bims-aukdir Biomed News
on Automated knowledge discovery in diabetes research
Issue of 2025–05–11
eleven papers selected by
Mott Given
  1. Validation of artificial intelligence algorithm LuxIA for screening of diabetic retinopathy from a single 45° retinal colour fundus images: the CARDS study.
  2. Latent space autoencoder generative adversarial model for retinal image synthesis and vessel segmentation.
  3. Deep learning segmentation of periarterial and perivenous capillary-free zones in optical coherence tomography angiography.
  4. Non-proliferative diabetic retinopathy detection using Rosmarus Quagga optimized explainable generative meta learning based deep convolutional neural network model.
  5. Recent trends in diabetes mellitus diagnosis: an in-depth review of artificial intelligence-based techniques.
  6. Development of a 5-Year Risk Prediction Model for Transition From Prediabetes to Diabetes Using Machine Learning: Retrospective Cohort Study.
  7. Machine Learning Analysis of Blood Glucose Regulation in Korean Male Workers with Type 2 Diabetes.
  8. Characterizing the clinical profile and prevalence of people with diabetes attended in the hospital setting by using unstructured healthcare data and natural language processing: the Diabetic@ study.
  9. Predicting isolated impaired glucose tolerance without oral glucose tolerance test using machine learning in Chinese Han men.
  10. Targeted Interventions Lead to Quality Improvement in Year 2 of an Artificial Intelligence-Based Diabetic Retinopathy Detection Program in Northern California.
  11. AI-driven personalized nutrition: RAG-based digital health solution for obesity and type 2 diabetes.
  1. BMJ Open Ophthalmol. 2025 May 08. pii: e002109. [Epub ahead of print]10(1):
    Validation of artificial intelligence algorithm LuxIA for screening of diabetic retinopathy from a single 45° retinal colour fundus images: the CARDS study.
    Rodrigo Abreu-Gonzalez, Gabriela Susanna-González, Joseph P M Blair, Romina M Lasagni Vitar, Carlos Ciller, Stefanos Apostolopoulos, Sandro De Zanet, José Natán Rodríguez Martín, Carlos Bermúdez, Alfonso Luis Calle Pascual, Elena Rigo, Enrique Cervera Taulet, Jose Juan Escobar-Barranco, Rosario Cobo-Soriano, Juan Donate-Lopez.
       OBJECTIVE: This study validated the artificial intelligence (AI)-based algorithm LuxIA for screening more-than-mild diabetic retinopathy (mtmDR) from a single 45° colour fundus image of patients with diabetes mellitus (DM, type 1 or type 2) in Spain. Secondary objectives included validating LuxIA according to the International Clinical Diabetic Retinopathy (ICDR) classification and comparing its performance between different devices.
    METHODS: In this multicentre, cross-sectional study, retinal colour fundus images of adults (≥18 years) with DM were collected from five hospitals in Spain (December 2021-December 2022). 45° colour fundus photographs were captured using non-mydriatic Topcon and ZEISS cameras. The Discovery platform (RetinAI) was used to collect images. LuxIA output was an ordinal score (1-5), indicating a classification as mtmDR based on an ICDR severity score.
    RESULTS: 945 patients with DM were included; the mean (SD) age was 64.6 (13.5) years. The LuxIA algorithm detected mtmDR with a sensitivity and specificity of 97.1% and 94.8%, respectively. The area under the receiver-operating characteristic curve was 0.96, indicating a high test accuracy. The 95% CI data for overall accuracy (94.8% to 95.6%), sensitivity (96.8% to 98.2%) and specificity (94.3% to 95.1%) indicated robust estimations by LuxIA, which maintained a concordance of classification (N=829, kappa=0.837, p=0.001) when used to classify Topcon images. LuxIA validation on ZEISS-obtained images demonstrated high accuracy (90.6%), specificity (92.3%) and lower sensitivity (83.3%) as compared with Topcon-obtained images.
    CONCLUSIONS: AI algorithms such as LuxIA are increasing testing feasibility for healthcare professionals in DR screening. This study validates the real-world utility of LuxIA for mtmDR screening.
    Keywords:  Diagnostic tests/Investigation; Imaging; Vision
    DOI:  https://doi.org/10.1136/bmjophth-2024-002109
  2. BMC Med Imaging. 2025 May 05. 25(1): 149
    Latent space autoencoder generative adversarial model for retinal image synthesis and vessel segmentation.
    K Radha, Yepuganti Karuna.
      Diabetes is a widespread condition that can lead to serious vision problems over time. Timely identification and treatment of diabetic retinopathy (DR) depend on accurately segmenting retinal vessels, which can be achieved through the invasive technique of fundus imaging. This methodology facilitates the systematic monitoring and assessment of the progression of DR. In recent years, deep learning has made significant steps in various fields, including medical image processing. Numerous algorithms have been developed for segmenting retinal vessels in fundus images, demonstrating excellent performance. However, it is widely recognized that large datasets are essential for training deep learning models to ensure they can generalize well. A major challenge in retinal vessel segmentation is the lack of ground truth samples to train these models. To overcome this, we aim to generate synthetic data. This work draws inspiration from recent advancements in generative adversarial networks (GANs). Our goal is to generate multiple realistic retinal fundus images based on tubular structured annotations while simultaneously creating binary masks from the retinal fundus images. We have integrated a latent space auto-encoder to maintain the vessel morphology when generating RGB fundus images and mask images. This approach can synthesize diverse images from a single tubular structured annotation and generate various tubular structures from a single fundus image. To test our method, we utilized three primary datasets, DRIVE, STARE, and CHASE_DB, to generate synthetic data. We then trained and tested a simple UNet model for segmentation using this synthetic data and compared its performance against the standard dataset. The results indicated that the synthetic data offered excellent segmentation performance, a crucial aspect in medical image analysis, where smaller datasets are often common. This demonstrates the potential of synthetic data as a valuable resource for training segmentation and classification models for disease diagnosis. Overall, we used the DRIVE, STARE, and CHASE_DB datasets to synthesize and evaluate the proposed image-to-image translation approach and its segmentation effectiveness.
    Keywords:  Data generation; Diabetic retinopathy (DR); Early diagnosis; Generative adversarial models; Vessel segmentation
    DOI:  https://doi.org/10.1186/s12880-025-01694-1
  3. J Biomed Opt. 2025 May;30(5): 056005
    Deep learning segmentation of periarterial and perivenous capillary-free zones in optical coherence tomography angiography.
    Mansour Abtahi, Behrouz Ebrahimi, Albert K Dadzie, Mojtaba Rahimi, Srishti Kolla, Yi-Ting Hsieh, Michael J Heiferman, Jennifer I Lim, Xincheng Yao.
       Significance: Automated segmentation of periarterial and perivenous capillary-free zones (CFZs) in optical coherence tomography angiography (OCTA) can significantly improve early detection and monitoring of diabetic retinopathy (DR), a leading cause of vision impairment, by identifying subtle microvascular changes.
    Aim: We aimed to develop and evaluate deep learning models, including convolutional neural networks (CNNs) and vision transformers (ViTs), for precise segmentation of periarterial and perivenous CFZs. Quantitative features derived from the segmented CFZs were assessed as potential biomarkers for DR.
    Approach: OCTA images from healthy controls, patients with diabetes but no DR (NoDR), and those with mild DR were utilized. Automated CFZ maps were generated using deep learning models such as UNet, UNet++, TransUNet, and Segformer. Quantitative features, including CFZ ratios, counts, and mean sizes, were analyzed to characterize disease progression.
    Results: UNet++ with EfficientNet-b7 achieved the best performance, with a mean intersection over union of 86.48% and a Dice coefficient of 89.87%. Quantitative analyses revealed significant differences in CFZ metrics between the control, NoDR, and mild DR groups, demonstrating their potential as sensitive biomarkers for early DR detection and monitoring.
    Conclusions: The study underscores the efficacy of deep learning models in automating CFZ segmentation and introduces quantitative features as biomarkers for DR. These findings support further exploration of CFZ analysis in retinal disease diagnostics and therapeutic monitoring.
    Keywords:  capillary-free zones; deep learning; diabetic retinopathy; optical coherence tomography angiography
    DOI:  https://doi.org/10.1117/1.JBO.30.5.056005
  4. Int Ophthalmol. 2025 May 08. 45(1): 178
    Non-proliferative diabetic retinopathy detection using Rosmarus Quagga optimized explainable generative meta learning based deep convolutional neural network model.
    Ajita Arvind Mahapadi, Vishal Shirsath, Ajitkumar Pundge.
       PURPOSE: Non-Proliferative Diabetic Retinopathy (NPDR) is a complication of diabetes disease where there is damage of the blood vessels in retina but with no signs of formation of new vessels. It is present in the earlier stages and therefore the control of diabetes combined with constant check-up can address the challenge. Existing models face several challenges such as heterogeneity of the lesion with regard to size, shape, and distribution. Therefore, to reduce those existing challenges, in this research, a novel model Rosmarus Quagga optimized Explainable generative Meta learning based Deep Convolutional Neural Network (RQ-EGMCN) is proposed for Non-Proliferative Diabetic Retinopathy. The main purpose of the proposed research is to develop and validate the effective diagnosis of severe DR with lesion recognition using the retinal images.
    METHODS: The presented approach develops the Rosmarus Quagga optimization, which exhibits the adaptive foraging behaviors are integrated along with the aspects of the leader-based feeding strategies to enhance the detection accuracy. Simultaneously, the proposed model employs explainable Convolutional Neural Networks to ensure interpretations which in turn provides a tradition of decision making by presenting the attention and saliency maps. The generative component allows to generate realistic retinal images for training purposes and meta-learning, when applied to new data, and accelerates learning while enhancing its' generalization potential.
    MAIN OUTCOME MEASURES: Moreover, the proposed model improves the NPDR diagnosis by minimizing the computational complexity, improving the accuracy and versatility of the model across different datasets.
    RESULTS: Experimental analysis show that the RQ-EGMCN model obtained the maximum accuracy, precision, and recall of 95.47%, 95.34%, and 95.24% for the diabetic retinopathy detection dataset, respectively.
    Keywords:  Explainable CNN; Meta learning; Proliferative diabetic retinopathy detection; Rosmarus Quagga optimization and modified patho-GAN
    DOI:  https://doi.org/10.1007/s10792-025-03528-z
  5. Diabetes Res Clin Pract. 2025 May 04. pii: S0168-8227(25)00235-9. [Epub ahead of print]224 112221
    Recent trends in diabetes mellitus diagnosis: an in-depth review of artificial intelligence-based techniques.
    Salman Khalid, Hojun Kim, Heung Soo Kim.
      Diabetes mellitus (DM) is a highly prevalent chronic condition with significant health and economic impacts; therefore, an accurate diagnosis is essential for the effective management and prevention of its complications. This review explores the latest advances in artificial intelligence (AI) focusing on machine learning (ML) and deep learning (DL) for the diagnosis of diabetes. Recent developments in AI-driven diagnostic tools were analyzed, with an emphasis on breakthrough methodologies and their real-world clinical applications. This review also discusses the role of various data sources, datasets, and preprocessing techniques in enhancing diagnostic accuracy. Key advancements in integrating AI into clinical workflows and improving early detection are highlighted along with challenges related to model interpretability, ethical considerations, and practical implementation. By offering a comprehensive overview of these advancements and their implications, this review contributes significantly to the understanding of how AI technologies can enhance the diagnosis of diabetes and support their integration into clinical practice, thereby aiming to improve patient outcomes and reduce the burden of diabetes.
    Keywords:  Artificial Intelligence; Deep learning; Diabetes Mellitus; Early Detection, Healthcare; Machine learning; Prediction
    DOI:  https://doi.org/10.1016/j.diabres.2025.112221
  6. J Med Internet Res. 2025 May 09. 27 e73190
    Development of a 5-Year Risk Prediction Model for Transition From Prediabetes to Diabetes Using Machine Learning: Retrospective Cohort Study.
    Yongsheng Zhang, Hongyu Zhang, Dawei Wang, Na Li, Haoyue Lv, Guang Zhang.
       BACKGROUND: Diabetes has emerged as a critical global public health crisis. Prediabetes, as the transitional phase with 5%-10% annual progression to diabetes, offers a critical window for intervention. The lack of a 5-year risk prediction model for diabetes progression among Chinese individuals with prediabetes limits clinical decision-making support.
    OBJECTIVE: This study aimed to develop and validate a machine learning-based 5-year risk prediction model of progression from prediabetes to diabetes for the Chinese population and establish an interactive web-based platform to facilitate high-risk patients identifying and early targeted interventions, ultimately reducing diabetes incidence and health care burdens.
    METHODS: A retrospective cohort study was conducted on 2 prediabetes cohorts from 2 Chinese medical centers (primary cohort: n=6578 and external validation cohort: n=2333) tracking from 2019 to 2024. Participants meeting the American Diabetes Association (ADA) criteria (prediabetes: hemoglobin A1c [HbA1c] level of 5.7%-6.4%; diabetes: HbA1c level of ≥6.5%) were identified. A total of 42 variables (demographics, physical measures, and hematologic biomarkers) were collected using standardized protocols. Patients were split into the training (70%) and test (30%) sets randomly in the primary cohort. Significant predictors were selected on the training set using recursive feature elimination methods, followed by prediction model development using 7 machine learning algorithms (logistic regression, random forest, support vector machine, multilayer perceptron, extreme gradient boosting machine, light gradient boosting machine, and categorical boosting machine [CatBoost]), optimized through grid search and 5-fold cross-validation. Model performance was assessed using the receiver operating characteristic curve, the precision-recall curves, accuracy, sensitivity, and specificity as well as multiple other metrics on both the test set and the external test set.
    RESULTS: During the follow-up of 5 years, 2610 (41.6%) participants and 760 (35.2%) participants progressed from prediabetes to diabetes, with mean annual progression rates of 8.34% and 7.04% in the primary cohort and the external cohort, respectively. Using 14 features selected using the recursive feature elimination-logistic algorithm, the CatBoost model achieved optimal performance in the test set and the external test set with an area under the receiver operating characteristic curve of 0.819 and 0.807, respectively. It also showed the best discrimination performance on the accuracy, negative predictive value (NPV), and F1-scores as well as the calibration performances in both the test set and the external test set. Then the Shapley Additive Explanations (SHAP) analysis highlighted the top 6 predictors (FBG, HDL, ALT/AST, BMI, age, and MONO), enabling targeted modification of these risk factors to reduce diabetes incidence.
    CONCLUSIONS: We developed a 5-year risk prediction model of progression from prediabetes to diabetes for the Chinese population, with the CatBoost model showing the best predictive performance, which could effectively identify individuals at high risk of diabetes.
    Keywords:  CatBoost; Chinese population; SHAP; clinical decision support; risk factors
    DOI:  https://doi.org/10.2196/73190
  7. Iran J Public Health. 2025 Apr;54(4): 888-890
    Machine Learning Analysis of Blood Glucose Regulation in Korean Male Workers with Type 2 Diabetes.
    Haewon Byeon.
      
    DOI:  https://doi.org/10.18502/ijph.v54i4.18429
  8. Diabetes Res Clin Pract. 2025 May 02. pii: S0168-8227(25)00228-1. [Epub ahead of print] 112214
    Characterizing the clinical profile and prevalence of people with diabetes attended in the hospital setting by using unstructured healthcare data and natural language processing: the Diabetic@ study.
    A J Blanco-Carrasco, J F Merino-Torres, M Rubio Almanza, G Canovas Molina, M A Brito-Sanfiel, D E Barajas Galindo, L A Cuellar Olmedo, D Mauricio, S Tofé Povedano, J A Balsa Barro, J J Aparicio Sánchez, M Sequera Mutiozabal, B Pimentel, A Pérez Domínguez, C Arias-Cabrales, V Fanjul, L Pérez de Isla, J F Navarro González.
       AIMS: This study aimed to evaluate the potential of unstructured electronic health records (EHRs) data, analyzed using natural language processing (NLP) and machine learning (ML), to describe the prevalence and clinical spectrum of diabetes mellitus (DM) in hospitals.
    METHODS: A multicenter, retrospective study was conducted using EHRs from eight Spanish hospitals (2013-2018). Unstructured data were extracted using EHRead® (NLP and ML) and SNOMED_CT. Individuals with type 1 or 2 DM (T1DM/T2DM) were identified, and a semi-supervised ML classifier was developed for unregistered types (UrDM). DM prevalence and related complications were analyzed in the final subpopulations (sT1DM/sT2DM).
    RESULTS: From 56,181,954 EHRs of 2,582,778 individuals, 638,730 were identified with DM: 75.4 % with UrDM, 21.3 % with T2DM, and 3.3 % with T1DM. The ML model reclassified 93.5 % as T2DM and 6.5 % as T1DM. Over 50 % of relevant variables like anthropometrics, lab values and treatments were missing. The prevalence of sT1DM/sT2DM was 2.6 %/38.4 %. Major comorbidities included hypertension, dyslipidemia, chronic kidney disease (CKD), ischemic heart disease, and chronic heart failure (CHF). CKD and CHF were the most frequent complications for sT1DM/sT2DM at 60 months.
    CONCLUSIONS: NLP and ML for profiling DM using EHRs unstructured data are helpful, but additional data and better EHR documentation are crucial.
    Keywords:  Diabetes Comorbidities; Diabetes Complications; Diabetes Mellitus (DM); Machine Learning (ML); Natural Language Processing (NLP); Unstructured Data
    DOI:  https://doi.org/10.1016/j.diabres.2025.112214
  9. Front Endocrinol (Lausanne). 2025 ;16 1514397
    Predicting isolated impaired glucose tolerance without oral glucose tolerance test using machine learning in Chinese Han men.
    Lin Wang, Jing Xie, Zhaoyan Gu, Xinyu Miao, Lichao Ma, Shuangtong Yan, Yanping Gong, Chunlin Li, Banruo Sun, Yue Ruan.
       Background: Isolated Impaired Glucose Tolerance (I-IGT) represents a specific prediabetic state that typically requires a standardized oral glucose tolerance test (OGTT) for diagnosis. This study aims to predict glucose tolerance status in Chinese Han men at fasting state using machine learning (ML) models with demographic, anthropometric, and laboratory data.
    Methods: The study population consisted of 1,117 Chinese Han men aged 50-87 years. Baseline variables including age, fasting plasma glucose (FPG), high blood pressure (HBP), body mass index (BMI), waist to hip ratio (WHR), total cholesterol (TC), triglyceride (TG), high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C) were collected from electronic medical records (EMRs) for machine learning model training and validation. Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Logistic Regression (LR), K-Nearest Neighbors (KNN), Naive Bayes (NB), Adaptive Boosting (AdaBoost) and Gradient Boosting Machines (GBM) were tested for machine learning model performance comparison. Model performance was evaluated using metrics including accuracy, recall, F1 score, positive predictive value (PPV), negative predictive value (NPV), and the area under the receiver operating characteristic curve (AUC). Shapley Additive Explanations (SHAP) and confusion matrix plots were used for model interpretation.
    Results: The RF model demonstrated the best overall performance with a 96.7% accuracy, recall of 91.4%, F1 score of 95.7%, PPV of 99.1%, and NPV of 95.6%. The AUC values for the SVM, DT, RF, LR, KNN, NB, AdaBoost, and GBM models were 0.97, 0.92, 0.96, 0.97, 0.88, 0.88, 0.97, and 0.97, respectively. While the RF model showed strong overall performance, the LR model had the highest AUC, indicating superior discriminatory power. FPG was identified as the most important predictor for I-IGT, followed by HDL, TC, HBP, BMI, and WHR. Individuals with FPG levels higher than 5.1 mmol/L were more likely to have I-IGT; the performance metrics for this cut-off value were: 89.35% accuracy, 89.79% recall, 85.22% F1 score, 81.09% PPV, 94.38% NPV, and 0.95 AUC.
    Conclusion: Machine learning models based on demographic and clinical characteristics offer a cost-effective method for predicting I-IGT in Chinese Han men aged over 50, without the need for an OGTT. These models could complement existing early diagnostic strategies, thereby enhancing the early detection and prevention of diabetes. Additionally, FPG alone could serve as an efficient screening tool for the early identification of I-IGT in clinical settings.
    Keywords:  fasting plasma glucose; isolated impaired glucose tolerance; machine learning models; oral glucose tolerance test; pre-diabetes
    DOI:  https://doi.org/10.3389/fendo.2025.1514397
  10. Retina. 2025 Apr 30.
    Targeted Interventions Lead to Quality Improvement in Year 2 of an Artificial Intelligence-Based Diabetic Retinopathy Detection Program in Northern California.
    Karen M Chen, Cindy S Zhao, Austen Knapp, Eliot Dow, Anuradha Phadke, Marilyn Tan, Kaniksha Desai, Christopher Or, Vinit Mahajan, Diana V Do, Prithvi Mruthyunjaya, Theodore Leng, David Myung.
       PURPOSE: This study evaluates the second-year outcomes of an AI-based diabetic retinopathy (DR) detection program (Stanford Teleophthalmology Autonomous Testing and Universal Screening (STATUS)) implemented in primary care and endocrinology clinics in Northern California. We focused on assessing improvements following implementation of an intervention-based framework to increase AI system gradability and patient encounters.
    METHODS: A retrospective analysis was conducted involving diabetic patients aged 18 years and older with no prior DR diagnosis or examination in the past year. These patients presented for routine DR screening in primary care or endocrinology clinics. In its second year, the STATUS program expanded to additional sites and introduced an intervention-based framework, including targeted training protocols, to enhance screening accuracy and efficiency. Our study measured AI system gradability and tracked patient encounters over Year 2.
    RESULTS: The AI system's gradability increased from 62.3% in Year 1 to 71.2% in Year 2, comparable to non-mydriatic gradability rates observed in clinical trials. Patient encounters increased by 21.9%, indicating expanded reach and improved accessibility. Interventions, including enhanced training protocols and camera utilization reports, effectively improved screening efficiency.
    CONCLUSION: The second-year outcomes of the STATUS AI-based DR screening program demonstrate significant improvements in image gradability by the AI system as well as in patient encounter numbers. These findings highlight the potential of interventional methods to continually improve the outcomes of AI-based screening programs and offer a scalable solution to the growing burden of diabetic retinopathy. The success of STATUS supports further integration and expansion of AI-based screening in clinical practice for early detection and management of DR, improving patient outcomes.
    Keywords:  AI-Human Hybrid; Artificial Intelligence; Clinical Artificial Intelligence; DR Screening Program; Diabetic Retinopathy; Primary Care; Teleophthalmology
    DOI:  https://doi.org/10.1097/IAE.0000000000004499
  11. PLOS Digit Health. 2025 May;4(5): e0000758
    AI-driven personalized nutrition: RAG-based digital health solution for obesity and type 2 diabetes.
    Anand K Gavai, Jos van Hillegersberg.
      Effective management of obesity and type 2 diabetes is a major global public health challenge that requires evidence-based, scalable personalized nutrition solutions. Here, we present an artificial intelligence (AI) driven dietary recommendation system that generates personalized smoothie recipes while prioritizing health outcomes and environmental sustainability. A key feature of the system is the "virtual nutritionist", an iterative validation framework that dynamically refines recipes to meet predefined nutritional and sustainability criteria. The system integrates dietary guidelines from the National Institute for Public Health and the Environment (RIVM), EUFIC, USDA FoodData Central, and the American Diabetes Association with retrieval-augmented generation (RAG) to deliver evidence-based recommendations. By aligning with the United Nations Sustainable Development Goals (SDGs), the system promotes plant-based, seasonal, and locally sourced ingredients to reduce environmental impact. We leverage explainable AI (XAI) to enhance user engagement through clear explanations of ingredient benefits and interactive features, improving comprehension across varying health literacy levels. Using zero-shot and few-shot learning techniques, the system adapts to user inputs while maintaining privacy through local deployment of the LLaMA3 model. In evaluating 1,000 recipes, the system achieved 80.1% adherence to health guidelines meeting targets for calories, fiber, and fats and 92% compliance with sustainability criteria, emphasizing seasonal and locally sourced ingredients. A prototype web application enables real-time, personalized recommendations, bridging the gap between AI-driven insights and clinical dietary management. This research underscores the potential of AI-driven precision nutrition to revolutionize chronic disease management by improving dietary adherence, enhancing health literacy, and offering a scalable, adaptable solution for clinical workflows, telehealth platforms, and public health initiatives, with the potential to significantly alleviate the global healthcare burden.
    DOI:  https://doi.org/10.1371/journal.pdig.0000758
This page is being maintained by Thomas Krichel. It was last updated on 2025‒05‒11 at 0:40.