bims-aukdir Biomed News
on Automated knowledge discovery in diabetes research
Issue of 2025–08–10
fourteen papers selected by
Mott Given
  1. Automated detection of diabetic retinopathy lesions in ultra-widefield fundus images using an attention-augmented YOLOv8 framework.
  2. The role of artificial intelligence in diabetic retinopathy screening in type 1 diabetes: A systematic review.
  3. Integration of Immunometabolic Composite Indices and Machine Learning for Diabetic Retinopathy Risk Stratification: Insights from NHANES 2011 - 2020.
  4. AI-Assisted identification of sex-specific patterns in diabetic retinopathy using retinal fundus images.
  5. Sharper Detection: Enhanced Techniques for Diabetic Retinopathy Grading.
  6. Artificial Intelligence and Machine Learning in Diabetic Foot Ulcer Care: Advances in Diagnosis, Treatment, Prognosis, and Novel Therapeutic Strategies.
  7. Machine learning for diabetic foot care: accuracy trends and emerging directions in healthcare AI.
  8. Development and Validation of DIANA (Diabetes Novel Subgroup Assessment tool): A web-based precision medicine tool to determine type 2 diabetes endotype membership and predict individuals at risk of microvascular disease.
  9. Metabolomics-based prediction model for diabetes: A comprehensive analysis of biomarkers and machine learning approaches.
  10. Predicting Diabetes Using Convolutional Neural Networks and EKG Entropy Analysis.
  11. NeSyDPP-4: discovering DPP-4 inhibitors for diabetes treatment with a neuro-symbolic AI approach.
  12. ICU Length of Stay Prediction for Patients with Diabetes Using Machine Learning and Clinical Notes.
  13. Comparing Deterministic and Stochastic Reinforcement Learning for Glucose Regulation in Type 1 Diabetes.
  14. Diaproteo: A supervised learning framework for early detection of diabetes mellitus based on proteomic profiles.
  1. Front Cell Dev Biol. 2025 ;13 1608580
    Automated detection of diabetic retinopathy lesions in ultra-widefield fundus images using an attention-augmented YOLOv8 framework.
    Lei-Si Hu, Jie Wang, Heng-Ming Zhang, Hai-Yu Huang.
       Objective: To enhance the automatic detection precision of diabetic retinopathy (DR) lesions, this study introduces an improved YOLOv8 model specifically designed for the precise identification of DR lesions.
    Method: This study integrated two attention mechanisms, convolutional exponential moving average (convEMA) and convolutional simple attention module (convSimAM), into the backbone of the YOLOv8 model. A dataset consisting of 3,388 ultra-widefield (UWF) fundus images obtained from patients with DR, each with a resolution of 2,600 × 2048 pixels, was utilized for both training and testing purposes. The performances of the three models-YOLOv8, YOLOv8+ convEMA, and YOLOv8+ convSimAM-were systematically compared.
    Results: A comparative analysis of the three models revealed that the original YOLOv8 model suffers from missed detection issues, achieving a precision of 0.815 for hemorrhage spot detection. YOLOv8+ convEMA improved hemorrhage detection precision to 0.906, while YOLOv8+ convSimAM achieved the highest value of 0.910, demonstrating the enhanced sensitivity of spatial attention. The proposed model also maintained comparable precision in detecting hard exudates while improving recall to 0.804. It demonstrated the best performance in detecting cotton wool spots and the epiretinal membrane. Overall, the proposed method provides a fine-tuned model specialized in subtle lesion detection, providing an improved solution for DR lesion assessment.
    Conclusion: In this study, we proposed two attention-augmented YOLOv8 models-YOLOv8+ convEMA and YOLOv8+ convSimAM-for the automated detection of DR lesions in UWF fundus images. Both models outperformed the baseline YOLOv8 in terms of detection precision, average precision, and recall. Among them, YOLOv8+ convSimAM achieved the most balanced and accurate results across multiple lesion types, demonstrating an enhanced capability to detect small, low-contrast, and structurally complex features. These findings support the effectiveness of lightweight attention mechanisms in optimizing deep learning models for high-precision DR lesion detection.
    Keywords:  YOLOv8; attention mechanisms; automatic detection; diabetic retinopathy; ultra-widefield fundus images
    DOI:  https://doi.org/10.3389/fcell.2025.1608580
  2. J Diabetes Complications. 2025 Jul 25. pii: S1056-8727(25)00192-8. [Epub ahead of print]39(10): 109139
    The role of artificial intelligence in diabetic retinopathy screening in type 1 diabetes: A systematic review.
    Francesco Sacchini, Stefano Mancin, Giovanni Cangelosi, Sara Morales Palomares, Gabriele Caggianelli, Francesco Gravante, Fabio Petrelli.
       BACKGROUND/OBJECTIVES: Diabetic retinopathy (DR) is one of the leading causes of blindness in adults worldwide and represents a critical complication in both type 1 (T1D) and type 2 (T2D) diabetes. Artificial Intelligence (AI) offers a promising opportunity to enhance both the accuracy of screening and the efficiency of ongoing care management, assisting healthcare providers in mitigating the incidence and complications of DR.
    METHODS: Systematic review of the literature was conducted following PRISMA guidelines. Searches were performed using PubMed-Medline, Scopus, and Embase databases, with the protocol registered on the Open Science Framework (OSF) database: (doi.org/10.17605/OSF.IO/TJ9UH). A predefined search strategy utilizing Boolean operators was applied, and two researchers independently selected articles, with a third resolving any discrepancies.
    RESULTS: Of the 2127 articles identified, 8 studies were included. The results highlighted that AI is particularly effective in enhancing the DR screening process in patients with T1D, offering rapid and reliable analysis. Healthcare providers reported positive feedback, noting its significant contribution to improving patient management.
    CONCLUSIONS: The integration of AI into DR care pathways shows substantial potential for improving early diagnosis and disease management, particularly for patients with T1D. Further research is required to optimize AI implementation and ensure its positive and sustainable impact on public health.
    Keywords:  Artificial intelligence; Diabetic retinopathy screening; Health technology assessment; Public health; Type 1 diabetes
    DOI:  https://doi.org/10.1016/j.jdiacomp.2025.109139
  3. Ophthalmol Sci. 2025 Nov-Dec;5(6):5(6): 100854
    Integration of Immunometabolic Composite Indices and Machine Learning for Diabetic Retinopathy Risk Stratification: Insights from NHANES 2011 - 2020.
    Cui Xuehao, Wen Dejia, Li Xiaorong.
       Objective: This study aimed to investigate the association between immunometabolic composite indices and diabetic retinopathy (DR) and to develop predictive models using machine learning (ML) techniques to improve early detection and risk stratification for DR.
    Design: A cross-sectional study.
    Subjects and Controls: Data from the National Health and Nutrition Examination Survey 2011-2020 were analyzed, involving 8249 participants categorized into healthy controls (n = 6830), diabetes without retinopathy (n = 918), and DR (n = 501).
    Methods: Immunometabolic indices reflecting insulin resistance, inflammation, and lipid metabolism were evaluated. Multivariate logistic regression models assessed associations with DR, and Bayesian kernel machine regression analyzed nonlinear interactions. Eight ML models, including ensemble methods, were developed to predict DR risk, with feature importance determined by SHapley Additive exPlanations.
    Main Outcome Measures: The primary outcome was DR status, classified according to the ETDRS criteria from fundus photography.
    Results: Key immunometabolic indices, notably Frailty Index (FRAILTY) and fasting serum insulin (FSI), were significantly associated with increased DR risk, whereas the metabolic score for insulin resistance (METS) showed a protective effect. Bayesian kernel machine regression highlighted complex interactions among indices. Machine learning models achieved high predictive accuracy, particularly XGBoost and LightGBM (area under the curve > 0.9). SHapley Additive exPlanations analyses identified FRAILTY, FSI, and METS as the most influential predictors.
    Conclusions: Immunometabolic dysregulation significantly contributes to DR progression beyond traditional risk factors such as hyperglycemia alone. Incorporating immunometabolic indices into predictive models substantially enhances DR risk stratification, facilitating personalized screening and intervention strategies. Machine learning approaches effectively identify high-risk individuals, underscoring their utility in clinical practice for early DR detection and targeted preventive care.
    Financial Disclosures: The author(s) have no proprietary or commercial interest in any materials discussed in this article.
    Keywords:  Bayesian kernel machine regression; Diabetic retinopathy; Immunometabolic indices; NHANES; Risk stratification
    DOI:  https://doi.org/10.1016/j.xops.2025.100854
  4. PLoS One. 2025 ;20(8): e0327305
    AI-Assisted identification of sex-specific patterns in diabetic retinopathy using retinal fundus images.
    Parsa Delavari, Gulcenur Ozturan, Eduardo V Navajas, Ozgur Yilmaz, Ipek Oruc.
      Diabetic retinopathy (DR) is a microvascular complication of diabetes that can lead to blindness if left untreated. Regular monitoring is crucial for detecting early signs of referable DR, and the progression to moderate to severe non-proliferative DR, proliferative DR (PDR), and macular edema (ME), the most common cause of vision loss in DR. Currently, aside from considerations during pregnancy, sex is not factored into DR diagnosis, management or treatment. Here we examine whether DR manifests differently in male and female patients, using a dataset of retinal images and leveraging convolutional neural networks (CNN) integrated with explainable artificial intelligence (AI) techniques. To minimize confounding variables, we curated 2,967 fundus images from a larger dataset of DR patients acquired from EyePACS, matching male and female groups for age, ethnicity, severity of DR, and hemoglobin A1C levels. Next, we fine-tuned two pre-trained VGG16 models-one trained on the ImageNet dataset and another on a sex classification task using healthy fundus images-achieving AUC scores of 0.72 and 0.75, respectively, both significantly above chance level. To uncover how these models distinguish between male and female retinas, we used the Guided Grad-CAM technique to generate saliency maps, highlighting critical retinal regions for correct classification. Saliency maps showed CNNs focused on different retinal regions by sex: the macula in females, and the optic disc and peripheral vasculature along the arcades in males. This pattern differed noticeably from the saliency maps generated by CNNs trained on healthy eyes. These findings raise the hypothesis that DR may manifest differently by sex, with women potentially at higher risk for developing ME, as opposed to men who may be at greater risk for PDR.
    DOI:  https://doi.org/10.1371/journal.pone.0327305
  5. Stud Health Technol Inform. 2025 Aug 07. 329 568-572
    Sharper Detection: Enhanced Techniques for Diabetic Retinopathy Grading.
    Ali Rafiei, Osmar R Zaiane.
      Diabetic retinopathy (DR) is a leading cause of blindness, and deep learning methods are extensively used for automated DR grading. This paper demonstrates how applying techniques such as oversampling, data augmentation, K-fold cross-validation, learning rate scheduling, and early stopping can significantly improve performance for models such as ResNet18 and ResNet50. Using the APTOS and EyePACS datasets, our results reveal a substantial accuracy improvement compared to prior methods, which rely on oversampling and data augmentation as baseline techniques. We obtained an accuracy of 97.78% on APTOS with ResNet18 and 93.80% on EyePACS with ResNet50, outperforming the baseline. The techniques presented offer a valuable framework that can complement and enhance the state-of-the-art methods for DR grading and beyond.
    Keywords:  Deep Learning; Diabetic Retinopathy; Model Improvement; ResNet
    DOI:  https://doi.org/10.3233/SHTI250904
  6. J Diabetes Sci Technol. 2025 Aug 03. 19322968251363632
    Artificial Intelligence and Machine Learning in Diabetic Foot Ulcer Care: Advances in Diagnosis, Treatment, Prognosis, and Novel Therapeutic Strategies.
    Abdulhamit Misir.
       BACKGROUND: Diabetic foot ulcers (DFUs) affect 19% to 34% of individuals with diabetes during their lifetime and account for over one million nontraumatic lower-limb amputations annually. Standard care often fails to detect early, subclinical changes, leading to delayed intervention and high mortality rates. This review examines how artificial intelligence (AI) and machine learning (ML) can extract complex patterns from diverse data modalities to advance DFU care.
    METHODS: We examined AI/ML applications in DFU care across four domains: diagnosis (automated image and thermogram classification, biomechanical risk stratification), treatment optimization (AI-driven offloading prescriptions, tele-rehabilitation, molecularly informed wound care), prognosis (neural network and random forest models for risk assessment), and novel strategy development (generative AI, transcriptomic target discovery, wearable digital biomarkers).
    RESULTS: Artificial intelligence/ML methodologies have demonstrated promising results in DFU image and thermogram analysis, with reported accuracies ranging from 81-97% across different studies. Biomechanical ML models show potential for dynamic risk stratification, and prognostic models achieve moderate performance with area under the curve values around 0.74-0.82. Generative AI approaches have shown promise for data augmentation, improving segmentation performance in limited datasets.
    CONCLUSION: Despite promising advances, several challenges impede clinical translation, including data standardization, model explainability, regulatory compliance, clinical workflow integration, prospective validation, and equitable implementation. Collaborative efforts among clinicians, data scientists, regulators, and patients are essential to translate AI-driven innovations into routine DFU management, potentially reducing amputations and improving outcomes for this global health burden.
    Keywords:  artificial intelligence; biomechanics; clinical decision support systems; diabetic foot ulcers; machine learning; prognostic modeling; thermography; wearable technology
    DOI:  https://doi.org/10.1177/19322968251363632
  7. Front Public Health. 2025 ;13 1613946
    Machine learning for diabetic foot care: accuracy trends and emerging directions in healthcare AI.
    Pei-Chun Lin, Tsai-Chung Li, Tzu-Hsuan Huang, Ying-Lin Hsu, Wen-Chao Ho, Jia-Lang Xu, Ching-Liang Hsieh, Zih-En Jhang.
       Background: Diabetic foot is a common and debilitating complication of diabetes that significantly impacts patients' quality of life and frequently leads to amputation. In parallel, artificial intelligence (AI), particularly machine learning (ML), has emerged as a powerful tool in healthcare, offering novel solutions for disease prediction, monitoring, and management. Despite growing interest, a systematic overview of machine learning applications in diabetic foot research is still lacking.
    Objective: This study aims to systematically analyze recent literature to identify key trends, focus areas, and methodological approaches in the application of machine learning to diabetic foot research.
    Data sources: A comprehensive literature search was conducted across three major databases: Web of Science (WoS), IEEE Xplore, and PubMed. The search targeted peer-reviewed journal articles published between 2020 and 2024 that focused on the intersection of machine learning and diabetic foot management.
    Eligibility criteria and study selection: Articles were included if they were indexed in the Science Citation Index (SCI) or Social Sciences Citation Index (SSCI), published in English. They explored the use of machine learning in diabetic foot-related applications. After removing duplicates and irrelevant entries, 25 original research articles were included for review.
    Results: There has been a steady increase in publications related to machine learning in diabetic foot research over the past 5 years. Among the 25 studies included, image analysis was the most prevalent theme (12 articles), dominated by thermal imaging applications (10 articles). General clinical imaging was less common (2 articles). Seven studies focused on structured clinical data analysis, while six explored IoT-based approaches such as smart insoles with integrated sensors for real-time foot monitoring. Citation analysis showed that Computers in Biology and Medicine and Sensors had the highest average citation rates among journals publishing multiple relevant studies.
    Conclusion: The integration of machine learning into diabetic foot research is rapidly evolving; it is characterized by growing diversity in data modalities and analytical techniques. Thermal imaging remains a key area of interest, while IoT innovations show promise for clinical translation. Future studies should aim to incorporate deep learning, genomic data, and large language models to further enhance the scope and clinical utility of diabetic foot research.
    Keywords:  artificial intelligence in healthcare; clinical data analysis; diabetic foot; internet of things; machine learning; thermal imaging
    DOI:  https://doi.org/10.3389/fpubh.2025.1613946
  8. PLOS Digit Health. 2025 Aug;4(8): e0000702
    Development and Validation of DIANA (Diabetes Novel Subgroup Assessment tool): A web-based precision medicine tool to determine type 2 diabetes endotype membership and predict individuals at risk of microvascular disease.
    Viswanathan Baskar, Mani Arun Vignesh, Sumanth C Raman, Arun Jijo, Bhavadharini Balaji, Nico Steckhan, Lena Maria Klara Roth, Moneeza K Siddiqui, Saravanan Jebarani, Ranjit Unnikrishnan, Viswanathan Mohan, Ranjit Mohan Anjana.
       BACKGROUND: Previous research has identified four distinct endotypes of type 2 diabetes in Asian Indians, which include Severe Insulin Deficient Diabetes (SIDD), Combined Insulin Resistant and Deficient Diabetes (CIRDD), Insulin Resistance and Obese Diabetes (IROD), and Mild Age-related Diabetes (MARD). DIANA (Diabetes Novel Subgroup Assessment) is an online precision medicine tool that can predict endotype membership of type 2 diabetes and individual risk for retinopathy and nephropathy.
    METHODOLOGY: The DIANA tool determines subgroup membership using a machine learning model (support vector machine) on T2D subgroups in the Asian Indian population. We used a support vector machine (SVM) model to classify type 2 diabetes patient endotypes, and the model is trained based on k-fold cross-validation. Its performance was compared with an algorithm determined based on conditional pre-determined cut-offs and weights for each clinical feature [age at diagnosis, BMI, waist, HbA1c, Serum Triglycerides, HDL-Cholesterol, (C-peptide fasting, C-peptide stimulated) - optional. This study employed local interpretable model-agnostic explanations (LIME) and SHapley Additive exPlanations (SHAP) to demystify the endotype prediction model. A random forest model was built to assess an individual's risk for nephropathy and retinopathy based on individual risk algorithms.
    FINDINGS: The SVM model has relatively high accuracy, specificity, sensitivity, and precision values compared to conditional pre-determined cut-offs 98% vs 63.6%, 99.8% vs 88%, 98.5% vs 65.1%, and 98.7% vs 63.4%. Clinician face value validation of the prediction by the SVM model reported an accuracy, specificity, sensitivity and precision compared to conditional pre-determined cut-offs 97% vs 85%, 95.3% vs 63%, 95.8% vs 73%, and 98.9% vs 66.9%. Additionally, our study demonstrated the impact of features on ML models through LIME and SHAP analyses. The accuracy of the random forest risk prediction model for nephropathy and retinopathy was 89.6% (p < 0.05) and 78.4% (p < 0.05), respectively.
    CONCLUSION: We conclude that, DIANA is an accurate, clinically explainable AI tool that clinicians can use to make informed decisions on risk assessment and provide precision management to individuals with new-onset type 2 diabetes.
    DOI:  https://doi.org/10.1371/journal.pdig.0000702
  9. Diabetes Res Clin Pract. 2025 Aug 04. pii: S0168-8227(25)00417-6. [Epub ahead of print] 112403
    Metabolomics-based prediction model for diabetes: A comprehensive analysis of biomarkers and machine learning approaches.
    Doaa Farid, Farhana Saleh, Tareq Mohammed, Atika Nigar, Abderrahmane Maaradji.
       AIMS: To develop a prediction model for diabetes using metabolomics data and to evaluate various machine learning approaches and identify the most effective framework for disease prediction.
    METHODS: A comprehensive analysis was conducted on the Qatar Biobank dataset comprising metabolomics profiles, instrument measurements, and clinical diagnoses from 450 Qatari nationals. Targeted metabolites were selected based on correlation strength with diabetes status. Five machine learning models (Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, and Neural Network) were evaluated for their predictive performance using metrics including accuracy, precision, recall, F1 score, and ROC AUC.
    RESULTS: Among 450 individuals, 9.33 % (n = 42) were diagnosed with diabetes. Correlation analysis identified 140 metabolites significantly associated with diabetes status (p < 0.05). The most strongly correlated metabolites included glucose (r = 0.281, p < 0.0001), mannose (r = 0.247, p < 0.0001), and 1,5-anhydroglucitol (r = -0.297, p < 0.0001). Logistic Regression demonstrated superior performance with the highest accuracy (93.3 %), F1 score (0.625), and ROC AUC (0.941) compared to other models.
    CONCLUSION: Metabolomics data can effectively predict diabetes status, with logistic regression providing the optimal balance of performance and interpretability. The identified metabolites offer potential biomarkers for early diabetes detection and monitoring. This model could serve as a valuable tool for clinical risk assessment and personalized preventive interventions.
    Keywords:  Biomarkers; Diabetes mellitus; Logistic regression; Machine learning; Metabolomics; Prediction model
    DOI:  https://doi.org/10.1016/j.diabres.2025.112403
  10. Stud Health Technol Inform. 2025 Aug 07. 329 946-950
    Predicting Diabetes Using Convolutional Neural Networks and EKG Entropy Analysis.
    Sayonara de Fátima F Barbosa, Fabio J Silva.
      Heart Rate Variability (HRV) is associated with diabetic complications. This analysis can quantify changes in heart rate variability, and it may help detect early alterations in diabetes. This study aimed to design and validate a Convolutional Neural Network (CNN) empowered with entropy metrics (RNC-Rica) for the diagnosis of diabetes through EKG recordings. The RNC-Rica model thus uses CNN architecture for two-dimensional convolution based feature extraction from EKG and simultaneous study of age, HRV measures and entropy measures. From these setups five test setups were evaluated by integrating combinations of convolutional layers and entropy. Through this approach, Test 4 was found to yield the best results (accuracy: 70.3%; sensitivity: 78.4%; specificity: 62.0%; positive predictive value: 68.0%; F1-Score: 72.8%), and the entropy metrics illustrate an improvement in model stability with entropy metrics. The model was capable of early detection in subclinical stages of diabetes. The results indicate that the regularized nearest centroid-Rica model augmented with entropy metrics is an effective tool for the early diagnosis of diabetes via EKG, with high sensitivity, and also statistically significant for clinical classification.
    Keywords:  Artificial Intelligence; Convolutional Neural Network; Diabetes; Health Informatics
    DOI:  https://doi.org/10.3233/SHTI250979
  11. Front Bioinform. 2025 ;5 1603133
    NeSyDPP-4: discovering DPP-4 inhibitors for diabetes treatment with a neuro-symbolic AI approach.
    Delower Hossain, Ehsan Saghapour, Jake Y Chen.
       Introduction: Diabetes Mellitus (DM) constitutes a global epidemic and is one of the top ten leading causes of mortality (WHO, 2019), projected to rank seventh by 2030. The US National Diabetes Statistics Report (2021) states that 38.4 million Americans have diabetes. Dipeptidyl Peptidase-4 (DPP-4) is an FDA-approved target for the treatment of type 2 diabetes mellitus (T2DM). However, current DPP-4 inhibitors may cause adverse effects, including gastrointestinal issues, severe joint pain (FDA safety warning), nasopharyngitis, hypersensitivity, and nausea. Moreover, the development of novel drugs and the in vivo assessment of DPP-4 inhibition are both costly and often impractical. These challenges highlight the urgent need for efficient in-silico approaches to facilitate the discovery and optimization of safer and more effective DPP-4 inhibitors.
    Methodology: Quantitative Structure-Activity Relationship (QSAR) modeling is a widely used computational approach for evaluating the properties of chemical substances. In this study, we employed a Neuro-symbolic (NeSy) approach, specifically the Logic Tensor Network (LTN), to develop a DPP-4 QSAR model capable of identifying potential small-molecule inhibitors and predicting bioactivity classification. For comparison, we also implemented baseline models using Deep Neural Networks (DNNs) and Transformers. A total of 6,563 bioactivity records (SMILES-based compounds with IC50 values) were collected from ChEMBL, PubChem, BindingDB, and GTP. Feature sets used for model training included descriptors (CDK Extended-PaDEL), fingerprints (Morgan), chemical language model embeddings (ChemBERTa-2), LLaMa 3.2 embedding features, and physicochemical properties.
    Results: Among all tested configurations, the Neuro-symbolic QSAR model (NeSyDPP-4) performed best using a combination of CDK extended and Morgan fingerprints. The model achieved an accuracy of 0.9725, an F1-score of 0.9723, an ROC AUC of 0.9719, and a Matthews correlation coefficient (MCC) of 0.9446. These results outperformed the baseline DNN and Transformer models, as well as existing state-of-the-art (SOTA) methods. To further validate the robustness of the model, we conducted an external evaluation using the Drug Target Common (DTC) dataset, where NeSyDPP-4 also demonstrated strong performance, with an accuracy of 0.9579, an AUC-ROC of 0.9565, a Matthews Correlation Coefficient (MCC) of 0.9171, and an F1-score of 0.9577.
    Discussion: These findings suggest that the NeSyDPP-4 model not only delivered high predictive performance but also demonstrated generalizability to external datasets. This approach presents a cost-effective and reliable alternative to traditional vivo screening, offering valuable support for the identification and classification of biologically active DPP-4 inhibitors in the treatment of type 2 diabetes mellitus (T2DM).
    Keywords:  DPP-4; QSAR; deep learning; drug discovery; machine learning; neuro-symbolic artificial intelligence
    DOI:  https://doi.org/10.3389/fbinf.2025.1603133
  12. Stud Health Technol Inform. 2025 Aug 07. 329 1130-1134
    ICU Length of Stay Prediction for Patients with Diabetes Using Machine Learning and Clinical Notes.
    Ling Zheng, Yuansi Hu, Andrew Catapano, Guozhong Zheng.
      Diabetes, a chronic disease, often leads to poor health outcomes and increased healthcare costs, particularly for patients admitted to ICU. Accurate early prediction of ICU length of stay (LOS) is vital for hospital resource management and patient outcome improvement. This study developed predictive models for ICU LOS in diabetic patients by integrating unstructured clinical notes with structured data, including demographics, diagnoses, lab tests, and ICU chart events. Using machine learning techniques, we addressed two tasks: predicting ICU days and classifying stays as long (≥10 days) or short (<10 days). Neural network using Doc2Vec word embedding achieved the best regression performance with an R2 of 0.3626 and mean absolute error of 1.54 days for short stays. For classification, logistic regression with TF-IDF text encoding achieved a best accuracy of 0.875. These results demonstrate the potential of combining structured and unstructured data with machine learning to enhance early ICU LOS predictions, supporting clinical decision-making and resource optimization.
    Keywords:  Clinical Notes; ICU; Length of Stay; Machine Learning; Word Embedding
    DOI:  https://doi.org/10.3233/SHTI251015
  13. Stud Health Technol Inform. 2025 Aug 07. 329 1039-1043
    Comparing Deterministic and Stochastic Reinforcement Learning for Glucose Regulation in Type 1 Diabetes.
    David Timms, Chirath Hettiarachchi, Hanna Suominen.
      Type 1 Diabetes (T1D) is a chronic condition affecting millions worldwide, requiring external insulin administration to regulate blood glucose levels and prevent serious complications. Artificial Pancreas Systems (APS) for managing T1D currently rely on manual input, which adds a cognitive burden on people with T1D and their carers. Research into alleviating this burden through Reinforcement Learning (RL) explores enabling the APS to autonomously learn and adapt to the complex dynamics of blood glucose regulation, demonstrating improvements in in-silico evaluations compared to traditional clinical approaches. This evaluation study compared the primary polarities of RL for glucose regulation, namely, stochastic (e.g., Proximal Policy Optimization (PPO) and deterministic (e.g., Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithms in-silico using quantitative and qualitative methods, patient specific clinical metrics, and the adult and adolescent cohorts of the U.S. Food and Drug Administration approved UVA/PADOVA 2008 model. Although the behavior of TD3 was easier to interpret, it did not typically outperform PPO, thereby challenging assessing their safety and suitability. This conclusion highlights the importance of improving RL algorithms in APS applications for both interpretability and predictive performance in future research.
    Keywords:  Artificial Pancreas; Deep Learning; Evaluation Study; Type 1 Diabetes
    DOI:  https://doi.org/10.3233/SHTI250997
  14. Digit Health. 2025 Jan-Dec;11:11 20552076251362281
    Diaproteo: A supervised learning framework for early detection of diabetes mellitus based on proteomic profiles.
    Hamza Shahab Awan, Fahad Alturise, Tamim Alkhalifah, Yaser Daanial Khan.
      Diabetes mellitus (DM) is a chronic metabolic disease that affects millions of people worldwide, posing major health risks and financial challenges. Early diagnosis and treatment are essential for reducing complications and improving patient outcomes. This research explores the application of supervised algorithms to predict DM using a variety of datasets such as clinical features, genetic markers, and lifestyle variables. This study proposes novel approaches and evaluates prediction models with classic machine learning algorithms and cutting-edge deep learning architecture. Performance metrics (accuracy, precision, recall, F1 score) reveal that the Extra Trees model for the independent test and Convolutional Neural Network (CNN) for 10-fold cross-validation, achieving 91.52% accuracy with an F1 score of 0.91 (Extra Trees) and 87.03% accuracy with an F1 score of 84.82% (CNN). In addition, other evaluation indicators demonstrated that the Extra Trees algorithm outperformed others, achieving the highest accuracy on the independent test. Our study shows that machine learning and deep learning approaches may accurately predict DM, demonstrating the potential for early intervention and personalized healthcare strategies.
    Keywords:  Biological computational; bioinformatics; diabetes mellitus; neural network; predictive analytics
    DOI:  https://doi.org/10.1177/20552076251362281
This page is being maintained by Thomas Krichel. It was last updated on 2025‒08‒10 at 0:16.