bims-aukdir Biomed News
on Automated knowledge discovery in diabetes research
Issue of 2025–06–08
eleven papers selected by
Mott Given
  1. Revolutionizing diabetes care: the role of artificial intelligence in prevention, diagnosis, and patient care.
  2. Artificial Intelligence for the Detection of Diabetic Retinopathy.
  3. Highly efficient homomorphic encryption-based federated learning for diabetic retinopathy classification.
  4. Development and validation of a 3-D deep learning system for diabetic macular oedema classification on optical coherence tomography images.
  5. Enhancing Antidiabetic Drug Selection Using Transformers: Machine-Learning Model Development.
  6. Machine learning-based coronary heart disease diagnosis model for type 2 diabetes patients.
  7. Determining the risk of gestational diabetes using machine learning: A study on first-trimester PAPP-A and β-hCG data.
  8. CGMformer: a novel deep-learning model promising for early detection of prediabetes to effectively prevent type 2 diabetes.
  9. Enhancing Care in Type 1 Diabetes with Artificial Intelligence Driven Clinical Decision Support Systems.
  10. The Application of Machine Learning in Warfarin Dose Precision for Diabetic Patients Treated with Statins: A Comparative Study.
  11. Machine Learning Insights Into Amputation Risk: Evaluating Wound Classification Systems in Diabetic Foot Ulcers.
  1. J Diabetes Metab Disord. 2025 Jun;24(1): 132
    Revolutionizing diabetes care: the role of artificial intelligence in prevention, diagnosis, and patient care.
    Madhav Kohli, Pallavi Pandey, Vikash Jakhmola, Supriyo Saha, Meenu Chaudhary, Arif Nur Muhammad Ansori, Arvind Negi.
      Millions of people worldwide have diabetes, a disease that is becoming more common and has substantial socioeconomic costs. Artificial intelligence (AI) improves diabetes management, diagnosis, and prevention. AI-powered tools enable early detection of diabetes and its complications, including diabetic retinopathy, using sophisticated algorithms and large-scale data analysis. Wearable devices and continuous glucose monitors, integrated with AI, facilitate personalized treatment plans and real-time insights, improving glycemic control and overall health outcomes. Advanced machine learning models demonstrate high accuracy in diagnosing and predicting diabetes, while automated insulin delivery systems and bolus calculators enhance insulin management, reducing risks of hypo- and hyperglycemia. Despite these advancements, challenges such as cost, accessibility, device interoperability, and ethical considerations persist. The development of new digital biomarkers, individualized clinical metrics, and patient-centric solutions is critical for optimizing care. While AI holds immense promise in alleviating the global diabetes burden, addressing these limitations through sustained innovation and collaboration is essential. This review underscores the transformative potential of AI in revolutionizing diabetes care, enabling advancement for enhanced prevention, precise diagnosis, and effective management strategies.
    Supplementary Information: The online version contains supplementary material available at 10.1007/s40200-025-01648-y.
    Keywords:  Artificial intelligence; Diabetes management; Glycaemic control; Patient-centric approach; Personalized treatment
    DOI:  https://doi.org/10.1007/s40200-025-01648-y
  2. Klin Monbl Augenheilkd. 2025 Jun 02.
    Artificial Intelligence for the Detection of Diabetic Retinopathy.
    Ansgar Beuse, Carsten Grohmann, Hauke M Schadwinkel, Christos Skevas, Martin S Spitzer.
      Screening and timely treatment can avoid the majority of severe vision loss and blindness from diabetic retinopathy. Artificial intelligence (AI) algorithms that detect DR from retinal photographs without human assessment might reduce the challenges of systematic screening. The German National Care Guideline recommends that individuals with diabetes receive annual or biennial eye examinations to detect treatable DR. Efficient and comprehensive screening of the growing diabetic population requires more and more resources. Artificial intelligence (AI) algorithms that detect DR from retinal photographs without human assessment might help in coping with the immense screening burden. Many of these AI algorithms have achieved good sensitivity and specificity for detecting treatable DR, as compared to human graders; however, many important challenges remain, such as acceptance, cost-effectiveness, liability issues, IT security, and reimbursement. AI-supported DR screening has so far only been used to a limited extent, even in countries with a developed digital infrastructure. These questions must be addressed before AI-based DR screening can be implemented on a large scale into clinical practice. This overview presents key concepts in development and currently approved AI applications for DR screening.
    DOI:  https://doi.org/10.1055/a-2545-1192
  3. J Med Imaging (Bellingham). 2025 May;12(3): 034504
    Highly efficient homomorphic encryption-based federated learning for diabetic retinopathy classification.
    Christopher Nielsen, Matthias Wilms, Nils D Forkert.
       Purpose: Diabetic retinopathy (DR) is the leading cause of blindness among working-age adults globally. Although machine learning (ML) has shown promise for DR diagnosis, ensuring model generalizability requires training on data from diverse populations. Federated learning (FL) offers a potential solution by enabling model training on decentralized datasets. However, privacy concerns persist in FL due to potential privacy breaches, such as gradient inversion attacks, which can be used to reconstruct sensitive training data and may discourage participation from patients.
    Approach: We developed and tested a computationally efficient FL framework that integrates homomorphic encryption (HE) to safeguard patient privacy using 6457 retinal fundus images from the APTOS-2019 and ODIR-5K datasets. First, features are extracted from distributed fundus images using RETFound, a large pretrained foundation model for retinal analysis. These encrypted features are then used to train a lightweight multiclass logistic regression head (MLRH) model for DR grade classification using FL.
    Results: Experimental results show that the MLRH model trained using FL achieves similar performance compared with a fully fine-tuned RETFound model on centralized data, with the area under the receiver operating characteristic curve scores of 0.93±0.01 on APTOS-2019 and 0.78±0.02 on ODIR-5K. Efficiency improvements include a 95.9-fold reduction in computation time and a 63.0-fold reduction in data transfer needs compared with fine-tuning the full RETFound model with FL. In addition, results showed that integrating HE effectively protects patient data against gradient inversion attacks.
    Conclusions: We advance privacy-preserving, ML-based DR screening technology, supporting the goal of equitable vision care worldwide.
    Keywords:  diabetic retinopathy; federated learning; fundus imaging; homomorphic encryption
    DOI:  https://doi.org/10.1117/1.JMI.12.3.034504
  4. BMJ Open. 2025 May 31. 15(5): e099167
    Development and validation of a 3-D deep learning system for diabetic macular oedema classification on optical coherence tomography images.
    Huishan Zhu, Jie Ji, Jian-Wei Lin, Ji Wang, Yi Zheng, Peiwen Xie, Cui Liu, Tsz Kin Ng, Jinqu Huang, Yongqun Xiong, Hanfu Wu, Leixian Lin, Mingzhi Zhang, Guihua Zhang.
       OBJECTIVES: To develop and validate an automated diabetic macular oedema (DME) classification system based on the images from different three-dimensional optical coherence tomography (3-D OCT) devices.
    DESIGN: A multicentre, platform-based development study using retrospective and cross-sectional data. Data were subjected to a two-level grading system by trained graders and a retina specialist, and categorised into three types: no DME, non-centre-involved DME and centre-involved DME (CI-DME). The 3-D convolutional neural networks algorithm was used for DME classification system development. The deep learning (DL) performance was compared with the diabetic retinopathy experts.
    SETTING: Data were collected from Joint Shantou International Eye Center of Shantou University and the Chinese University of Hong Kong, Chaozhou People's Hospital and The Second Affiliated Hospital of Shantou University Medical College from January 2010 to December 2023.
    PARTICIPANTS: 7790 volumes of 7146 eyes from 4254 patients were annotated, of which 6281 images were used as the development set and 1509 images were used as the external validation set, split based on the centres.
    MAIN OUTCOMES: Accuracy, F1-score, sensitivity, specificity, area under receiver operating characteristic curve (AUROC) and Cohen's kappa were calculated to evaluate the performance of the DL algorithm.
    RESULTS: In classifying DME with non-DME, our model achieved an AUROCs of 0.990 (95% CI 0.983 to 0.996) and 0.916 (95% CI 0.902 to 0.930) for hold-out testing dataset and external validation dataset, respectively. To distinguish CI-DME from non-centre-involved-DME, our model achieved AUROCs of 0.859 (95% CI 0.812 to 0.906) and 0.881 (95% CI 0.859 to 0.902), respectively. In addition, our system showed comparable performance (Cohen's κ: 0.85 and 0.75) to the retina experts (Cohen's κ: 0.58-0.92 and 0.70-0.71).
    CONCLUSIONS: Our DL system achieved high accuracy in multiclassification tasks on DME classification with 3-D OCT images, which can be applied to population-based DME screening.
    Keywords:  Diabetic retinopathy; Diagnostic Imaging; Vetreoretinal
    DOI:  https://doi.org/10.1136/bmjopen-2025-099167
  5. JMIR Med Inform. 2025 Jun 02. 13 e67748
    Enhancing Antidiabetic Drug Selection Using Transformers: Machine-Learning Model Development.
    Hisashi Kurasawa, Kayo Waki, Tomohisa Seki, Eri Nakahara, Akinori Fujino, Nagisa Shiomi, Hiroshi Nakashima, Kazuhiko Ohe.
       Background: Diabetes affects millions worldwide. Primary care physicians provide a significant portion of care, and they often struggle with selecting appropriate medications.
    Objective: This study aimed to develop a model that accurately predicts what drug an endocrinologist would prescribe based on the current measurements. The goal was to create a system that would assist nonspecialists in choosing medications, thereby potentially improving diabetes treatment outcomes. Based on the performance of previous studies, we set a performance target of achieving a receiver operating characteristic area under the curve (ROC-AUC) above 0.95.
    Methods: A transformer-based encoder-decoder model predicts whether 44 types of diabetes drugs will be prescribed. The model uses sequences of age, sex, history for 12 laboratory tests, and prescribed drug history as inputs. We assessed the model using the electronic health records from 7034 patients with diabetes seeing endocrinologists between 2012 and 2022 at the University of Tokyo Hospital. We assessed model performance trained on data subsets spanning different time periods (2, 5, and 10 years) using micro- and macro-averaged ROC-AUC on a hold-out test set comprising data solely from 2022. The model's performance was compared against LightGBM.
    Results: The model trained on data from the past 5 years (2017-2021) yielded the best predictive performance, achieving a microaverage (95% CI) ROC-AUC of 0.993 (0.992-0.994) and a macroaverage (95% CI) ROC-AUC of 0.988 (0.980-0.993). The model achieved an ROC-AUC above 0.95 for 43 out of 44 drugs. These results surpassed the predefined performance target and outperformed both previous studies and the LightGBM model's microaverage ROC-AUC of 0.988 (0.985-0.990) in terms of prediction accuracy. Furthermore, training the model with short-term data from the past 5 years yielded high accuracy compared to using data from the past 10 years, suggesting that learning from more recent prescribing patterns might be advantageous.
    Conclusions: The proposed model demonstrates the feasibility of accurately predicting the next prescribed drugs. This model, trained from the past prescriptions of endocrinologists, has the potential to provide information that can assist nonspecialists in making diabetes-treatment decisions. Future studies will focus on incorporating important factors such as prescription contraindications and constraints to enhance safety, as well as leveraging large-scale clinical data across multiple hospitals to improve the generalizability of the model.
    Keywords:  AI; artificial intelligence; diabetes; drug selection; machine learning; transformer
    DOI:  https://doi.org/10.2196/67748
  6. Front Endocrinol (Lausanne). 2025 ;16 1550793
    Machine learning-based coronary heart disease diagnosis model for type 2 diabetes patients.
    Yingxi Chen, Chunyu Wang, Xiaozhu Liu, Minjie Duan, Tianyu Xiang, Haodong Huang.
       Background: To establish a classification model for assisting the diagnosis of type 2 diabetes mellitus (T2DM) complicated with coronary heart disease (CHD).
    Methods: Patients with T2DM who underwent coronary angiography (CA) were enrolled from seven affiliated hospitals of Chongqing Medical University. Statistical differences in clinical variables between T2DM with or without CHD patients were verified using univariate analysis. The original data was divided into a training set and a validation set in a 7:3 ratio. The training set data were used to screen features using Logistic regression, Lasso regression, or recursive feature elimination (RFE). Five machine learning algorithms, including Logistic regression, Support Vector Machine (SVM), Random Forest (RF), eXtreme gradient boosting (XgBoost), and Light Gradient Boosting Machine (LightGBM), were selected for modeling. The performance of the models was verified through 5-fold cross-validation and the training set.
    Results: Clinical data were collected from 1943 patients with T2DM complicated with CHD and 574 T2DM patients without CHD. Univariate analysis identified 20 optimal risk factors, four of the risk factors had over 30% missing values, we ultimately included 16 risk factors. Logistic regression screened eight features, Lasso regression screened ten features, the RFE method screened eight, fourteen, sixteen, and thirteen features for SVM, RF, XgBoost, and LightGBM, respectively. Among all models, the XgBoost model based on features selected by RFE+LightGBM demonstrated the best performance, achieving an AUC of 0.814 (95% CI, 0.779-0.847), accuracy of 0.799 (95% CI, 0.771-0.827), precision of 0.841 (95% CI, 0.812-0.868), recall of 0.920 (95% CI, 0.898-0.941), and F1-score of 0.879 (95% CI, 0.859-0.897) in the testing set.
    Conclusions: Based on T2DM data and machine learning theory, a Bayesian-optimized XgBoost model was established using the RFE+LightGBM method. This model effectively determines whether T2DM patients have CHD.
    Keywords:  coronary heart diseases; diabetic comorbidities; diagnosis model; machine learning; type 2 diabetes mellitus
    DOI:  https://doi.org/10.3389/fendo.2025.1550793
  7. Int J Gynaecol Obstet. 2025 Jun 02.
    Determining the risk of gestational diabetes using machine learning: A study on first-trimester PAPP-A and β-hCG data.
    Uğurcan Zorlu, Burak Elmas, Gonca Türker Ergün, Serap Topkara Sucu, Elif Ozan, Ece Aydoğdu, Dilek Şahin, Özlem Moraloğlu Tekin.
       OBJECTIVE: To evaluate the predictive potential of first-trimester biomarkers-pregnancy-associated plasma protein-A (PAPP-A) and free β-human chorionic gonadotropin (β-hCG)-combined with maternal body mass index (BMI), using machine learning (ML) algorithms for the early detection of gestational diabetes mellitus (GDM).
    METHODS: A retrospective cohort study was conducted with 400 pregnant women who underwent first-trimester screening at Ankara Bilkent City Hospital. Demographic, clinical, and biochemical data, including PAPP-A, free β-hCG, and BMI, were collected. ML models, including random forest, gradient boosting, and logistic regression, were employed to predict GDM risk. Data standardization, model training, and performance evaluation were performed using metrics such as accuracy, F1 score, and receiver operating characteristics area under the curve (ROC-AUC).
    RESULTS: The combination of PAPP-A, free β-hCG, and BMI significantly enhanced GDM prediction accuracy across all ML models. Gradient boosting achieved the highest performance with an ROC-AUC of 0.715 and an accuracy of 71.3%, demonstrating the robust predictive value of these variables. PAPP-A alone showed limited predictive capacity (ROC-AUC: 0.632), but its integration with BMI improved model performance substantially. Cut-off analyses identified key thresholds for PAPP-A (<1.02) and BMI (>26.12) for effective risk stratification.
    CONCLUSION: This study underscores the potential of integrating first-trimester biomarkers with ML algorithms for early GDM prediction. By using routinely collected clinical data, this approach offers a cost-effective and scalable solution for improving maternal and neonatal health outcomes. Future research should validate these findings in diverse populations and explore the incorporation of additional biomarkers to further refine predictive models.
    Keywords:  body mass index; free β‐human chorionic gonadotropin; gestational diabetes mellitus; gradient boosting; machine learning; pregnancy‐associated plasma protein‐A
    DOI:  https://doi.org/10.1002/ijgo.70264
  8. Natl Sci Rev. 2025 Jun;12(6): nwaf188
    CGMformer: a novel deep-learning model promising for early detection of prediabetes to effectively prevent type 2 diabetes.
    Xiao-Ding Peng.
      
    DOI:  https://doi.org/10.1093/nsr/nwaf188
  9. Horm Res Paediatr. 2025 Jun 05. 1-22
    Enhancing Care in Type 1 Diabetes with Artificial Intelligence Driven Clinical Decision Support Systems.
    Revital Nimri, Moshe Phillip.
      Background Type 1 diabetes (T1D) is a chronic condition that requires significant daily self-management and long-term clinical care, involving insulin therapy, glucose monitoring , dietary control and education. The majority of tasks associated with diabetes care are becoming amenable to the application of artificial intelligence (AI) driven clinical decision support systems (AI-CDSS). By integrating data from multiple sources, including smartphone apps, smart watches, activity trackers, continuous glucose monitors (CGM), insulin pumps and smartpens, AI-CDSS can empower people with T1D to make daily self-management of T1D more personalized, more predictive and more proactive. For healthcare professionals (HCPs), AI-CDSS are already changing approaches to risk prediction, detection and assessment of presymptomatic T1D. When necessary, AI-CDSS can help HCPs to prioritize necessary clinical management approaches for people with T1D, as well as streamlining service delivery and allocating resources more effectively. Summary AI technologies are anticipated to provide valuable support for people with T1D in their daily life with diabetes. Equally, AI-CDSS can have high value for HCPs and healthcare services in the screening, monitoring and management of T1D. However, these benefits will require that AI-driven tools become part of routine clinical care for people with T1D and their HCPs, including validation in clinical studies and regulatory pathways. Just as important is the need for training and education in the application of AI-CDSS to achieve the outcomes that match the significant potential of these technologies. Key messages • AI technologies have the capability to provide data-driven, personalized treatment recommendations for people with T1D. • AI-CDSS can assist HCPs by analyzing patient data to offer insights and recommendations for treatment adjustment in T1D, on an individual basis. • To realize the promise of AI-CDSS in T1D, significant challenges exist for the trust and adoption of these AI-driven tools, as well as ensuring equity of access and application, clinical efficacy and regulatory compliance.
    DOI:  https://doi.org/10.1159/000546713
  10. Cardiovasc Drugs Ther. 2025 May 31.
    The Application of Machine Learning in Warfarin Dose Precision for Diabetic Patients Treated with Statins: A Comparative Study.
    Mobina Ghajar, Mandana Sadat Ghafourian, Sara Tarkiani, Atousa Naser Tork, Amin Ramezani, Behrouz Zolfaghari, Mohammad Ghasemi Palangi.
       PURPOSE: To evaluate the impact of statin therapy on warfarin dose requirements in diabetic patients and to assess the performance of various machine learning algorithms in predicting optimal warfarin dosing.
    METHODS: The datasets available for total participants of 628 (216 diabetics and 412 non-diabetic patients) were analyzed. We categorized the patients according to height, weight, gender, race, and age, plasma international normalized ratio (INR) on reported therapeutic dose of warfarin, target INR, warfarin dose, statin therapy, and indications for warfarin. Various models were tested on data of patients from the International Warfarin Pharmacogenetics Consortium (IWPC). Data preprocessing involves structuring and handling missing values. Six predictive models, including least absolute shrinkage and selection operator (LASSO), k-nearest neighbors (KNN), support vector regression (SVR), linear regression (LR), decision tree, and random forest (RF), were employed in predicting optimal warfarin dosage. The best dose for each patient will be predicted using one of the six regression models.
    RESULTS: This comparative study showed that the mean (and the standard deviation) of warfarin dose for diabetic and non-diabetic patients were 38.73 (15.37) and 34.50 (18.27) mg per week, respectively. Furthermore, the impact of various statin they use is considered and patient undergoing atorvastatin and rosuvastatin therapy against the necessity of high dose warfarin if the diabetic patients use lovastatin and fluvastatin.
    CONCLUSION: Diabetic patients under statin therapy, considering the specific statin used, require different warfarin dose. Through the application of advanced machine learning, models as dosing predictors may attenuate the adverse effects of warfarin.
    Keywords:  Diabetics; International normalized ratio (INR); Machine learning (ML); Statins; Thrombotic disorders; Warfarin dosage
    DOI:  https://doi.org/10.1007/s10557-025-07690-5
  11. Int Wound J. 2025 Jun;22(6): e70515
    Machine Learning Insights Into Amputation Risk: Evaluating Wound Classification Systems in Diabetic Foot Ulcers.
    Farideh Mostafavi, Mohammad Reza Amini, Yadollah Mehrabi, Ensieh Nasli Esfahani, Seyed Saeed Hashemi Nazari.
      This study compares the performance of various wound classification systems to determine which system most effectively predicts amputation risk in diabetic foot ulcer (DFU) patients. Additionally, it identifies the key clinical and socioeconomic factors that influence this risk. A total of 616 DFUs from 400 outpatient participants in a prospective cohort study were followed over 6 months. Ten machine learning (ML) algorithms were employed to evaluate the predictive accuracy of various wound classification systems. The SHapley Additive exPlanations (SHAP) method was used to interpret the predictions of the selected model. The DIAFORA (diabetic foot risk assessment) and WIFI (Wound, Ischaemia and foot Infection) classification systems demonstrated the highest predictive power for predicting amputation within 6 months. SHAP analysis revealed that wound penetration to bone, presence of ischaemia and infection, renal failure, delayed first specialist visit, longer diabetes duration, high baseline HbA1c, low education levels and high body mass index were significant risk factors for amputation. Conversely, higher education levels served as a protective factor. Occupation showed variable effects, with private-sector employment associated with increased risk, while being a housewife was linked to lower risk. Infection and ischaemia are significant factors affecting DFU outcomes. Addressing treatment adherence barriers and implementing tailored interventions that consider patients' occupational needs can reduce amputation rates.
    Keywords:  amputation; diabetic foot; machine learning; ulcer
    DOI:  https://doi.org/10.1111/iwj.70515
This page is being maintained by Thomas Krichel. It was last updated on 2025‒06‒08 at 3:35.