bims-aukdir Biomed News
on Automated knowledge discovery in diabetes research
Issue of 2025–06–22
fifteen papers selected by
Mott Given



  1. Ophthalmol Sci. 2025 Sep-Oct;5(5):5(5): 100755
       Objective: To develop and validate an artificial intelligence (AI)-based system, Diabetic Retinopathy Analysis Model Assistant (DRAMA), for diagnosing diabetic retinopathy (DR) across multisource heterogeneous datasets and aimed at improving the diagnostic accuracy and efficiency.
    Design: This was a cross-sectional study conducted at Zhejiang University Eye Hospital and approved by the ethics committee.
    Subjects: The study included 1500 retinal images from 957 participants aged 18 to 83 years. The dataset was divided into 3 subdatasets: color fundus photography, ultra-widefield imaging, and portable fundus camera. Images were annotated by 3 experienced ophthalmologists.
    Methods: The AI system was built using EfficientNet-B2, pretrained on the ImageNet dataset. It performed 11 multilabel tasks, including image type identification, quality assessment, lesion detection, and diabetic macular edema (DME) detection. The model used LabelSmoothingCrossEntropy and AdamP optimizer to enhance robustness and convergence. The system's performance was evaluated using metrics such as accuracy, sensitivity, specificity, and area under the curve (AUC). External validation was conducted using datasets from different clinical centers.
    Main Outcome Measures: The primary outcomes measured were the accuracy, sensitivity, specificity, and AUC of the AI system in diagnosing DR.
    Results: After excluding 218 poor-quality images, DRAMA demonstrated high diagnostic accuracy, with EfficientNet-B2 achieving 87.02% accuracy in quality assessment and 91.60% accuracy in lesion detection. Area under the curves were >0.95 for most tasks, with 0.93 for grading and DME detection. External validation showed slightly lower accuracy in some tasks but outperformed in identifying hemorrhages and DME. Diabetic Retinopathy Analysis Model Assistant diagnosed the entire test set in 86 ms, significantly faster than the 90 to 100 minutes required by humans.
    Conclusions: Diabetic Retinopathy Analysis Model Assistant, an AI-based multitask model, showed high potential for clinical integration, significantly improving the diagnostic efficiency and accuracy, particularly in resource-limited settings.
    Financial Disclosures: The author(s) have no proprietary or commercial interest in any materials discussed in this article.
    Keywords:  Artificial intelligence; Deep learning; Diabetic retinopathy; Multisource heterogeneous dataset; Multitask learning
    DOI:  https://doi.org/10.1016/j.xops.2025.100755
  2. Cureus. 2025 May;17(5): e84314
      Gestational diabetes mellitus (GDM) significantly increases the risk of developing type 2 diabetes (T2D) postpartum. Early identification of high-risk women using machine learning (ML) models could enable targeted interventions and improve outcomes. This systematic review aims to evaluate the performance, predictive features, and methodological quality of ML models designed to predict the transition from GDM to T2D. A comprehensive search was conducted across PubMed, Scopus, IEEE Xplore, and Web of Science, yielding 178 records. After removing duplicates and screening for eligibility, 13 studies were included. Data on study characteristics, ML algorithms, predictive features, model performance, and validation methods were extracted. Risk of bias was assessed using the PROBAST (Prediction model Risk of Bias Assessment Tool). The included studies demonstrated variable performance, with area under the curve (AUC) values ranging from 0.72 to 0.92. Models incorporating omics data outperformed clinical-only models. Key predictive features included age, b ody m ass i ndex (BMI), glycemic measures, and pregnancy-specific factors. However, only 38% of studies employed robust external validation, and small sample sizes limited generalizability in some cases. Risk of bias assessment revealed low overall bias, though analytical validation methods were often unclear or insufficient. ML models, particularly those integrating omics data, show strong potential for predicting T2D risk in women with prior GDM. However, heterogeneity in validation methods and limited external validation highlight the need for standardized reporting and larger, diverse cohorts to enhance clinical applicability. Future research should focus on developing reproducible, generalizable models to guide personalized prevention strategies.
    Keywords:  gestational diabetes mellitus; machine learning; prediction models; systematic review; type 2 diabetes
    DOI:  https://doi.org/10.7759/cureus.84314
  3. Front Med (Lausanne). 2025 ;12 1542860
       Background: Machine learning technology that uses available clinical data to predict diabetic retinopathy (DR) can be highly valuable in medical settings where fundus cameras are not accessible.
    Objective: This study aimed to develop and compare machine learning algorithms for predicting DR without fundus image.
    Methods: We used data from Korea National Health and Nutrition Examination Survey (2008-2012 and 2017-2021) and enrolled individuals aged ≥ 20 years with diabetes who received fundus examination. Predictive models for DR were developed using logistic regression and three machine learning algorithms: extreme gradient boosting, decision tree, and random forest. Model performance was evaluated using area under the receiver operating characteristic curve (AUC) and accuracy for the diagnosis of DR, and feature importance was determined using Shapley Additive Explanations (SHAP).
    Results: Among the 3,026 diabetic participants (male, 50.7%; mean age, 63.7 ± 10.5 years), 671 (22.2%) had DR. The random forest model, using 16 variables, achieved the highest AUC of 0.748 (95% confidence interval, 0.705-0.790) with a sensitivity 0.669, specificity of 0.729 and an accuracy of 0.715. As interpreted by SHAP, HbA1c, fasting glucose levels, duration of diabetes, and body mass index were identified as common key determinants influencing the model's outcomes.
    Conclusion: The DR prediction models using machine learning techniques demonstrated reliable performance even without fundus imaging, with the random forest model showing particularly strong results. These models could assist in managing DR by identifying high-risk patients, enabling timely ophthalmic referrals.
    Keywords:  Korea; diabetic retinopathy; machine learning; prediction; random forest algorithms
    DOI:  https://doi.org/10.3389/fmed.2025.1542860
  4. Clin Exp Ophthalmol. 2025 Jun 14.
       BACKGROUND: Diabetic retinopathy is a leading cause of preventable blindness worldwide. Meanwhile, artificial intelligence is rapidly growing in clinical utility within medicine. This scoping review aims to identify and summarise existing literature on the barriers and enablers of clinical applications of artificial intelligence systems for the screening of diabetic retinopathy.
    METHODS: Utilising a systematic approach and the PRISMA-ScR protocol for conducting scoping reviews, searches were performed in MEDLINE, Embase, Emcare, Cochrane, CINAHL, ProQuest, Scopus and grey literature (Australian Indigenous Health InfoNet). Two reviewers independently reviewed the records. A third reviewer provided consensus. Data extraction and synthesis in narrative form ensued.
    RESULTS: A total of 3844 articles were screened, of which 18 were selected. Published between 2018 and 2023, the selected studies varied in study design and were conducted across 10 countries. Several barriers and enablers were identified and categorised into four domains: healthcare system, healthcare professional, healthcare user and information technology. Within the healthcare system, clinical efficiency was reported on most frequently. Concerning the healthcare professional, education was most frequently discussed. Within healthcare user, studies most frequently identified factors pertaining to patient outcomes, while diagnostic performance was most frequently explored under the information technology domain.
    CONCLUSIONS: As evidence for the efficacy of artificial intelligence for diabetic retinopathy screening grows, barriers to and enablers for its uptake in clinical practice are paramount considerations. Translating the knowledge of systems, provider, consumer and technological factors informs clinical strategies, ultimately facilitating the sustainable and effective implementation of this novel technology for screening practices.
    Keywords:  artificial intelligence; barriers; diabetic retinopathy; enablers; implementation science
    DOI:  https://doi.org/10.1111/ceo.14567
  5. Glob Epidemiol. 2025 Jun;9 100209
       Background: Timely identification and treatment of Diabetic Retinopathy (DR) is critical in avoiding vision loss. DR screening is challenging, especially in resource-limited areas where trained ophthalmologists are scarce. AI solutions show promise in addressing this challenge. In this study, the performance metrics of an AI solution (MadhuNetrAI) developed in India was evaluated for referring and grading DR.
    Methods: MadhuNetrAI was developed de novo by the All India Institute of Medical Sciences (AIIMS) and Wadhwani AI (WIAI). It was tested on 1078 fundus images (from AIIMS Delhi and an unannotated subset of publicly available EyePACS images) against two ophthalmologists and an adjudicator serving as independent gold-standard annotators, wherein the disease status of the patients remained unknown.
    Findings: MadhuNetrAI demonstrated high sensitivity (93·2 %; CI: 89·5 %-95·6 %) and specificity (95·3 %; CI: 93·7 %-96·6 %) in detecting referable DR (moderate, severe, proliferative DR). The area-under-the-curve for referring DR against the gold standard was 0·97 (CI: 0·95-0·99) indicating excellent diagnostic performance. The agreement in grading DR severity was high (kappa = 0·89, CI: 0·86-0·91). The model performed comparably in detecting DR too.
    Interpretation: MadhuNetrAI's ability to grade DR severity and identify referrable cases could bring DR patients to care much earlier. Further research and clinical trials are needed to ensure its reliability and generalizability across diverse populations and image qualities.
    Funding: MadhuNetrAI was developed by technical and programmatic teams at WIAI, with inputs and contributions by the clinical team at AIIMS, and funded by USAID. The authors have no financial or non-financial conflicts of interest to disclose.
    DOI:  https://doi.org/10.1016/j.gloepi.2025.100209
  6. Diabetol Metab Syndr. 2025 Jun 18. 17(1): 227
       BACKGROUND: Diabetes mellitus, a global health concern with severe complications, demands early detection and precise staging for effective management. Machine learning approaches, combined with bioinformatics, offer promising avenues for enhancing diagnostic accuracy and identifying key biomarkers.
    METHODS: This study employed a multi-class classification framework to classify patients across four health states: healthy, prediabetes, type 2 Diabetes Mellitus (T2DM) without complications, and T2DM with complications. Three models were developed using molecular markers, biochemical markers, and a combined model of both. Five machine learning classifiers were applied: Random Forest (RF), Extra Tree Classifier, Quadratic Discriminant Analysis, Naïve Bayes, and Light Gradient Boosting Machine. To improve the robustness and precision of the classification, Recursive Feature Elimination with Cross-Validation (RFECV) and a fivefold cross-validation were used. The multi-class classification approach enabled effective discrimination between the four diabetes stages.
    RESULTS: The top contributing features identified for the combined model through RFECV included three molecular markers-miR342, NFKB1, and miR636-and two biochemical markers the albumin-to-creatinine ratio and HDLc, indicating their strong association with diabetes progression. The Extra Trees Classifier achieved the highest performance across all models, with an AUC value of 0.9985 (95% CI: [0.994-1.000]). This classifier outperformed other models, demonstrating its robustness and applicability for precise diabetes staging.
    CONCLUSION: These findings underscore the value of integrating machine learning with molecular and biochemical markers for the accurate classification of diabetes stages, supporting a potential shift toward more personalized diabetes management.
    Keywords:  Diabetes mellitus; Extra tree classifier; Machine learning; RNA; T2DM
    DOI:  https://doi.org/10.1186/s13098-025-01786-6
  7. J Med Ultrasound. 2025 Apr-Jun;33(2):33(2): 116-124
       Background: Studies have demonstrated that a qualitatively and quantitatively assessed hyperechoic deltoid muscle on ultrasound (US) was accurate for the earlier detection of type 2 diabetes (T2D). We aim to demonstrate the utility of automated skeletal muscle US radiomics and machine learning for the earlier detection of T2D and prediabetes (PreD) as a supplement to traditional hemoglobin A1c (HbA1c) testing.
    Methods: A sample of 1191 patients who underwent shoulder US was collected with five cohorts: 171 "normal" (without T2D), 69 "screening" (negative pre-US, but positive HbA1c post-US), 190 "risk" (negative, but clinically high-risk and referred for HbA1c), 365 with "PreD" (pre-US), and 396 with "diabetes" (pre-US). Analysis was performed on deltoid muscle US images. Automatic detection identified the deltoid region of interest. Radiomics features, race, age, and body mass index were input to a gradient-boosted decision tree model to predict if the patient was either low-risk or moderate/high-risk for T2D.
    Results: Combining selected radiomics and clinical features resulted in a mean area under the receiver operating characteristic (AUROC) of 0.86 with 71% sensitivity and 96% specificity. In a subgroup of only patients with obesity, combining radiomics and clinical features achieved an AUROC of 0.92 with 82% sensitivity and 95% specificity.
    Conclusion: US radiomics and machine learning yielded promising results for the detection of T2D using skeletal muscle. Given the increasing use of shoulder US and the increasingly high number of undiagnosed patients with T2D, skeletal muscle US and radiomics analysis has the potential to serve as a supplemental noninvasive screening tool for the opportunistic earlier detection of T2D and PreD.
    Keywords:  Deltoid muscle; diabetic myopathy; machine learning; muscle insulin resistance; musculoskeletal ultrasound; myosteatosis; prediabetes; radiomics; skeletal muscle; type 2 diabetes mellitus
    DOI:  https://doi.org/10.4103/jmu.jmu_12_24
  8. Front Cell Dev Biol. 2025 ;13 1603958
       Purpose: To develop a machine learning model to predict anatomical response to anti-VEGF therapy in patients with diabetic macular edema (DME).
    Methods: This retrospective study included patients with DME who underwent intravitreal anti-VEGF treatment between January 2023 and February 2025. Baseline data included optical coherence tomography (OCT) features and blood-based metabolic and hematologic markers. The primary outcome was defined as a ≥20% reduction in central retinal thickness (CRT) post-treatment. Feature selection was performed using univariate logistic regression and LASSO regression. Five machine learning algorithms-logistic regression, decision tree, multilayer perceptron, random forest, and support vector machine-were trained and validated. Model performance was evaluated using accuracy, sensitivity, specificity, Area Under the Receiver Operating Characteristic Curve (AUC), and decision curve analysis. The best-performing model was further interpreted using SHAP analysis, and a nomogram was constructed for clinical application.
    Results: Among the 37 baseline variables, five key predictors were identified: preoperative CRT >400 μm, presence of retinal edema, presence of subretinal fluid (SRF), disorganization of the inner retinal layers (DRIL), and ellipsoid zone (EZ) integrity. The logistic regression model achieved the best performance with an accuracy of 0.83, sensitivity of 0.85, specificity of 0.79, and an AUC of 0.90 (95% CI: 0.81-0.99). SHAP analysis revealed that preoperative retinal edema, DRIL, SRF, and CRT had the strongest positive contributions, while intact EZ was a negative predictor of CRT reduction. A nomogram was developed to facilitate individualized clinical decision-making.
    Conclusion: We successfully developed a predictive model for anatomical response to anti-VEGF therapy in DME patients. The model identified key features associated with treatment outcomes, providing a valuable tool for personalized therapeutic planning. Further validation in multicenter cohorts is warranted to confirm generalizability and enhance model robustness.
    Keywords:  anti-vegf; diabetic macular edema; diabetic retinopathy; nomogram; predictive model
    DOI:  https://doi.org/10.3389/fcell.2025.1603958
  9. Front Endocrinol (Lausanne). 2025 ;16 1614657
       Background: Diabetic peripheral neuropathy (DPN) is a common and debilitating complication of type 2 diabetes mellitus (T2DM), significantly impacting patients' quality of life and increasing healthcare burdens. Early prediction and intervention are critical to mitigating its impact.
    Methods: This study analyzed 1,544 diabetic patients from the First Affiliated Hospital of Shandong First Medical University, who were randomly divided into a training cohort (n = 1,082) and a testing cohort (n = 462) using a 7:3 split ratio. Feature selection was performed using both Boruta and LASSO algorithms, and the intersection of the selected variables was used as the final predictor set. Eight key predictors were identified from 23 variables, including diabetes duration, uric acid, HbA1c, NLR, smoking status, SCR, LDH, and hypertension. Nine machine learning models were developed and compared for DPN risk prediction.
    Results: Stochastic Gradient Boosting (SGBT) demonstrated the best performance (training AUC: 0.933, 95% CI: 0.921-0.946; testing AUC: 0.811, 95% CI: 0.776-0.843). Shapley Additive Explanations (SHAP) analysis provided interpretability, highlighting the clinical importance of diabetes duration and HbA1c among other predictors.
    Conclusion: This study establishes a robust predictive tool for early DPN detection, laying the foundation for improved prevention and management strategies.
    Keywords:  clinical data; diabetic peripheral neuropathy; interpretable; machine learning; risk prediction model
    DOI:  https://doi.org/10.3389/fendo.2025.1614657
  10. J Phys Chem Lett. 2025 Jun 15. 6355-6363
      Accurate and early detection of C-peptide, a stable biomarker indicative of diabetes, is crucial for disease diagnosis, treatment, and prevention. This study explores a novel detection methodology using solid-state nanopore technology coupled with machine learning for the sensitive identification of C-peptide molecules. Solid-state nanopores were fabricated via focused ion beam milling and systematically tested to analyze ionic current blockade characteristics of C-peptide and fetal bovine serum during its translocation. A comprehensive five-dimensional signal analysis incorporating current blockade amplitude, dwell time, standard deviation, kurtosis, and skewness significantly enhanced the discriminative capability of the nanopore sensor. Employing Support Vector Machine classification with a radial basis function kernel, the proposed platform achieved an outstanding identification accuracy of 99.63% for distinguishing C-peptide events from serum background. These results highlight the potential of solid-state nanopore technology integrated with advanced machine learning as a rapid, sensitive, and portable platform for early diabetes diagnostics and continuous biomarker monitoring.
    DOI:  https://doi.org/10.1021/acs.jpclett.5c01281
  11. Endocr Pract. 2025 Jun 16. pii: S1530-891X(25)00924-3. [Epub ahead of print]
      Automated insulin delivery (AID) systems have revolutionized diabetes care by integrating continuous glucose monitoring (CGM), insulin pumps, and advanced algorithms to improve glycemic outcomes and reduce user burden. Early commercial AID systems were developed with a conservative approach, prioritizing safety and regulatory approval over full automation or extensive customization. While these systems significantly improved diabetes management, they still face limitations, including incomplete automation, accessibility barriers, and the need for better adaptation to diverse user needs and lifestyles. These challenges are catalyzing development of next-generation AID technologies with a focus on achieving full automation, greater personalization, and broader accessibility. This review examines key limitations of current AID systems and explores future directions, including fully closed-loop control, novel insulin formulations, multi-hormonal systems, advanced sensor technologies, and integration of wearable and artificial intelligence (AI) tools. By addressing these challenges, future AID systems have the potential to deliver better effectiveness and equity in diabetes care for all individuals requiring insulin therapy.
    Keywords:  Artificial Intelligence; Artificial Pancreas; Automated Insulin Delivery (AID); Closed-loop; Continuous Glucose Monitoring; Diabetes Technology; Hybrid closed-loop; Insulin Pump
    DOI:  https://doi.org/10.1016/j.eprac.2025.05.752
  12. Am J Transl Res. 2025 ;17(5): 3951-3960
       BACKGROUND: Diabetes is a chronic condition that significantly impacts the cardiovascular system and various other organs. Photoplethysmogram (PPG) signals have been shown to correlate with variations in vascular blood flow and the presence of atherosclerosis. To effectively explore the complex nonlinear relationship between PPG signals and diabetes, we propose an automatic detection model based on the fusion of PPG features.
    METHODS: The proposed model consists of two main components: 1. Dynamic Fusion Feature Extraction: Short PPG signal window segments are processed using the SGR spatial encoding algorithm to extract dynamic fusion features. 2. Feature Representation Learning: Multi-scale convolutional layers (MCNN) are employed to learn feature representations, while the Vision Transformer (ViT) model is utilized to capture global contextual semantic features.
    RESULTS: The model was trained and validated on a self-collected medical dataset. The experimental results demonstrate that the classification model, which integrates short time window information, significantly improves detection performance. Specifically, the multi-period sequence input model achieves an accuracy of 91.11%, with a Receiver Operating Characteristic (ROC) curve area of 0.9341, indicating strong diagnostic capability.
    CONCLUSION: This study is a retrospective case-control study that collected clinical data from three groups of people: those with normal glucose levels, those with poorly controlled diabetes, and those with well-controlled diabetes. The study aims to utilize deep learning algorithms for the early prevention and screening of diabetes.
    Keywords:  MCNN; PPG; deep learning; diabetes; transformer
    DOI:  https://doi.org/10.62347/ZRMW1346
  13. Endocrinol Metab (Seoul). 2025 Jun 18.
       Background: The long-term association between adrenal gland volume (AGV) and type 2 diabetes (T2D) remains unclear. We aimed to determine the association between deep learning-based AGV and current glycemic status and incident T2D.
    Methods: In this observational study, adults who underwent abdominopelvic computed tomography (CT) for health checkups (2011-2012), but had no adrenal nodules, were included. AGV was measured from CT images using a three-dimensional nnU-Net deep learning algorithm. We assessed the association between AGV and T2D using a cross-sectional and longitudinal design.
    Results: We used 500 CT scans (median age, 52.3 years; 253 men) for model development and a Multi-Atlas Labeling Beyond the Cranial Vault dataset for external testing. A clinical cohort included a total of 9708 adults (median age, 52.0 years; 5,769 men). The deep learning model demonstrated a dice coefficient of 0.71±0.11 for adrenal segmentation and a mean volume difference of 0.6± 0.9 mL in the external dataset. Participants with T2D at baseline had a larger AGV than those without (7.3 cm3 vs. 6.7 cm3 and 6.3 cm3 vs. 5.5 cm3 for men and women, respectively, all P<0.05). The optimal AGV cutoff values for predicting T2D were 7.2 cm3 in men and 5.5 cm3 in women. Over a median 7.0-year follow-up, T2D developed in 938 participants. Cumulative T2D risk was accentuated with high AGV compared with low AGV (adjusted hazard ratio, 1.27; 95% confidence interval, 1.11 to 1.46).
    Conclusion: AGV, measured using deep learning algorithms, is associated with current glycemic status and can significantly predict the development of T2D.
    Keywords:  Adrenal glands; Deep learning; Diabetes mellitus, type 2; Radiographic image interpretation, computer-assisted
    DOI:  https://doi.org/10.3803/EnM.2025.2336
  14. J Diabetes Metab Disord. 2025 Dec;24(2): 151
      Diabetes is known as a chronic illness with severe consequences. The rising morbidity rates predict a stunning growth in the global diabetes population, approaching 642 million by 2040, implying that one out of every ten people will be affected. This worrying number highlights the critical need for collaborative efforts from industry and academics to accelerate innovation and foster growth in diabetes risk prediction, eventually saving lives. As the frequency of life-threatening diseases, such as diabetes, rises, Medical Decision Support Systems (MDSS) continue to prove their usefulness in supporting healthcare professionals, particularly physicians, in clinical decision-making procedures. Due to the advancement of technology, machine-learning techniques have made headlines in the early prediction of diabetes. In this paper, we employed machine learning techniques and the Analysis of Variance (ANOVA) method to explore associations between regional body fat distribution and diabetes mellitus in a community adult population, aiming to assess predictive capabilities. We used individual standard classifiers and ensemble learning methods to conduct a retrospective analysis of a portion of data based on body composition. To address the class imbalance problem in the target variable, we also applied three oversampling methods to provide more accurate predictions via learning algorithms. The results demonstrate that XGBoost, based on the Adaptive Synthetic Sampling (ADASYN) method, outperforms the state-of-the-art by achieving an accuracy value of 92.04%. This model exhibits more effectiveness for diabetes prediction compared to other models.
    Keywords:  ADASYN; ANOVA; Diabetes prediction; Machine learning
    DOI:  https://doi.org/10.1007/s40200-025-01661-1