bims-aukdir Biomed News
on Automated knowledge discovery in diabetes research
Issue of 2025–07–27
twelve papers selected by
Mott Given



  1. Endocr Pract. 2025 Jul 17. pii: S1530-891X(25)00966-8. [Epub ahead of print]
      Artificial intelligence (AI) is rapidly transforming clinical medicine, and its potential impact on diabetes care is particularly noteworthy. In recent years, both traditional machine learning approaches and deep learning algorithms have been applied to improve screening for complications of diabetes such as retinopathy, macular edema, and neuropathy, predict disease progression risk, and enhance clinical decision support systems for diagnosis, prognosis, and treatment optimization. AI-driven solutions are also emerging to identify noninvasive biomarkers for detecting diabetes and prediabetes, analyze the macronutrient content of meals using image-based deep learning methods, integrate novel risk prediction tools within electronic health records, and optimize automated insulin delivery (AID) systems. These advancements hold promise for streamlining patient care, personalizing treatment plans, and ultimately improving clinical outcomes. In this narrative review, we examine the latest AI applications in diabetes care, exploring their capabilities, limitations, and the future directions necessary to realize their full potential to improve the care of people living with diabetes.
    Keywords:  Artificial Intelligence; Automated Insulin; Continuous Glucose Monitoring; Delivery; Diabetes; Machine Learning; Technology
    DOI:  https://doi.org/10.1016/j.eprac.2025.07.008
  2. Comput Biol Med. 2025 Jul 18. pii: S0010-4825(25)01066-2. [Epub ahead of print]196(Pt B): 110715
      Diabetes is one of the most common diseases worldwide and requires accurate diagnosis. Patients with diabetes are often affected by diabetic retinopathy (DR), which can lead to low vision, vision loss, or blindness. Therefore, a robust computer-aided diagnosis system is needed to provide better treatment to patients. This review mainly focuses on the works related to diagnosing DR from retinal fundus images. A total of 128 research papers have been reviewed from 1986 to 2025. The survey is divided into two parts: one for the grading/classification of DR and the other for DR lesions segmentation. This survey article introduces the details of eye diseases, followed by the background details of DR and different imaging techniques required to diagnose DR, like fundus imaging, multifocal electroretinogram, and optical coherence tomography. Details of well-known DR datasets since 2009 are also provided, including their complete statistical information and potential dataset biases. Furthermore, the approaches used for grading and segmentation tasks from the early 1980s to recent developments are discussed. The reviewed papers are based on traditional and deep learning based methods used in DR diagnosis. In traditional methods, the researchers used image preprocessing, mathematical morphology, fuzzy system, active contour, features extraction methods, evolutionary approaches, and machine learning based classifiers. In deep learning, researchers have used convolutional neural network (CNN), long short-term memory, vision transformer, contrastive learning, federated learning, and Explainable Artificial Intelligence (XAI) based approaches for diagnosis. In this article, we have emphasized almost all the significant work done in diagnosing DR disease, the datasets used, and performance of methods on those datasets. The comparative analysis of the methods is also done to help researchers obtain future directions for further research in the area of medical disease identification, especially DR disease detection. The challenges of AI and its associated ethical implications are also discussed in the article to provide direction for future work.
    Keywords:  Classification; Deep learning; Diabetic retinopathy; Machine learning; Segmentation
    DOI:  https://doi.org/10.1016/j.compbiomed.2025.110715
  3. Am J Lifestyle Med. 2025 Jul 17. 15598276251359185
      Objective: To examine the applications of artificial intelligence (AI) in lifestyle medicine focused on diabetes care as a narrative review. Methods: Relevant keywords were identified and searched using PubMed to find relevant studies on AI in diabetes lifestyle management. Results: AI applications in diabetes care were divided into four primary categories: 1- predictive models for diabetes risk and complications, which can utilize random forest and deep learning, demonstrating high accuracy rates (>80%); 2- personalized lifestyle recommendations, which can utilize clustering techniques and causal forest analysis to adapt interventions, leading to enhanced glycemic control and weight reduction; 3- remote monitoring and self-management tools, which can utilize digital twin technology and machine learning for behavior modeling, showing improved patient adherence and clinical results; and 4- clinical decision support systems, which assess various data sources for enhanced diagnosis and treatment suggestions. Conclusion: AI technologies show significant potential in improving diabetes care through multiple modalities that offer scalable cost-effective solutions, improved patient outcomes, and more efficient resource distribution.
    Keywords:  artificial intelligence; clinical decision support; deep learning; diabetes care; lifestyle medicine; machine learning; personalized interventions
    DOI:  https://doi.org/10.1177/15598276251359185
  4. Artif Intell Med. 2025 Jul 19. pii: S0933-3657(25)00156-3. [Epub ahead of print]168 103221
      Diabetic retinopathy (DR) is a leading cause of blindness worldwide, requiring early detection and accurate grading for effective intervention. Advances in artificial intelligence (AI), computer vision, machine learning, and deep learning (DL) have enabled automated detection and classification of DR through various imaging modalities. This review comprehensively evaluates 91 studies employing AI-based methods in the detection and classification of DR using fundus color photography, optical coherence tomography (OCT), OCT-angiography (OCTA), and fundus fluorescein angiography, providing a holistic understanding of their strengths, challenges, and limitations. Additionally, this review compares the characteristics of 23 public datasets for DR. Across modalities, DL approaches generally outperform traditional methods. Among the studies reviewed, 81% utilized fundus images, followed by 9% using OCT, 6% using OCTA, and 2% incorporating multiple modalities. Regarding classification tasks, 62% used AI for multi-way classification, 28% for binary classification, and 10% incorporated both. The paper concludes with future directions, including explainable AI frameworks, multimodal data integration, and suggested protocols to integrate into existing healthcare workflows.
    Keywords:  Artificial intelligence; Computer vision; Deep learning; Diabetic retinopathy; Machine learning; Retinal imaging modalities
    DOI:  https://doi.org/10.1016/j.artmed.2025.103221
  5. Ewha Med J. 2025 Apr;48(2): e32
       Purpose: This study developed and evaluated a feature-based ensemble model integrating the synthetic minority oversampling technique (SMOTE) and random undersampling (RUS) methods with a random forest approach to address class imbalance in machine learning for early diabetes detection, aiming to improve predictive performance.
    Methods: Using the Scikit-learn diabetes dataset (442 samples, 10 features), we binarized the target variable (diabetes progression) at the 75th percentile and split it 80:20 using stratified sampling. The training set was balanced to a 1:2 minority-to-majority ratio via SMOTE (0.6) and RUS (0.66). A feature-based ensemble model was constructed by training random forest classifiers on 10 two-feature subsets, selected based on feature importance, and combining their outputs using soft voting. Performance was compared against 13 baseline models, using accuracy and area under the curve (AUC) as metrics on the imbalanced test set.
    Results: The feature-based ensemble model and balanced random forest both achieved the highest accuracy (0.8764), followed by the fully connected neural network (0.8700). The ensemble model had an excellent AUC (0.9227), while k-nearest neighbors had the lowest accuracy (0.8427). Visualizations confirmed its superior discriminative ability, especially for the minority (high-risk) class, which is a critical factor in medical contexts.
    Conclusion: Integrating SMOTE, RUS, and feature-based ensemble learning improved classification performance in imbalanced diabetes datasets by delivering robust accuracy and high recall for the minority class. This approach outperforms traditional resampling techniques and deep learning models, offering a scalable and interpretable solution for early diabetes prediction and potentially other medical applications.
    Keywords:  Area under curve; Computer neural networks; Deep learning; Diabetes mellitus; Random forest
    DOI:  https://doi.org/10.12771/emj.2025.00353
  6. PLoS One. 2025 ;20(7): e0326579
      Deep learning models for diagnostic applications require large amounts of sensitive patient data, raising privacy concerns under centralized training paradigms. We propose FedGAN, a federated learning framework for synthetic medical image generation that combines Generative Adversarial Networks (GANs) with cross-silo federated learning. Our approach pretrains a DCGAN on abdominal CT scans and fine-tunes it collaboratively across clinical silos using diabetic retinopathy datasets. By federating the GAN's discriminator and generator via the Federated Averaging (FedAvg) algorithm, FedGAN generates high-quality synthetic retinal images while complying with HIPAA and GDPR. Experiments demonstrate that FedGAN achieves a realism score of 0.43 (measured by a centralized discriminator). This work bridges data scarcity and privacy challenges in medical AI, enabling secure collaboration across institutions.
    DOI:  https://doi.org/10.1371/journal.pone.0326579
  7. World J Diabetes. 2025 Jul 15. 16(7): 104789
       BACKGROUND: Diabetic foot ulcer (DFU) is a serious and destructive complication of diabetes, which has a high amputation rate and carries a huge social burden. Early detection of risk factors and intervention are essential to reduce amputation rates. With the development of artificial intelligence technology, efficient interpretable predictive models can be generated in clinical practice to improve DFU care.
    AIM: To develop and validate an interpretable model for predicting amputation risk in DFU patients.
    METHODS: This retrospective study collected basic data from 599 patients with DFU in Beijing Shijitan Hospital between January 2015 and June 2024. The data set was randomly divided into a training set and test set with fivefold cross-validation. Three binary variable models were built with the eXtreme Gradient Boosting (XGBoost) algorithm to input risk factors that predict amputation probability. The model performance was optimized by adjusting the super parameters. The predictive performance of the three models was expressed by sensitivity, specificity, positive predictive value, negative predictive value and area under the curve (AUC). Visualization of the prediction results was realized through SHapley Additive exPlanation (SHAP).
    RESULTS: A total of 157 (26.2%) patients underwent minor amputation during hospitalization and 50 (8.3%) had major amputation. All three XGBoost models demonstrated good discriminative ability, with AUC values > 0.7. The model for predicting major amputation achieved the highest performance [AUC = 0.977, 95% confidence interval (CI): 0.956-0.998], followed by the minor amputation model (AUC = 0.800, 95%CI: 0.762-0.838) and the non-amputation model (AUC = 0.772, 95%CI: 0.730-0.814). Feature importance ranking of the three models revealed the risk factors for minor and major amputation. Wagner grade 4/5, osteomyelitis, and high C-reactive protein were all considered important predictive variables.
    CONCLUSION: XGBoost effectively predicts diabetic foot amputation risk and provides interpretable insights to support personalized treatment decisions.
    Keywords:  Amputation risk stratification; Clinical risk prediction; Diabetic foot ulcer; Machine learning; SHapley Additive exPlanation; eXtreme Gradient Boosting
    DOI:  https://doi.org/10.4239/wjd.v16.i7.104789
  8. Biomed Phys Eng Express. 2025 Jul 24.
      Low-quality fundus images pose significant challenges for diabetic retinopathy (DR) classification due to noise, blurred boundaries, and the loss of high-frequency details, which hinder both global contextual understanding and local fine-grained feature extraction. To address these limitations, this work proposes a Hierarchical Frequency-Spatial Feature Fusion Network (HFSF-Net) that effectively integrates frequency-domain and spatial-domain information for robust DR classification. Initially, this paper introduces the Spatial Prior-Aware Transformer (SPAT) Block, which incorporates spatial prior knowledge to direct the attention distribution, enabling precise localization of the complex distribution of lesion regions in low-quality fundus images. Subsequently, a novel Wavelet-Enhanced Self-Attention (WESA) module is developed, which utilizes wavelet transforms to extract and enhance high-frequency components such as microvascular textures and edges. Based on WESA, the Wavelet-Enhanced Transformer (WET) Block is constructed to strengthen the ability to recover local details in degraded images. Furthermore, a Hierarchical Frequency-Spatial Fusion (HFSF) module is designed to hierarchically integrate multi-scale features, mitigating information redundancy and resolving feature conflicts between domains. Through this architecture, the model achieves a balanced representation of global and local information. The experiments conducted on the APTOS and DDR datasets yield ACC values of 0.8117 and 0.8021, and Kappa scores of 0.7158 and 0.6731, respectively. Although the model does not achieve exceptionally high Recall, its consistently strong performance in other key metrics supports the claim that the proposed architecture enables a balanced representation of global and local information. Furthermore, the experimental results validate the effectiveness and robustness of HFSF-Net in classifying diabetic retinopathy from low-quality fundus images.
    Keywords:  Diabetic retinopathy; Frequency-spatial feature fusion; Low-quality fundus images; Transformer network
    DOI:  https://doi.org/10.1088/2057-1976/adf3b5
  9. Proteomes. 2025 Jul 01. pii: 29. [Epub ahead of print]13(3):
       BACKGROUND: Type 2 diabetes mellitus (T2DM) is an epidemic chronic disease that affects millions of people worldwide. This study aims to explore the impact of T2DM on the tear proteome, specifically investigating whether alterations occur before the development of diabetic retinopathy.
    METHODS: Flush tear samples were collected from healthy subjects and subjects with preDM and T2DM. Tear proteins were processed and analyzed by mass spectrometry-based shotgun proteomics using a data-independent acquisition parallel acquisition serial fragmentation (diaPASEF) approach. Machine learning algorithms, including random forest, lasso regression, and support vector machine, and statistical tools were used to identify potential biomarkers.
    RESULTS: Machine learning models identified 17 proteins with high importance in classification. Among these, five proteins (cystatin-S, S100-A11, submaxillary gland androgen-regulated protein 3B, immunoglobulin lambda variable 3-25, and lambda constant 3) exhibited differential abundance across these three groups. No correlations were identified between proteins and clinical assessments of the ocular surface. Notably, the 17 important proteins showed superior prediction accuracy in distinguishing all three groups (healthy, preDM, and T2DM) compared to the five proteins that were statistically significant.
    CONCLUSIONS: Alterations in the tear proteome profile were observed in adults with preDM and T2DM before the clinical diagnosis of ocular abnormality, including retinopathy.
    Keywords:  machine learning; mass spectrometry; proteomics; tear film; type 2 diabetes mellitus
    DOI:  https://doi.org/10.3390/proteomes13030029
  10. Sci Rep. 2025 Jul 23. 15(1): 26811
      The increasing prevalence of type 2 diabetes (T2D) is a significant health concern worldwide. Effective and personalized treatment strategies are essential for improving patient outcomes and reducing healthcare costs. Machine learning (ML) has the potential to create clinical decision support systems (CDSS) that assist clinicians in making prediction-informed treatment decisions. This study aims to develop a novel predictive-prescriptive analytics framework that leverages ML to enhance medication prescriptions for T2D patients. The framework is designed as a data-driven CDSS to determine the best medication strategies based on individual patient profiles, including demographics, comorbidities, and medications. Utilizing a comprehensive dataset of electronic health records from 17,773 patients across various U.S. Veterans Administration Medical Centers collected over 12 years, the study employs the Bayesian Network (BN) as the ML model of choice. The BN's unique dual capability serves both predictive and prescriptive functions. Several BN learning algorithms are applied to map the relationships among patient features and decision variables for predicting the outcome. The prescriptive stage includes three strategies, i.e., forward, backward, and guideline-based, to identify optimal treatment recommendations. Next, the complex treatment pathways identified through the prescriptive stage were illustrated using rule-based and decision-tree presentations to improve interpretability for actionable insights and clinical usability. Finally, our empirical analysis examines the alignment between recommended treatment strategies and actual physician prescriptions. ML exhibited strong predictive performance with a precision of 0.789, a recall of 0.879, and an F1-score of 0.831. The recommended treatment strategies aligned with physician prescriptions in simpler treatment scenarios. However, the alignment decreased as the complexity of medication prescription increased, highlighting the challenges of achieving physician compliance with optimal strategies in complex scenarios. This underscores the greater need for CDSS, particularly in situations involving complex combination therapy. This study presents a novel ML-based CDSS framework for personalized T2D treatment. Leveraging ML, the framework offers a promising approach to optimizing medication prescriptions and improving patient outcomes.
    Keywords:  Bayesian network; Data-driven optimization; Machine learning; Personalized medicine; Predictive-prescriptive analytics; Type 2 diabetes
    DOI:  https://doi.org/10.1038/s41598-025-12310-1
  11. Sci Rep. 2025 Jul 23. 15(1): 26765
      Diabetes is a chronic condition brought on by either an inability to use insulin effectively or a lack of insulin produced by the body. If left untreated, this illness can be lethal to a person. Diabetes can be treated and a good life can be led with early diagnosis. The conventional method of identifying diabetes utilizing clinical and physical data is laborious, hence an automated method is required. An ensemble deep learning model is presented in this research for the diagnosis of diabetes which includes three steps. Preprocessing is the first step, which includes cleaning, normalizing, and organizing the data so that it can be fed into deep learning models. The second step involves employing two neural networks to retrieve features. Convolutional neural network (CNN) is the first neural network utilized for extracting the spatial characteristics of the data, while Long Short-Term Memory (LSTM) networks-more specifically, an LSTM Stack-are used to comprehend the time-dependent flow of the data based on medical information from patients. The last step is combining the two feature sets that the CNN and LSTM models have acquired to create the input for the MLP (Multi-layer Perceptron) classifier. To diagnose sickness, the MLP model serves as a meta-learner to combine and convert the data from the two feature extraction algorithms into the target variable. According to the implementation results, the suggested approach outperformed the compared approaches in terms of average accuracy and precision, achieving 98.28% and 0.99%, respectively, indicating a very great capacity to identify diabetes.
    Keywords:  Convolutional neural network; Deep neural networks; Diabetes disease; Ensemble learning; Long short-term memory
    DOI:  https://doi.org/10.1038/s41598-025-12151-y
  12. Ewha Med J. 2025 Apr;48(2): e34
       Purpose: Accurate prediction of blood glucose variability is crucial for effective diabetes management, as both hypoglycemia and hyperglycemia are associated with increased morbidity and mortality. However, conventional predictive models rely primarily on patient-specific biometric data, often neglecting the influence of patient-provider interactions, which can significantly impact outcomes. This study introduces Cyclic Dual Latent Discovery (CDLD), a deep learning framework that explicitly models patient-provider interactions to improve prediction of blood glucose levels. By leveraging a real-world intensive care unit (ICU) dataset, the model captures latent attributes of both patients and providers, thus improving forecasting accuracy.
    Methods: ICU patient records were obtained from the MIMIC-IV v3.0 critical care database, including approximately 5,014 instances of patient-provider interaction. The CDLD model uses a cyclic training mechanism that alternately updates patient and provider latent representations to optimize predictive performance. During preprocessing, all numeric features were normalized, and extreme glucose values were capped at 500 mg/dL to mitigate the effect of outliers.
    Results: CDLD outperformed conventional models, achieving a root mean square error of 0.0852 on the validation set and 0.0899 on the test set, which indicates improved generalization. The model effectively captured latent patient-provider interaction patterns, yielding more accurate glucose variability predictions than baseline approaches.
    Conclusion: Integrating patient-provider interaction modeling into predictive frameworks can increase blood glucose prediction accuracy. The CDLD model offers a novel approach to diabetes management, potentially paving the way for artificial intelligence-driven personalized treatment strategies.
    Keywords:  Biometry; Blood glucose; Deep learning; Diabetes mellitus; Hyperglycemia; United States
    DOI:  https://doi.org/10.12771/emj.2025.00332