bims-aukdir Biomed News
on Automated knowledge discovery in diabetes research
Issue of 2025–08–31
29 papers selected by
Mott Given



  1. Sci Rep. 2025 Aug 21. 15(1): 30706
      Diabetic retinopathy is a serious ocular complication that poses a significant threat to patients' vision and overall health. Early detection and accurate grading are essential to prevent vision loss. Current automatic grading methods rely heavily on deep learning applied to retinal fundus images, but the complex, irregular patterns of lesions in these images, which vary in shape and distribution, make it difficult to capture the subtle changes. This study introduces RadFuse, a multi-representation deep learning framework that integrates non-linear RadEx-transformed sinogram images with traditional fundus images to enhance diabetic retinopathy detection and grading. Our RadEx transformation, an optimized non-linear extension of the Radon transform, generates sinogram representations to capture complex retinal lesion patterns. By leveraging both spatial and transformed domain information, RadFuse enriches the feature set available to deep learning models, improving the differentiation of severity levels. We conducted extensive experiments on two benchmark datasets, APTOS-2019 and DDR, using three convolutional neural networks (CNNs): ResNeXt-50, MobileNetV2, and VGG19. RadFuse showed significant improvements over fundus-image-only models across all three CNN architectures and outperformed state-of-the-art methods on both datasets. For severity grading across five stages, RadFuse achieved a quadratic weighted kappa of 93.24%, an accuracy of 87.07%, and an F1-score of 87.17%. In binary classification between healthy and diabetic retinopathy cases, the method reached an accuracy of 99.09%, precision of 98.58%, and recall of 99.64%, surpassing previously established models. These results demonstrate RadFuse's capacity to capture complex non-linear features, advancing diabetic retinopathy classification and promoting the integration of advanced mathematical transforms in medical image analysis. The source code will be available at https://github.com/Farida-Ali/RadEx-Transform/tree/main .
    DOI:  https://doi.org/10.1038/s41598-025-14944-7
  2. Bioengineering (Basel). 2025 Aug 03. pii: 840. [Epub ahead of print]12(8):
      The retina offers a unique window into both ocular and systemic health, motivating the development of AI-based tools for disease screening and risk assessment. In this study, we present a comprehensive evaluation of six state-of-the-art deep neural networks, including convolutional neural networks and vision transformer architectures, on the Brazilian Multilabel Ophthalmological Dataset (BRSET), comprising 16,266 fundus images annotated for multiple clinical and demographic labels. We explored seven classification tasks: Diabetes, Diabetic Retinopathy (2-class), Diabetic Retinopathy (3-class), Hypertension, Hypertensive Retinopathy, Drusen, and Sex classification. Models were evaluated using precision, recall, F1-score, accuracy, and AUC. Among all models, the Swin-L generally delivered the best performance across scenarios for Diabetes (AUC = 0.88, weighted F1-score = 0.86), Diabetic Retinopathy (2-class) (AUC = 0.98, weighted F1-score = 0.95), Diabetic Retinopathy (3-class) (macro AUC = 0.98, weighted F1-score = 0.95), Hypertension (AUC = 0.85, weighted F1-score = 0.79), Hypertensive Retinopathy (AUC = 0.81, weighted F1-score = 0.97), Drusen detection (AUC = 0.93, weighted F1-score = 0.90), and Sex classification (AUC = 0.87, weighted F1-score = 0.80). These results reflect excellent to outstanding diagnostic performance. We also employed gradient-based saliency maps to enhance explainability and visualize decision-relevant retinal features. Our findings underscore the potential of deep learning, particularly vision transformer models, to deliver accurate, interpretable, and clinically meaningful screening tools for retinal and systemic disease detection.
    Keywords:  convolutional neural networks; deep learning; explainable AI; fundus images; retinal disease; vision transformers
    DOI:  https://doi.org/10.3390/bioengineering12080840
  3. Br J Ophthalmol. 2025 Aug 22. pii: bjo-2025-327447. [Epub ahead of print]
       PURPOSE: To investigate the diagnostic accuracy, feasibility and end-user experiences of an artificial intelligence (AI)-based, automated diabetic retinopathy (DR) screening model in real-world, Australian primary care and endocrinology clinics.
    METHODS: In a pragmatic trial conducted across five sites including general practice and endocrinology clinics, from August 2021 to June 2023, patients aged ≥50 years, and those aged ≥18 years with diabetes were screened using an AI-integrated, non-mydriatic fundus camera. The AI instantly analysed the retinal images for referable DR. Patients detected with referable DR or ungradable images were referred to eyecare professionals. The accuracy of the AI grading was assessed against gold standard human grading. A satisfaction survey was administered among the participants and care providers.
    RESULTS: Among 863 participants enrolled (mean (SD) age: 62.6 (13.2) years; 53.0% women), the AI system achieved high accuracy of 93.3% (95% CI: 91.4% to 95.5%) for referable DR detection, with a sensitivity of 83.7% (95% CI: 78.2% to 88.3%), specificity of 96.1% (95% CI: 94.7% to 97.2%) and an area under the receiver operating characteristic curve of 0.899 (95% CI: 0.874 to 0.924). The proportion of ungradable images was lower according to the AI grading (13.4%) compared with human grading (15.6%). Most patients (86%) and care providers (85%) expressed high satisfaction with the AI system.
    CONCLUSIONS: The AI-assisted DR screening model was accurate and well received by patients and staff in Australian primary care and endocrinology clinics. This opportunistic screening model holds promise for enhancing early DR detection in non-eyecare settings, potentially preventing vision loss due to DR on a considerable scale.
    Keywords:  Diagnostic tests/Investigation; Humans; Imaging; Public health; Retina
    DOI:  https://doi.org/10.1136/bjo-2025-327447
  4. Sensors (Basel). 2025 Aug 13. pii: 5019. [Epub ahead of print]25(16):
      Diabetic retinopathy (DR), a leading cause of vision loss worldwide, poses a critical challenge to healthcare systems due to its silent progression and the reliance on labor-intensive, subjective manual screening by ophthalmologists, especially amid a global shortage of eye care specialists. Addressing the pressing need for scalable, objective, and interpretable diagnostic tools, this work introduces RetinoDeep-deep learning frameworks integrating hybrid architectures and explainable AI to enhance the automated detection and classification of DR across seven severity levels. Specifically, we propose four novel models: an EfficientNetB0 combined with an SPCL transformer for robust global feature extraction; a ResNet50 ensembled with Bi-LSTM to synergize spatial and sequential learning; a Bi-LSTM optimized through genetic algorithms for hyperparameter tuning; and a Bi-LSTM with SHAP explainability to enhance model transparency and clinical trustworthiness. The models were trained and evaluated on a curated dataset of 757 retinal fundus images, augmented to improve generalization, and benchmarked against state-of-the-art baselines (including EfficientNetB0, Hybrid Bi-LSTM with EfficientNetB0, Hybrid Bi-GRU with EfficientNetB0, ResNet with filter enhancements, Bi-LSTM optimized using Random Search Algorithm (RSA), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), and a standard Convolutional Neural Network (CNN)), using metrics such as accuracy, F1-score, and precision. Notably, the Bi-LSTM with Particle Swarm Optimization (PSO) outperformed other configurations, achieving superior stability and generalization, while SHAP visualizations confirmed alignment between learned features and key retinal biomarkers, reinforcing the system's interpretability. By combining cutting-edge neural architectures, advanced optimization, and explainable AI, this work sets a new standard for DR screening systems, promising not only improved diagnostic performance but also potential integration into real-world clinical workflows.
    Keywords:  EfficientNetB0; SHAP explainability; SPCL transformer; ant colony optimization; bidirectional LSTM; data augmentation; diabetic retinopathy; particle swarm optimization
    DOI:  https://doi.org/10.3390/s25165019
  5. Commun Med (Lond). 2025 Aug 23. 5(1): 368
       BACKGROUND: Diabetic retinopathy (DR) is the leading cause of blindness worldwide, making early prediction of DR progression crucial for effectively preventing visual loss. This study introduces a prediction framework DRForecastGAN (Diabetic Retinopathy Forecast Generative Adversarial Network), and investigates its clinical value in predicting DR development.
    METHODS: DRForecastGAN model, consisting of a generator, discriminator, and registration network, was trained, validated, and tested in training (12,852 images), internal validation (2734 images), and external test (8523 images) datasets. A pre-trained ResNet50 classification model identified the DR severity on synthetic images. The performance of the proposed DRForecastGAN model was compared with the CycleGAN and Pix2Pix models in image reality and DR severity of the synthesized fundus images by calculating Fréchet Inception Distance (FID), Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and area under the curve (AUC).
    RESULTS: DRForecastGAN model has the lowest FID, highest PSNR and highest SSIM on internal validation (FID: 27.3 vs. 32.8 vs. 34.4; PSNR: 25.3 vs. 17.0 vs. 16.9; SSIM: 0.93 vs. 0.79 vs. 0.65) and external test (FID: 37.6 vs.45.1 vs.48.4; PSNR: 20.7 vs.15.2 vs.14.7; SSIM: 0.86 vs.0.69 vs.0.63) datasets compared with Pix2Pix and CycleGAN models. In the prediction of DR severity, our DRForecastGAN model outperforms both Pix2Pix and CycleGAN models, achieving the highest AUC values on both internal validation (0.87 vs. 0.76 vs. 0.75) and external test (0.85 vs. 0.70 vs. 0.69) datasets.
    CONCLUSIONS: The proposed DRForecastGAN model can effectively visualize DR development by synthesizing future fundus images, offering potential utility for both treatment and ongoing monitoring of DR.
    DOI:  https://doi.org/10.1038/s43856-025-01092-2
  6. Front Endocrinol (Lausanne). 2025 ;16 1649988
       Aim: We aimed to develop and internally validate a machine learning (ML)-based model for the prediction of the risk of type 2 diabetes mellitus (T2DM) in children with obesity.
    Methods: In total, 292 children with obesity and T2DM were enrolled between July 2023 and February 2024 and followed for at least 1 year. Eight ML algorithms (Decision Tree, Logistic Regression, Support Vector Machine (SVM), Multilayer Perceptron, Adaptive Boosting, Random Forest, Gradient Boosting Decision Tree, and Extreme Gradient Boosting) were compared for their capacity to identify key clinical and laboratory characteristics of T2DM in children and to create a risk prediction model.
    Results: Forty-nine children were diagnosed with T2DM during the follow-up period. The SVM algorithm was the best predictor of T2DM, with the largest area under the receiver operating characteristic curve (0.98) and accuracy (93.2%). The SVM algorithm identified eight predictors: BMI, creatinine, prealbumin, glucose (180 min), glycosylated hemoglobin A1c, thyrotropin, total thyroxine (T4), and free T4 concentrations. Thus, an ML-based prediction model accurately identifies children with obesity at high risk of T2DM. If externally validated, this tool could facilitate early, personalized interventions aimed at preventing T2DM.
    Discussion: The rising prevalence of obesity in childhood is associated with an increase in the risk of early-onset T2DM. Therefore, the early identification of individuals at high risk is crucial to prevent the development of this disease. In a comparative analysis of the performance of multiple ML algorithms, we found that the SVM algorithm was the best predictor of the development of T2DM.
    Keywords:  children; machine learning; obesity; risk prediction model; type 2 diabetes mellitus
    DOI:  https://doi.org/10.3389/fendo.2025.1649988
  7. Front Endocrinol (Lausanne). 2025 ;16 1611499
       Objective: Osteoporosis is a common complication in patients with type 2 diabetes mellitus (T2DM), yet its screening rate remains low. This study aimed to develop and validate a cost-effective and interpretable machine learning (ML) model to predict the risk of osteoporosis in patients with T2DM.
    Methods: This retrospective study included 1560 inpatients who underwent dual-energy X-ray absorptiometry (DXA) between January 2022 and December 2023 at Panyu Hospital of Chinese Medicine. Demographic information and laboratory test results obtained within 24 hours of hospital admission were collected. Potential predictive features were identified using univariate analysis, least absolute shrinkage and selection operator (LASSO) regression, and the Boruta algorithm. Eight supervised ML algorithms were applied to construct predictive models. Model performance was evaluated based on the area under the receiver operating characteristic curve (AUC), calibration plots, decision curve analysis (DCA), accuracy, sensitivity, specificity, and F1 score. The SHapley Additive exPlanations (SHAP) method was used to interpret the model and visualize feature importance.
    Results: Ten predictive features were selected based on the intersection of the three feature selection methods. Among the tested models, logistic regression achieved the best overall performance, with an AUC of 0.812, an accuracy of 0.762, a sensitivity of 0.809, a specificity of 0.761, and an F1 score of 0.771 in the validation set. Calibration plots and DCA curves demonstrated good agreement and the highest net clinical benefit. SHAP analysis identified age, sex, alkaline phosphatase, uric acid, hemoglobin, and neutrophil count as the six most influential features. An easy-to-use, web-based risk calculator was developed based on the logistic model and is available at: https://t2dm.shinyapps.io/t2dm-osteoporosis/.
    Conclusion: We developed an interpretable and accessible ML-based online tool that enables preliminary screening of osteoporosis risk in patients with T2DM using routine blood indicators. This tool may assist clinicians in early risk identification and reduce the underdiagnosis of osteoporosis.
    Keywords:  explainable machine learning; osteoporosis; predictive model; risk assessment; type 2 diabetes mellitus
    DOI:  https://doi.org/10.3389/fendo.2025.1611499
  8. BMC Med Inform Decis Mak. 2025 Aug 27. 25(1): 315
      Early diagnosis and screening of diabetic retinopathy (DR) are crucial for reducing medical burdens and conserving healthcare resources. This study introduces an advanced AI-assisted recognition system designed to enhance the detection of DR lesions through innovative automatic learning methods. Central to our approach are agnostic text instruction templates, which facilitate zero-shot DR detection by integrating text embeddings with visual information. Our system performs comprehensive lesion detection by leveraging similarity mapping at both the image and patch levels, enabling it to identify a wide range of diabetic retinopathy (DR) lesions without the need for extensive annotated data. This AI-assisted system distinguishes itself from traditional fully supervised models and few-shot learning approaches by addressing the complexities of DR image annotation and safeguarding patient privacy. To validate the system's effectiveness, we conducted extensive experiments across five internal and publicly available test sets, as well as an external test set captured using smartphone devices. Our evaluation involved performance analysis of various pre-training methods, including detailed patch-level visualizations and t-SNE clustering techniques to assess the quality of feature embeddings. The results of our zero-shot experiments reveal that our system outperforms conventional transfer learning-based DR detection methods. This superiority is evident in both the pre-training and testing phases, showcasing the system's ability to deliver accurate and reliable DR lesion detection while circumventing the limitations of traditional approaches.
    DOI:  https://doi.org/10.1186/s12911-025-03117-1
  9. JMIR Med Inform. 2025 Aug 21. 13 e72938
       Background: Several studies have used electronic health records (EHRs) to build machine learning models predicting the likelihood of developing gestational diabetes mellitus (GDM) later in pregnancy, but none have described validation of the GDM "label" within the EHRs.
    Objective: This study examines the accuracy of GDM diagnoses in EHRs compared with a clinical team database (CTD) and their impact on machine learning models.
    Methods: EHRs from 2018 to 2022 were validated against CTD data to identify true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN). Logistic regression models were trained and tested using both EHR and validated labels, whereafter simulated label noise was introduced to increase FP and FN rates. Model performance was assessed using the area under the receiver operating characteristic curve (ROC AUC) and average precision (AP).
    Results: Among 3952 patients, 3388 (85.7%) were correctly identified with GDM in both databases, while 564 cases lacked a GDM label in EHRs, and 771 were missing a corresponding CTD label. Overall, 32,928 (87.5%) of cases were TN, 3388 (9%) TP, 771 (2%) FP, and 564 (1.5%) FN. The model trained and tested with validated labels achieved an ROC AUC of 0.817 and an AP of 0.450, whereas the same model tested using EHR labels achieved 0.814 and 0.395, respectively. Increased label noise during training led to gradual declines in ROC AUC and AP, while noise in the test set, especially elevated FP rates, resulted in marked performance drops.
    Conclusions: Discrepancies between EHR and CTD diagnoses had a limited impact on model training but significantly affected performance evaluation when present in the test set, emphasizing the importance of accurate data validation.
    Keywords:  electronic health records; gestational diabetes; label noise; machine learning; pregnancy; validation
    DOI:  https://doi.org/10.2196/72938
  10. Biomedicines. 2025 Aug 07. pii: 1926. [Epub ahead of print]13(8):
      Objective: Retinal capillary dropout, characterized by acellular capillaries or "ghost vessels," is an early pathological sign of diabetic retinopathy (DR) that remains undetectable through standard clinical imaging techniques until visible morphological changes, such as microaneurysms or hemorrhages, occur. This study aims to develop a non-destructive artificial intelligence (AI)-based method using fluorescein angiography (FA) images to detect early-stage, silent retinal capillary dropout. Methods: We utilized 94 FA images and corresponding destructive retinal capillary density measurements obtained through retinal trypsin digestion from 51 Nile rats. Early capillary dropout was defined as having an acellular capillary density of ≥18 counts per mm2. A DenseNet based deep learning model was trained to classify images into early capillary dropout or normal. A Bayesian framework incorporating diabetes duration was used to enhance model predictions. RNA sequencing was conducted on retinal vasculature to identify molecular markers associated with capillary early dropout. Results: The AI-based FA imaging model demonstrated an accuracy of 80.85%, sensitivity of 84.21%, specificity of 75.68%, and an AUC of 0.86. Integration of diabetes duration into a Bayesian predictive framework further improved the model's performance (AUC = 0.90). Transcriptomic analysis identified 43 genes significantly upregulated in retinal tissues preceding capillary dropout. Notably, inflammatory markers such as Bcl2a1, Birc5, and Il20rb were among these genes, indicating that inflammation might play a critical role in early DR pathogenesis. Conclusions: This study demonstrates that AI-enhanced FA imaging can predict silent retinal capillary dropout before conventional clinical signs of DR emerge. Combining AI predictions with diabetes duration data significantly improves diagnostic performance. The identified gene markers further highlight inflammation as a potential driver in early DR, offering novel insights and potential therapeutic targets for preventing DR progression.
    Keywords:  Bayesian framework; RNA-seq; biomarkers; image-AI; retinal capillary dropout prediction
    DOI:  https://doi.org/10.3390/biomedicines13081926
  11. Sci Rep. 2025 Aug 20. 15(1): 30572
    Gatekeeper Consortium
      
    Keywords:  Deep learning; Interstitial glucose prediction; Multimodal AI; Time series modelling
    DOI:  https://doi.org/10.1038/s41598-025-16371-0
  12. J Imaging. 2025 Aug 19. pii: 279. [Epub ahead of print]11(8):
      Accurate and early classification of retinal diseases such as diabetic retinopathy, cataract, and glaucoma is essential for preventing vision loss and improving clinical outcomes. Manual diagnosis from fundus images is often time-consuming and error-prone, motivating the development of automated solutions. This study proposes a deep learning-based classification model using a pretrained EfficientNetB3 architecture, fine-tuned on a publicly available Kaggle retinal image dataset. The model categorizes images into four classes: cataract, diabetic retinopathy, glaucoma, and healthy. Key enhancements include transfer learning, data augmentation, and optimization via the Adam optimizer with a cosine annealing scheduler. The proposed model achieved a classification accuracy of 95.12%, with a precision of 95.21%, recall of 94.88%, F1-score of 95.00%, Dice Score of 94.91%, Jaccard Index of 91.2%, and an MCC of 0.925. These results demonstrate the model's robustness and potential to support automated retinal disease diagnosis in clinical settings.
    Keywords:  CNN; EfficientNetB0; cataract; deep learning; diabetic retinopathy; eye disease classification; fundus images; glaucoma; image augmentation; medical image analysis
    DOI:  https://doi.org/10.3390/jimaging11080279
  13. Curr Opin Ophthalmol. 2025 Aug 27.
       PURPOSE OF REVIEW: Artificial intelligence (AI) is transforming retina care, with deep learning (DL) models shaping a new era of improved screening accessibility, diagnostic precision, and personalized disease monitoring. This review highlights recent AI-powered clinical applications in diabetic retinopathy (DR), and age-related macular degeneration (AMD) care.
    RECENT FINDINGS: Since the FDA's authorization of the first autonomous AI system for DR screening in 2018, multiple platforms have emerged, expanding access to diabetic eye care. Real-world studies have confirmed a significant improvement in screening adherence and diagnostic accuracy, illustrating AI's tangible impact on public health. Meanwhile, newly certified AI technologies that meet European regulatory standards are increasingly guiding clinical decision-making in the management of AMD and diabetic macular edema through automated analysis of optical coherence tomography (OCT) images. Most recently, FDA-authorized home OCT platforms are transforming AMD monitoring, enabling proactive and remote management of retinal fluid.
    SUMMARY: As AI increasingly empowers patients and providers, its widespread success still depends on ongoing work, including thorough validation, outcome-based metrics, and improved workflow integration. The next decade will reveal whether AI in retina care transitions from a promising innovation to an essential and indispensable tool in modern retina.
    Keywords:  age-related macular degeneration; artificial intelligence; diabetic retinopathy; imaging; retina
    DOI:  https://doi.org/10.1097/ICU.0000000000001167
  14. J Diabetes Sci Technol. 2025 Aug 25. 19322968251355967
       BACKGROUND: Artificial intelligence (AI) has emerged as a transformative tool for advancing gestational diabetes mellitus (GDM) care, offering dynamic, data-driven methods for early detection, management, and personalized intervention.
    OBJECTIVE: This systematic review aims to comprehensively explore and synthesize the use of AI models in GDM care, including screening, diagnosis, management, and prediction of maternal and neonatal outcomes. Specifically, we examine (1) study designs and population characteristics; (2) the use of AI across different aspects of GDM care; (3) types of input data used for AI modeling; and (4) AI model types, validation strategies, and performance metrics.
    METHODS: A systematic search was conducted across six electronic databases, identifying 126 eligible studies published up to February 2025. Data extraction and quality appraisal were independently conducted by six reviewers, using a modified version of the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool for risk of bias assessment.
    RESULTS: Among 126 studies, 75% employed retrospective designs, with sample sizes ranging from 17 to over 100 000 participants. Most AI applications (85%) focused on early GDM prediction, while fewer addressed management, outcomes, or monitoring. Classical machine learning dominated (84%), with logistic regression and random forest frequently used. Internal validation was common (68%), but external validation was rare (6%). Our risk of bias appraisal indicated an overall moderate-to-good methodological quality, with notable deficiencies in analysis reporting.
    CONCLUSIONS: AI demonstrates strong potential to improve GDM prediction, screening, and management. Nonetheless, broader validation, enhanced model interpretability, and prospective studies in diverse populations are needed to translate these technologies into clinical practice.
    Keywords:  artificial intelligence; diabetes; gestational diabetes; pregnancy; women’s health
    DOI:  https://doi.org/10.1177/19322968251355967
  15. Diagnostics (Basel). 2025 Aug 09. pii: 1996. [Epub ahead of print]15(16):
      Background/Objectives: Diabetic foot ulcers (DFUs) remain a critical complication of diabetes, with high rates of amputation when not diagnosed early. Despite advancements in medical imaging, current DFU detection methods are often limited by their computational complexity, poor generalizability, and delayed diagnostic performance. This study presents a novel hybrid diagnostic framework that integrates traditional feature extraction methods with deep learning (DL) to improve the early real-time computer-aided detection (CAD) of DFUs. Methods: The proposed model leverages plantar thermograms to detect early thermal asymmetries associated with DFUs. It uniquely combines the oriented FAST and rotated BRIEF (ORB) algorithm with the Bag of Features (BOF) method to extract robust handcrafted features while also incorporating deep features from pretrained convolutional neural networks (ResNet50, AlexNet, and EfficientNet). These features were fused and input into a lightweight deep neural network (DNN) classifier designed for binary classification. Results: Our model demonstrated an accuracy of 98.51%, precision of 100%, sensitivity of 98.98%, and AUC of 1.00 in a publicly available plantar thermogram dataset (n = 1670 images). An ablation study confirmed the superiority of ORB + DL fusion over standalone approaches. Unlike previous DFU detection models that rely solely on either handcrafted or deep features, our study presents the first lightweight hybrid framework that integrates ORB-based descriptors with deep CNN representations (e.g., ResNet50 and EfficientNet). Compared with recent state-of-the-art models, such as DFU_VIRNet and DFU_QUTNet, our approach achieved a higher diagnostic performance (accuracy = 98.51%, AUC = 1.00) while maintaining real-time capability and a lower computational overhead, making it highly suitable for clinical deployment. Conclusions: This study proposes the first integration of ORB-based handcrafted features with deep neural representations for DFU detection from thermal images. The model delivers high accuracy, robustness to noise, and real-time capabilities, outperforming existing state-of-the-art approaches and demonstrating strong potential for clinical deployment.
    Keywords:  deep learning (DL); deep neural network (DNN); diabetic foot; diabetic foot ulcers (DFUs); plantar thermograms; thermal images
    DOI:  https://doi.org/10.3390/diagnostics15161996
  16. Medicina (Kaunas). 2025 Aug 01. pii: 1403. [Epub ahead of print]61(8):
      Background and Objectives: Diabetes is a global public health challenge, with increasing prevalence worldwide. The implementation of artificial intelligence (AI) in the management of this condition offers potential benefits in improving healthcare outcomes. This study primarily investigates the barriers and facilitators perceived by healthcare professionals in the adoption of AI. Secondarily, by analyzing both quantitative and qualitative data collected, it aims to support the potential development of AI-based programs for diabetes management, with particular focus on a possible bottom-up approach. Materials and Methods: A scoping review was conducted following PRISMA-ScR guidelines for reporting and registered in the Open Science Framework (OSF) database. The study selection process was conducted in two phases-title/abstract screening and full-text review-independently by three researchers, with a fourth resolving conflicts. Data were extracted and assessed using Joanna Briggs Institute (JBI) tools. The included studies were synthesized narratively, combining both quantitative and qualitative analyses to ensure methodological rigor and contextual depth. Results: The adoption of AI tools in diabetes management is influenced by several barriers, including perceived unsatisfactory clinical performance, high costs, issues related to data security and decision-making transparency, as well as limited training among healthcare workers. Key facilitators include improved clinical efficiency, ease of use, time-saving, and organizational support, which contribute to broader acceptance of the technology. Conclusions: The active and continuous involvement of healthcare workers represents a valuable opportunity to develop more effective, reliable, and well-integrated AI solutions in clinical practice. Our findings emphasize the importance of a bottom-up approach and highlight how adequate training and organizational support can help overcome existing barriers, promoting sustainable and equitable innovation aligned with public health priorities.
    Keywords:  artificial intelligence; diabetes; healthcare workers; public health; scoping review
    DOI:  https://doi.org/10.3390/medicina61081403
  17. J Imaging. 2025 Aug 18. pii: 278. [Epub ahead of print]11(8):
      Ocular disease (OD) represents a complex medical condition affecting humans. OD diagnosis is a challenging process in the current medical system, and blindness may occur if the disease is not detected at its initial phase. Recent studies showed significant outcomes in the identification of OD using deep learning (DL) models. Thus, this work aims to develop a multi-classification DL-based model for the classification of seven ODs, including normal (NOR), age-related macular degeneration (AMD), diabetic retinopathy (DR), glaucoma (GLU), maculopathy (MAC), non-proliferative diabetic retinopathy (NPDR), and proliferative diabetic retinopathy (PDR), using color fundus images (CFIs). This work proposes a custom model named the ocular disease detection model (ODDM) based on a CNN. The proposed ODDM is trained and tested on a publicly available ocular disease dataset (ODD). Additionally, the SMOTE Tomek (SM-TOM) approach is also used to handle the imbalanced distribution of the OD images in the ODD. The performance of the ODDM is compared with seven baseline models, including DenseNet-201 (R1), EfficientNet-B0 (R2), Inception-V3 (R3), MobileNet (R4), Vgg-16 (R5), Vgg-19 (R6), and ResNet-50 (R7). The proposed ODDM obtained a 98.94% AUC, along with 97.19% accuracy, a recall of 88.74%, a precision of 95.23%, and an F1-score of 88.31% in classifying the seven different types of OD. Furthermore, ANOVA and Tukey HSD (Honestly Significant Difference) post hoc tests are also applied to represent the statistical significance of the proposed ODDM. Thus, this study concludes that the results of the proposed ODDM are superior to those of baseline models and state-of-the-art models.
    Keywords:  AMD; CFI; deep learning; diabetic retinopathy; eye disease; ocular disease
    DOI:  https://doi.org/10.3390/jimaging11080278
  18. Diabetol Metab Syndr. 2025 Aug 27. 17(1): 361
       BACKGROUND: This study aimed to evaluate the predictive performance of an artificial intelligence (AI)-based algorithm in estimating the risk of cardio-cerebrovascular complications in patients with type 2 diabetes mellitus (T2D).
    METHODS: Medical records of 532 T2D patients from the Diabetology Unit in Padova, Italy, were analyzed using the Metaclinic AI Prediction Module, which estimates the probability of heart and cerebrovascular organ damage. For patients identified as "Very high" (n = 63) or "Low" (n = 122) risk for heart disease, additional clinical and instrumental data on their cardiac history were collected. The level of agreement between AI predictions and traditional clinical-instrumental diagnostics was assessed using Cohen's κ coefficient.
    RESULTS: In the "Very high" risk group, the agreement between AI predictions and clinical diagnostics for heart disease was poor (κ = 0.00), while prediction for cerebrovascular disease showed excellent agreement (κ = 0.89). Similarly, in the "Low" risk group, agreement for heart disease remained poor (κ = 0.00), but agreement for cerebrovascular disease was again high (κ = 0.83).
    CONCLUSIONS: A marked difference was observed in the algorithm's performance. While the AI showed strong predictive ability for cerebrovascular complications, it failed to reliably predict heart disease risk. These results suggest that the algorithm may be clinically valuable for cerebrovascular risk assessment but needs refinement for cardiac prediction.
    Keywords:  Artificial intelligence; Cerebrovascular disease; Cohen’s kappa; Heart disease; Risk evaluation; Type 2 diabetes mellitus
    DOI:  https://doi.org/10.1186/s13098-025-01910-6
  19. BioData Min. 2025 Aug 21. 18(1): 57
      This study explores diabetic foot (DF), a severe complication in diabetes, by combining deep learning (DL) and machine learning (ML) to develop a multi-model prediction tool. Early identification of high-risk DF patients can reduce disability and mortality. The research also aims to create an integrated application to assist clinicians in precise, efficient risk assessment for early intervention. In this multicenter retrospective study, 6,180 elderly diabetic patients (aged 60-85) were enrolled from 11 community hospitals in Shanghai in 2024. Lasso regression was used to identify 16 key DF risk factors, including age, MMSE score, lower limb discomfort, ABI, and hematocrit. Fourteen ML models (RF, XGBoost, CART, MLP, etc.) and three DL models (DNN, CNN, Transformer) were trained, with hyperparameters optimized via cross-validation and grid search. An application was developed integrating these models, offering both single and batch prediction options with visualization tools for clinical use.Experimental results showed the Logistic regression ensemble model achieved robust performance, with AUC values of 0.943 (validation set, 95% CI: 0.935-0.951) and 0.938 (test set, 95% CI: 0.929-0.947), along with high accuracy, precision, recall, and F1 scores. SHAP analysis revealed key predictive features including ABI results, lower limb discomfort, and MMSE score. The developed app integrates multiple models, compares their predictions for different clinical scenarios, and enhances prediction transparency and reliability.The multi-model approach demonstrates strong predictive performance for DF risk, offering clinicians an intuitive and accurate assessment tool tailored to individual patients. By combining multiple models, we enhance result stability and clinical applicability compared to single-model approaches. Future work will focus on algorithm optimization, expanded datasets, and real-time monitoring integration to enable more precise, dynamic risk evaluation for improved DF prevention and early intervention.
    Keywords:  Artificial intelligence; Machine learning; Prediction model
    DOI:  https://doi.org/10.1186/s13040-025-00477-2
  20. Braz J Med Biol Res. 2025 ;pii: S0100-879X2025000100662. [Epub ahead of print]58 e14986
      It is unclear who benefits the most from atherosclerotic cardiovascular disease (ASCVD) screening imaging. This study aimed to identify features associated with positive coronary artery calcium scores (CACS) in individuals with diabetes using machine learning (ML) techniques. ELSA-Brasil is a cohort study with 15,105 participants aged 35 to 74 years in six Brazilian cities. We analyzed 25 sociodemographic, medical history, symptom-related, and laboratory variables from 585 participants from the São Paulo investigation center with CACS data and no overt cardiovascular disease at baseline. We used six ML algorithms to build models to identify individuals with positive CACS. Feature importance was determined by SHapley Additive exPlanations (SHAP) values. The best performer ML algorithm was the XGBoost Classifier (accuracy: 94.8%). Age (SHAP: 0.220), systolic blood pressure (SHAP: 0.102), and body mass index (SHAP: 0.075) were the most important variables to identify ASCVD in individuals with diabetes in XGBoost models. Considering all ML models in our analysis, age, systolic blood pressure, and sex were frequently influential variables. We obtained high accuracy with our best model, using information generally present in current clinical practice. ML models may help clinicians select patients with characteristics most probably associated with a positive CAC. Age, systolic blood pressure, body mass index, and sex may be useful markers to identify those at higher risk for subclinical ASCVD.
    DOI:  https://doi.org/10.1590/1414-431X2025e14986
  21. Nutrients. 2025 Aug 12. pii: 2610. [Epub ahead of print]17(16):
      Background: Severe hypoglycemia (SH) is a critical complication in children and adolescents with type 1 diabetes (T1D), associated with cognitive impairment, coma, and significant psychosocial burden. Despite advances in glucose monitoring, predicting SH remains challenging, as most models focus on milder hypoglycemic events. Objective: To develop a machine learning model for early prediction of SH using continuous glucose monitoring (CGM) data in children and adolescent T1D patients. Methodology: This retrospective study analyzed CGM data from 67 patients (37 SH episodes, 1430 non-SH segments). Glycemic curves were segmented into 5-day windows, and 21 features were extracted, including glycemic mean, variability, time below range (TBR < 60 mg/dL), and PCA components of glucose trends. A support vector machine (SVM) model was trained using repeated cross-validation to predict SH 15 min before onset. Model performance was evaluated using sensitivity, specificity, balanced classification rate (BCR), and area under the ROC curve (AUC). Results: The model achieved robust performance, with a median AUC of 90% (IQR: 87-93%) and median BCR of 84% (IQR: 80-89%). Sensitivity and specificity exceeded 80%, demonstrating reliable detection of impending SH. However, the positive predictive value (PPV) was low (12%), with false alarms frequently triggered during descending glucose trends or near-hypoglycemic values (end glucose <54 mg/dL). SH episodes were stratified into two subgroups: group 1 (<45 mg/dL, n = 26) and group 2 (>52 mg/dL, n = 15). Notably, false alarms occurred at a median interval of 25 days, minimizing alarm fatigue. Conclusions: These findings confirm the feasibility of SH prediction in clinical practice, prioritizing high-risk events over milder hypoglycemia. By alerting patients and medical teams early on, this tool could facilitate individualized treatment adjustments, reduce the risk of serious hypoglycemic events, and thus contribute to more personalized management of pediatric diabetes, while improving patients' quality of life.
    Keywords:  continuous glucose monitoring; machine learning; predictive modeling; severe hypoglycemia; type 1 diabetes
    DOI:  https://doi.org/10.3390/nu17162610
  22. Healthcare (Basel). 2025 Aug 15. pii: 2007. [Epub ahead of print]13(16):
      Background: China has the largest number of patients with type 2 diabetes (T2D) worldwide, and the chronic complications and economic burden associated with T2D are becoming increasingly severe. Developing accurate and widely applicable risk prediction models is of great significance for the early identification of and intervention in high-risk populations. However, current Chinese models still have many shortcomings in terms of methodological design and clinical application. Objective: This study conducts a systematic review and narrative synthesis of existing risk prediction models for type 2 diabetes in China, aiming to identify issues with existing models and provide references with which Chinese scholars can develop higher-quality risk prediction models. Methods: This study followed the PRISMA guidelines to conduct a systematic search of the literature related to T2D risk prediction models in China published in English journals from October 2019 to October 2024. The databases included PubMed, CNKI and Web of Science. Included studies had to meet criteria such as clear modeling objectives, detailed model development and validation processes, and a focus on non-diabetic populations in China. A total of 20 studies were ultimately selected and comprehensively analyzed based on model type, variable selection, validation methods, and performance metrics. Results: The 20 included studies employed various modeling methods, including statistical and machine learning approaches. The AUC values of the models ranged from 0.728 to 0.977, indicating overall good predictive capability. However, only one study conducted external validation, and 45% (9/20) of the studies binned continuous variables, which may have reduced the models' generalization ability and predictive performance. Additionally, most models did not include key variables such as lifestyle, socioeconomic factors, and cultural background, resulting in limited data representativeness and adaptability. Conclusions: Chinese T2DM risk prediction models remain in the developmental stage, with issues such as insufficient validation, inconsistent variable handling, and incomplete coverage of key influencing factors. Future research should focus on strengthening multicenter external validation, standardizing modeling processes, and incorporating multidimensional social and behavioral variables to enhance the clinical utility and cross-population applicability of these models. Registration ID: CRD420251072143.
    Keywords:  China; external validation; generalization ability; machine learning; risk prediction model; type 2 diabetes
    DOI:  https://doi.org/10.3390/healthcare13162007
  23. IEEE J Biomed Health Inform. 2025 Aug 27. PP
      Nutritional intervention can improve glycemic control for type 2 diabetes mellitus (T2DM), and thus accurately predicting post-prandial glycemic responses (PPGRs) to each meal is essential. PPGRs can vary significantly between individuals, even when consuming the same foods, due to the diverse and complex nature of individual characteristics. However, to date, system-scale studies investigating the variability of PPGRs in people living with T2DM are scarce. This research collected meal logs, continuous glucose monitoring records, clinicodemographic profiles, and gut microbiota data comprising over 2,000 real-life meals across 88 individuals with T2DM, revealing causal relationships in the diet-microbiome-PPGR interplay. Furthermore, we developed a multimodal deep learning predictive PPGR model that integrates heterogeneous input data. The proposed model achieves R of 0.62 and 0.66 for 2- and 4-h PPGR prediction, respectively, significantly surpassing the perfor-mance of the carbohydrate single predictor and state-of-the-art machine learning algorithms. This model substantially improved the prediction in the subgroup of low responders to carbohydrates, a traditionally challenging population for accurate prediction using carbohydrate-based methods. This advancement empowers personalized PPGR prediction, laying the foundation for precision nutrition and better glycemic management for individuals with T2DM.
    DOI:  https://doi.org/10.1109/JBHI.2025.3602827
  24. Spectrochim Acta A Mol Biomol Spectrosc. 2025 Aug 13. pii: S1386-1425(25)01130-8. [Epub ahead of print]345 126823
      Type 2 diabetes mellitus (T2DM) is associated with increased skeletal fragility, yet standard clinical assessments often fail to detect diabetes-induced changes in bone quality. Raman spectroscopy (RS), a label-free and non-destructive technique, offers insight into bone composition, and its full spectral profile may reveal changes not captured by traditional compositional metrics. Prior to propagating a crack from a micro-notch to failure, we acquired RS data from human cortical bone samples extracted from fresh-frozen, cadaveric femurs: 60 non-diabetic donors and 60 T2DM donors (equal number of females and males between 50 years and 97 years of age). Eight ML models, including random forest (RF), support vector regression (SVR), ridge regression (RR), partial least squares (PLS), Extreme Gradient Boosting (XGBoost), gradient boosting machine (GBM), Adaptive Boosting (Adaboost), and Stacking, were evaluated for their ability to predict fracture toughness properties from the RS data. Using full-spectrum input, stacking regression yielded the best performance for predicting both crack initiation toughness (R2 = 0.81, RMSE = 0.08) and the final J-integral or energy required to propagate crack to failure (R2 = 0.86, RMSE = 0.14). These findings demonstrate that full-spectrum RS combined with ML can capture subtle, functionally relevant alterations in bone composition, enabling prediction of mechanical properties that are otherwise inaccessible. This is the first study to apply RS and ML for fracture toughness prediction in the context of T2DM, demonstrating the potential of spectroscopic approaches to improve assessment of bone quality in metabolic disease.
    Keywords:  Artificial intelligence; Bone biomechanics; Bone mechanical properties; Bone quality; Fracture toughness; Machine learning; Raman spectroscopy; Type 2 diabetes mellitus
    DOI:  https://doi.org/10.1016/j.saa.2025.126823
  25. Sci Rep. 2025 Aug 20. 15(1): 30636
      Personalized blood glucose (BG) prediction in Type 1 Diabetes (T1D) is challenged by significant inter-patient heterogeneity. To address this, we propose BiT-MAML, a hybrid model combining a Bidirectional LSTM-Transformer with Model-Agnostic Meta-Learning. We evaluated our model using a rigorous Leave-One-Patient-Out Cross-Validation (LOPO-CV) on the OhioT1DM dataset, ensuring a fair comparison against re-implemented LSTM and Edge-LSTM baselines. The results show our model achieved a mean RMSE of 24.89 mg/dL for the 30 min prediction horizon, marking a substantial improvement of 19.3% over the standard LSTM and 14.2% over the Edge-LSTM. Notably, our model also achieved the lowest standard deviation (±4.60 mg/dL), indicating more consistent and generalizable performance across the patient cohort. A key finding of our study is the confirmation of significant performance variability across individuals, a known clinical challenge. This was evident as our model's 30 min RMSE ranged from an excellent 19.64 mg/dL to a more challenging 30.57 mg/dL, reflecting the inherent difficulty of personalizing predictions rather than model instability. From a clinical safety perspective, Clarke Error Grid Analysis confirmed the model's robustness, with over 92% of predictions falling within the clinically acceptable Zones A and B. This study concludes that the development of effective personalized BG prediction requires not only advanced model architectures but also robust evaluation methods that transparently report the full spectrum of performance, providing a realistic pathway toward reliable clinical tools.
    Keywords:  Bidirectional long short term memory; Blood glucose prediction; Deep learning; Model-agnostic meta-learning; Transformer; Type 1 diabetes
    DOI:  https://doi.org/10.1038/s41598-025-13491-5
  26. Ann Med. 2025 Dec;57(1): 2536204
       BACKGROUND: Chronic systemic inflammation is a key contributor to cardiometabolic complications in diabetes mellitus (DM) and prediabetes (PreDM). Composite inflammatory indices-including neutrophil-to-lymphocyte ratio (NLR), monocyte-to-lymphocyte ratio (MLR), systemic inflammation response index (SIRI), systemic immune-inflammation index (SII), platelet-to-hemoglobin ratio (PHR), and aggregate inflammation systemic index (AISI)-have shown prognostic value for mortality. However, their integrated assessment using machine learning and quantification at the population level remain limited.
    METHODS: In this retrospective cohort study, 11,304 adults with DM or PreDM from the National Health and Nutrition Examination Survey (NHANES, 2005-2018) were analyzed. The primary outcomes were all-cause and cardiovascular mortality. Associations between inflammatory indices and mortality were evaluated using Cox proportional hazards models. Predictive performance was assessed via Extreme Gradient Boosting (XGBoost), and population attributable fractions (PAFs) estimated the mortality burden related to systemic inflammation.
    RESULTS: NLR, MLR, SIRI, SII, and AISI were independently associated with all-cause and cardiovascular mortality. MLR showed the strongest association (HR: 2.948 and 3.717 for all-cause and CVD mortality, respectively). XGBoost identified SIRI, SII, AISI, MLR, and NLR as key predictors, with SIRI ranked highest for cardiovascular mortality. Inclusion of inflammatory indices improved model discrimination and calibration. PAF analysis suggested that 10-20% of mortality reduction could be attributed to improved inflammatory profiles.
    CONCLUSION: Systemic inflammatory indices are independent predictors of mortality in individuals with DM or PreDM. Their integration into machine learning models enhances risk prediction and may inform population-level strategies for cardiometabolic risk stratification.
    Keywords:  Diabetes mellitus; mortality; national health and nutrition examination survey; population attributable fraction; prediabetes; systemic inflammatory indices
    DOI:  https://doi.org/10.1080/07853890.2025.2536204
  27. Front Endocrinol (Lausanne). 2025 ;16 1588718
       Background: Changes in certain metabolites are linked to an increased risk of type I diabetes (T1D), making metabolite analysis a valuable tool for T1D diagnosis and treatment. This study aimed to identify a metabolic signature linked with T1D.
    Methods: Untargeted metabolomic profiling was performed using liquid chromatography-mass spectrometry (LC-MS) on peripheral blood samples from T1D patients (n = 45) and healthy controls (n = 40). Data preprocessing and quality control were conducted using MetaboAnalyst 4.0. Differential metabolites (DMs) were identified via Wilcoxon rank-sum test (P< 0.05), and key diagnostic markers were selected using least absolute shrinkage and selection operator (LASSO) regression. A streptozotocin (STZ)-induced diabetic rat model was used for in vivo validation.
    Results: A total of 157 annotated metabolites were detected (58 in ESI- and 99 in ESI+ mode). Twenty-six DMs were identified, including 25 upregulated and 1 downregulated in the T1D group, mainly involving Acylcarnitines and xanthine metabolites. LASSO regression selected Hydroxyhexadecanoyl carnitine, Propionylcarnitine, and Valerylcarnitine as candidate markers. In the rat model, Hydroxyhexadecanoyl carnitine and Valerylcarnitine demonstrated strong diagnostic performance, with AUCs of 0.9383 (95% CI: 0.8786-0.9980) and 0.8395 (95% CI: 0.7451-0.9338), respectively (P< 0.01).
    Conclusion: Hydroxyhexadecanoyl carnitine and Valerylcarnitine are closely linked to altered lipid oxidation in T1D and show strong potential as diagnostic biomarkers. These findings provide new insights into the metabolic basis of T1D and offer promising targets for early detection.
    Keywords:  LASSO; LC-MS; metabolic markers; metabolomics; type 1 diabetes
    DOI:  https://doi.org/10.3389/fendo.2025.1588718