical management, AI combined with digital health tools [e.g., continuous glucose monitoring (CGM), wearable devices, and mobile health (mHealth) apps] facilitates remote monitoring, medication optimization, and personalized interventions, improving treatment adherence and health management efficiency. At the public health level, AI optimizes resource allocation and disease burden assessment, promoting chronic disease prevention and control model transformation. Future efforts should prioritize developing low-resource-adapted tools, strengthening data privacy protection tailored to LMICs, and addressing algorithmic fairness and the digital divide to ensure safe, equitable, and sustainable AI application in global T2DM management. Overall, AI and digital health integration is driving T2DM management towards an intelligent and precision-based era, with the potential to reduce disparities in LMICs.

Keywords:  artificial intelligence; data privacy; digital health; personalized medicine; public health management; type 2 diabetes mellitus

DOI:  https://doi.org/10.3389/fpubh.2026.1756755
NPJ Digit Med. 2026 Mar 05.

Autonomous AI-assisted diabetic retinopathy screening at primary care is associated with increased presentation to eye care by at risk patients.

Ariel Leong, Risa M Wolf, Roomasa Channa, Jiangxia Wang, Harold Lehmann, Michael D Abramoff, T Y Alvin Liu.

Adult patients with diabetes (n = 3745) seen at Johns Hopkins Medicine primary care sites were referred to the Wilmer Eye Institute either based on a primary care provider referral or autonomous AI diagnostic result (referral was made after a positive or non-diagnostic result). An inverse-probability-weighted regression, which incorporated propensity score matching on social determinants of health and relevant clinical variables, showed that implementation of an autonomous AI-assisted diabetic screening program in a primary care clinic was associated with increased presentation to eye care specialist care by African-Americans (p = 0.02). This is significant because African-Americans have traditionally been less likely to undergo annual screening exams and more likely to present with more severe forms of diabetic retinopathy (DR). The results suggest a potential association between office-based, AI-assisted DR screening and improved downstream ophthalmic access for African-American patients. However, given that the analysis was exploratory, this association should be interpreted cautiously and further validated.

DOI: https://doi.org/10.1038/s41746-026-02460-5
IEEE J Biomed Health Inform. 2026 Mar 04. PP

MVTT-GMamba: A Multimodal Graph Reasoning Framework for Anti-VEGF Efficacy Prediction in Diabetic Macular Edema.

Shijing Wu, Yifan Zheng, Tianyu Chen, Gang Wu, Qinkun Zhi, Ailing Sui, Haixia Bai, Junhui Shen, Li Zhang, Yao Wang, Jian Ma, Xiaoyun Fang, Yanming Zhu, Feiwei Qin, Zhiqing Chen.

Accurately predicting individual responses to Anti-Vascular Endothelial Growth Factor (Anti-VEGF) efficacy in diabetic macular edema (DME) remains a critical challenge in personalized ophthalmic care. Existing methods often rely on unimodal data or suffer from ineffective multimodal feature extraction and fusion, leading to modality redundancy and performance degradation. To address these limitations, we propose MVTT-GMamba, a novel multimodal learning framework that integrates optical coherence tomography (OCT) images and structured clinical indicators for early and accurate Anti-VEGF efficacy prediction. At the core of MVTT-GMamba is a feature-wise heterogeneous graph reasoning paradigm that explicitly models inter-patient and inter-feature relations, together with an adaptive, graph-guided prediction head that progressively anneals structural priors into the classifier. Building on this core, we adopt domain-tailored MambaVision and TabTransformer encoders and an early cross-attention fusion module to realize fine-grained multimodal representation learning. Extensive experiments on both a private clinical dataset (DMETHERA-ECSAHZU) and the public APTOS2021 benchmark demonstrate that MVTT-GMamba consistently outperforms state-of-the-art methods across all evaluation metrics. In addition, Grad-CAM visualizations reveal that the model attends to clinically relevant retinal regions, providing enhanced interpretability. Code is available at: https://github.com/DME666/DME.

DOI: https://doi.org/10.1109/JBHI.2026.3670360
Cureus. 2026 Jan;18(1): e102559

Grading Diabetic Retinopathy Using Comparative Assessment: A Pilot Study Comparing Paired Image Comparisons With Direct Grading.

Mohammed Al-Roubaie.

  Introduction Accurate grading of diabetic retinopathy is essential for effective screening, clinical decision-making, and evaluation of automated diagnostic systems. Conventional grading relies on categorical severity scales, which are subject to inter- and intra-observer variability, particularly among less-experienced or junior graders and in cases with subtle disease features. Comparative assessment using paired image comparisons may offer a complementary approach by reframing grading as a relative severity judgement and potentially reducing grading variability. Methods This pilot study evaluated retinal fundus photographs obtained from a publicly available dataset. Ninety images spanning the spectrum of diabetic retinopathy severity were graded using two approaches: direct grading according to the International Clinical Diabetic Retinopathy Severity Scale and comparative assessment using paired image comparisons. Both methods were performed twice by a junior clinician following structured training to assess repeatability. Classification performance for discrimination between the presence and absence of diabetic retinopathy was compared using confusion matrices and McNemar's test. Results Comparative assessment demonstrated higher overall accuracy and improved specificity compared with direct grading across repeated grading rounds, while maintaining high sensitivity. Paired image comparison showed greater consistency between grading attempts, whereas direct grading exhibited greater variability. Differences in classification performance between methods were statistically significant. Conclusion In this pilot study, comparative assessment using paired image comparisons outperformed conventional direct grading for discrimination between the presence and absence of diabetic retinopathy when applied by a junior grader. These findings suggest that relative severity judgement may represent a viable alternative or adjunct to traditional categorical grading systems, particularly in contexts where grading variability is a concern. Larger studies involving multiple graders and real-world screening images are required to validate these findings and define the clinical role of comparative assessment.

Keywords:  blind image quality assessment; confusion matrix; deep learning; diabetic retinopathy; fundus photography; image grading; mcnemar test; paired comparisons; tele-ophthalmology screening

DOI:  https://doi.org/10.7759/cureus.102559
Front Artif Intell. 2026 ;9 1684752

A review of optimization strategies for deep and machine learning in diabetic macular edema.

A M Mutawa, Khalid Sabti, Bibin Shalini Sundaram Thankaleela, Seemant Raizada.

  Diabetic macular edema (DME) is a primary contributor to visual impairment in diabetic patients, necessitating precise and prompt analysis for optimal treatment. Recent breakthroughs in deep learning (DL) and machine learning (ML) have yielded promising outcomes in ophthalmic image analysis. However, researchers often overlook the significance of optimization algorithms in enhancing the efficacy of their models for DME-related tasks. This review aims to consolidate, seek, discover, assess, and integrate existing work on the application of DL and ML, with emphasis on the integration and impact of optimization algorithms in enhancing their efficacy, robustness, and performance for DME in the fields of computer science and engineering. The population, intervention, comparison, and outcome framework was employed in this study to facilitate a clear and comprehensive analysis. The procedural superiority of the included investigations was evaluated using the Joanna Briggs Institute Critical Appraisal Tools for assessing methodological quality. The Auto-Metric Graph Neural Network achieved the greatest accuracy of 99.57% for combined diabetic retinopathy-DME grading, illustrating the higher efficacy of hybrid architectures augmented by meta-heuristic optimizers, such as Jaya and ant colony optimization. Successful deployment, however, depends on overcoming hurdles, such as the low mean average precision in lesion identification (0.1540) in YOLO-based models in the test set performance, and improved clinical interpretability to foster clinician trust. A Sankey diagram visually analyzes the flow of quantities between different entities of the survey.
Systematic review registration: B. (2025, November 2). A Review of Optimization Strategies for Deep and Machine Learning in DME. Retrieved from osf.io/qsh4j.

Keywords:  DME; deep learning; machine learning; optimization; soft computing; sunburst diagram

DOI:  https://doi.org/10.3389/frai.2026.1684752
Eur J Med Res. 2026 Mar 06.

Screening, validation, and transcriptional regulation analysis of oxidative stress-related biomarkers in gestational diabetes mellitus: SH3BP5, ITGAM, PRRG1, and MIS12.

Haiyu He, Wen Wu.

   BACKGROUND: Gestational diabetes mellitus (GDM) is a common pregnancy complication linked to adverse outcomes, highlighting the need for new diagnostic markers. This study aimed to identify oxidative stress-related genes as potential biomarkers for GDM using integrated bioinformatics and experimental validation.
METHODS: The GSE70493 dataset was obtained from the Gene Expression Omnibus (GEO) database and analyzed using weighted gene co-expression network analysis (WGCNA), functional enrichment, and differential expression analysis. Reactive oxygen species (ROS) activity scores for each sample were calculated using single-sample gene set enrichment analysis (ssGSEA). ROS-associated differentially expressed genes (DEGs) were further screened using the least absolute shrinkage and selection operator (LASSO), support vector machine-recursive feature elimination (SVM-RFE), and RandomForest algorithms to identify pivotal genes. A diagnostic radial-kernel support vector machine (SVM) classifier was constructed and rigorously evaluated through a 5 × 5 nested cross-validation framework on the training set, followed by validation in an independent external cohort (GSE249311). A transcription factor (TF)-gene regulatory network was established via the JASPAR database on the NetworkAnalyst 3.0 platform. The biological role of PRRG1 in GDM was also explored using cellular experiments.
RESULTS: WGCNA identified 7 co-expression modules, among which the green, pink, and black modules showed a strong positive correlation with ROS scores. Enrichment analysis showed that the module genes were mainly implicated in protein hydrolysis and processing, cell adhesion molecule binding, and various immune-related pathways. 765 DEGs, including 470 downregulated genes and 295 upregulated genes, were screened between GDM samples and control samples. Machine learning algorithm identified four hub genes (SH3BP5, ITGAM, PRRG1, and MIS12). When the four hub genes were combined, ROC curves showed that the hub genes exhibited strong diagnostic value for GDM. In GDM, SH3BP5, MIS12, and ITGAM were low-expressed, while PRRG1 was high-expressed. The TF-gene regulatory network showed that the hub genes could regulate multiple transcription factors separately. In vitro experiments demonstrated that PRRG1 knockdown significantly enhanced the viability, migration, and invasion of GDM cells.Table: Table [3, 4] was received; however, no citation was provided in the manuscript. Please provide the location of where to insert the citation in the main body of the text. Otherwise, kindly advise us on how to proceed. Please note that tables should be cited in ascending numerical order in the main body of the textDear editor, please consider removing the Tables 3 and 4 in the paper, since the data of molecular docking are absent in the current paper.
CONCLUSION: We provided four novel biomarkers targeting oxidative stress for the treatment of GDM.

Keywords:  Diagnostic biomarkers; Gestational diabetes mellitus; Oxidative stress; Transcription factor-gene regulatory networks

DOI:  https://doi.org/10.1186/s40001-026-03966-0
Front Endocrinol (Lausanne). 2026 ;17 1712085

An integrative analysis of cardiac autonomic neuropathy and nephropathy risk assessed with SUDOSCAN in individuals with type 2 diabetes.

Claudiu Cobuz, Mădălina Ungureanu-Iuga, Maricela Cobuz.

   Introduction: The use of non-invasive, rapid screening methods to detect diabetes mellitus complications, such as neuropathy, is a growing trend in modern medicine. This study aimed to investigate the relationship between SUDOSCAN-derived Cardiac Autonomic Neuropathy (CAN) and Nephropathy (Nephro) scores in individuals with type 2 diabetes mellitus and to evaluate the potential of artificial neural networks in predicting these scores.
Methods: A cross-sectional study was conducted, and 150 individuals were included in the statistical analysis to determine the risk of CAN and nephropathy in individuals with type 2 diabetes mellitus using the SUDOSCAN device. The relationships between SUDOSCAN-derived scores and covariate factors (age, sex, diabetes duration, and body mass index) were established through Spearman correlations, a general linear model, and an artificial neural network (ANN).
Results: The results indicated that individuals with diabetes are at higher risk of both cardiac autonomic neuropathy and nephropathy, which are strongly interconnected, mainly due to factors like age, BMI, and blood pressure rather than traditional glycemic markers. A strong inverse correlation was observed between CAN and nephropathy scores (r = -0.83, p < 0.05), highlighting a shared mechanism such as endothelial dysfunction and metabolic stress. The CAN score model showed slightly better predictive performance (RMSE 5.36, MAE 4.11) than the nephropathy model (RMSE 5.91, MAE 7.55), while artificial neural networks achieved outstanding classification performance (AUC ≥ 0.97).
Discussion: When used together, the highly sensitive CAN model can be employed for initial screening to prevent missing cases, while the highly-specific Nephro model can confirm risk and minimize false positives, thereby creating an optimal two-step risk stratification strategy. Thus, ANN-based systems can assist clinicians in guiding decisions by prioritizing individuals for further testing, tailoring treatments, and optimizing follow-up care in diabetic nephropathy.

Keywords:  artificial neural network; diabetes mellitus; electrochemical skin conductance; prediction; type 2 diabetes

DOI:  https://doi.org/10.3389/fendo.2026.1712085
J Diabetes Sci Technol. 2026 Feb 28. 19322968261424270

Evaluating Accuracy of ChatGPT-4o in Automated Carbohydrate Estimation From Images as a Self-Management Tool for Adolescents With Type 1 Diabetes.

Asta Risak Johansen, Isabella Kjær Laursen, Vár Jacobsen, Zacharias Henriksson Møller, Simon Lebech Cichosz.

   INTRODUCTION: Type 1 diabetes mellitus (T1D) requires precise carbohydrate estimation to manage blood glucose and prevent chronic and acute complications to hyperglycemia or hypoglycemia. This study evaluates the accuracy of ChatGPT in estimating carbohydrate content in images of meals, compared with the considered gold standard of manually counting carbohydrates.
METHOD: Carbohydrate content of 60 fruits and vegetables (F&V) and 60 composite meals was manually counted as the reference standard. Images (n = 240), with and without a size reference, were uploaded to ChatGPT-4o with a standardized prompt in separate sessions. ChatGPT's estimates were then compared with the manual counts to assess accuracy.
RESULTS: The performance of ChatGPT-4o compared with the manual calculation was assessed primarily using mean absolute error, percentage of agreement (PoA), and Bland-Altman analysis. ChatGPT-4o achieved a PoA of 93.3% for F&V's estimates, increasing to 95% with a size reference, while composite meal estimates yielded a PoA of 46.7%, reducing to 43.3% with a size reference, based on a ±10 g carbohydrates limit. Bland-Altman analysis showed a slight bias tendency in both ChatGPT-4o's estimates of F&V and composite meals with a size reference. ChatGPT-4o's estimate for F&V and composite meals without a size reference exhibited a systematic bias, with both overestimation and underestimation of the carbohydrate content.
CONCLUSION: This study suggests that adolescents living with T1D should employ ChatGPT-4o for carbohydrate estimating with caution. ChatGPT-4o showed inaccuracies in its application to composite meals, increasing the risk of inaccurate insulin administration and potentially causing postprandial hyperglycemia or hypoglycemia.

Keywords:  ChatGPT; artificial intelligence; carbohydrate counting; postprandial hyperglycemia and hypoglycemia; primary health care; type 1 diabetes

DOI:  https://doi.org/10.1177/19322968261424270