bims-aukdir 2025-11-02 papers

bims-aukdir

Biomed News

on Automated knowledge discovery in diabetes research

Issue of 2025–11–02
27 papers selected by
Mott Given

Diabetes Prediction Using Feature Selection Algorithms and Boosting-Based Machine Learning Classifiers.
FISM: harnessing deep learning and reinforcement learning for precision detection of microaneurysms and retinal exudates for early diabetic retinopathy diagnosis.
Describing the Performance and the Infrastructure Requirements of the Existing Artificial Intelligence (AI)-Based Diabetic Retinopathy (DR) Screening Algorithms for Diabetic Patients: an Umbrella Review.
Response to: "Letter to the editor "The role of artificial intelligence in diabetic retinopathy screening in type 1 diabetes: A systematic review".
A machine learning model for predicting the risk of diabetic nephropathy in individuals with type 2 diabetes mellitus.
Interpretable Machine Learning Framework for Diabetes Prediction: Integrating SMOTE Balancing with SHAP Explainability for Clinical Decision Support.
In-Context Learning for Data-Efficient Diabetic Retinopathy Detection via Multimodal Foundation Models.
DRCNN-Lesion Proxy: a hybrid CNN with lesion-inspired feature simulation for diabetic retinopathy severity classification.
Artificial Intelligence and Diabetes Prevention.
Deploying machine learning models in clinical settings: a real-world feasibility analysis for a model identifying adult-onset type 1 diabetes initially classified as type 2.
A CNN-Transformer fusion network for Diabetic retinopathy image classification.
[An adaptive multi-label classification model for diabetic retinopathy lesion recognition].
Diabetic Retinopathy Screening Approaches in Developing Countries: A Systematic Review and Meta-Analysis.
Pixel-Attention W-Shaped Network for Joint Lesion Segmentation and Diabetic Retinopathy Severity Staging.
Dynamics of the atherogenic index of plasma and diabetes conversion: Insights from machine learning and longitudinal pattern analysis in middle-aged and older Chinese adults with prediabetes.
Leveraging XGBoost and explainable AI for accurate prediction of type 2 diabetes.
Predicting Mild Cognitive Impairment in Type 2 Diabetes: A Machine Learning Approach.
Enhancing AI-based diabetic retinopathy screening in low- and middle-income countries with synthetic data.
Refine XGBoost with SHAP explainability for non-invasive early detection of diabetic kidney disease: Estimated cardiac output as a potential indicator.
Clinical diagnosis of diabetes using machine learning and surface-enhanced Raman spectroscopy liquid biopsy: an exploratory study.
Automated FAZ segmentation and diabetic retinopathy classification using OCTA images.
Triglyceride-glucose index and mortality in congestive heart failure with diabetes: a machine learning predictive model.
Autonomous Artificial Intelligence in Diabetic Retinopathy Testing-Lessons Learned on Successful Health System Adoption.
Reliable biomarkers for diabetic nephropathy using machine learning-assisted contrast-enhanced ultrasonography and clinical characteristics.
[Research on type 2 diabetes prediction algorithm based on photoplethysmography].
Response to Letter to the Editor: Commentary on "Multimodal MRI analysis selecting key brain features for machine learning based classification of diabetic neuropathic pain and phenotypes".
Automatic diagnosis of type 2 diabetes mellitus with mild cognitive impairment using artificial intelligence based on routine T1-weighted MRI.

Diagnostics (Basel). 2025 Oct 17. pii: 2622. [Epub ahead of print]15(20):

Diabetes Prediction Using Feature Selection Algorithms and Boosting-Based Machine Learning Classifiers.

Fatima Rahman, Sheyum Hossain, Jun-Jiat Tiang, Abdullah-Al Nahid.

  Background: Diabetes mellitus is a significant primary global health concern that requires accurate diagnosis at an early stage to prevent severe complications. However, accurate prediction remains challenging due to limited, noisy, and imbalanced datasets. This study proposes a novel machine learning framework for improved diabetes prediction, addressing key challenges such as inadequate feature selection, class imbalance, and data preprocessing. Methods: This proposed work systematically evaluates five feature selection algorithms-Recursive Feature Elimination, Grey Wolf Optimizer, Particle Swarm Optimizer, Genetic Algorithm, and Boruta-using cross-validation and SHAP analysis to enhance feature interpretability. Classification is performed using two boosting algorithms: the light gradient boosting machine algorithm (LGBM) and the extreme gradient boosting algorithm (XGBoost). Results: The proposed framework, using the five most important features selected by the Boruta feature selection algorithm, outperformed other configurations with the LightGBM classifier, achieving an accuracy of 85.16%, an F1-score of 85.41%, and a 54.96% reduction in training time. Conclusions: Additionally, we have benchmarked our approach against recent studies and validated its effectiveness on both the Pima Indian Diabetes Dataset and the newly released DiaHealth dataset, demonstrating robust and accurate early diabetes detection across diverse clinical datasets. This approach offers a cost-effective, interpretable, and clinically relevant solution for early diabetes detection by reducing the number of input features, providing transparent feature importance, and achieving high predictive accuracy with efficient model training.

Keywords:  boosting classifier algorithms; diabetes prediction; feature selection algorithms (FSAs); machine learning; medical diagnostics

DOI:  https://doi.org/10.3390/diagnostics15202622
BioData Min. 2025 Oct 30. 18(1): 75

FISM: harnessing deep learning and reinforcement learning for precision detection of microaneurysms and retinal exudates for early diabetic retinopathy diagnosis.

Abbas Rehman, Gu Naijie, Stephen Ojo, Thomas I Nathaniel, Nagwan Abdel Samee, Muhammad Umer, Mona M Jamjoom.

  Diabetic retinopathy (DR) is a primary cause of blindness globally and its treatment and management depend on accurate and timely identification. Current approaches for DR detection and segmentation repeatedly fall short in accuracy and sturdiness highlighting the essential for advanced computational methods. In this study propose a deep learning model Fundus Images Segmentation Model (FISM) designed to precisely detect microaneurysms and retinal exudates dangerous indicators of DR. Employing the Diabetic Retinopathy Dataset (DDR), our model utilizes both the segmentation and grading subsets, comprising over 13,000 fundus images annotated with comprehensive lesion-level and DR severity information, enabling robust training for both detection and classification tasks. The preprocessing pipeline contains band separation generative adversarial network (GAN) based data augmentation and extensive normalization techniques. The FISM architecture is derived from the Segment Anything Model (SAM) exclusively integrating transformer layers and patch embedding techniques. The model begins with patch embedding followed by transformer blocks to capture both local and global relationships within retinal images. The architecture employs transfer learning, domain-specific fine-tuning customized loss functions and attention mechanisms to optimize feature extraction and segmentation accuracy. The image encoder and Mask decoder modules work in tandem to transform input retinal images into precise segmentation Masks, highlighting regions affected by DR. Beyond deep learning, the framework also integrates reinforcement learning to constructively direct the exploration of regions of interest so that the model is capable of highlighting areas of interest to a diagnosis. This form of adaptive attention is an improvement in the precision of detection and computational cost. Results show that FISM outperforms state-of-the-art methods, achieving 96.32% accuracy, 95.14% precision, 95.25% recall and a 96.33% F1-score. The model demonstrates an AUC of 96.32%, specificity of 94.13%, segmentation Dice coefficient of 94.21% and IoU of 96.01%. These metrics indicate superior performance in both detection and segmentation tasks for early diabetic retinopathy diagnosis.

Keywords:  Attention mechanism; Data augmentation; Deep learning; Diabetic retinopathy; FISM; Generative adversarial networks; Image segmentation; Microaneurysms; Reinforcement learning; Retinal exudates

DOI:  https://doi.org/10.1186/s13040-025-00485-2
J Med Syst. 2025 Oct 30. 49(1): 149

Describing the Performance and the Infrastructure Requirements of the Existing Artificial Intelligence (AI)-Based Diabetic Retinopathy (DR) Screening Algorithms for Diabetic Patients: an Umbrella Review.

Rachel Kabunga, Justus Asasira, Sheilah Njuki, Atwine Daniel, Katharine Morley, Michael Morley, Fred Kaggwa, Justin C Cikomola, Arunga Simon.

  AI-based diabetic retinopathy (DR) screening algorithms have been evaluated in many countries and have shown promise in expanding access to screening, especially in low- and middle-income countries (LMICs). However, the literature lacks guidance on which algorithms are best suited for these settings. This umbrella review summarizes current evidence on the performance, infrastructure needs, and global implementation of AI-based DR screening tools. Following the Preferred Reporting Items for Systematic Review (PRISMA) guidelines, systematic reviews were identified through searches in PubMed, Embase, ScienceDirect, Scopus, and Google Scholar up to April 18, 2024. Eligible studies were reviewed, and findings were presented in tables and graphics. Twenty systematic reviews were included. Most algorithms were developed, validated, and used in high-income countries, with none developed or implemented in Africa. More than 400 algorithms were identified, of which 161 had some form of clinical validation, and 31 were validated in real-world settings. Sensitivity ranged from 66.0% to 100.0%, specificity from 59.5% to 98.7%, and AUROC from 87.8% to 99.1%. Only 12 algorithms have received regulatory approval, and 11 of them are currently used in clinical practice. AI-based DR screening models hold promise as diagnostic tools across diverse clinical settings, particularly where ophthalmic resources are limited. However, successful implementation depends on appropriate infrastructure, local validation, and regulatory support. Addressing the significant gaps in algorithm development and validation in Africa is essential to ensure equitable access and effective use of AI in DR screening.

Keywords:  Artificial intelligence; Deep learning screening; Diabetic retinopathy; Machine learning; Screening

DOI:  https://doi.org/10.1007/s10916-025-02280-2
J Diabetes Complications. 2025 Oct 06. pii: S1056-8727(25)00233-8. [Epub ahead of print] 109180

Response to: "Letter to the editor "The role of artificial intelligence in diabetic retinopathy screening in type 1 diabetes: A systematic review".

Giovanni Cangelosi, Francesco Sacchini, Stefano Mancin, Sara Morales Palomares, Gabriele Caggianelli, Francesco Gravante, Fabio Petrelli.

DOI: https://doi.org/10.1016/j.jdiacomp.2025.109180
Front Endocrinol (Lausanne). 2025 ;16 1587932

A machine learning model for predicting the risk of diabetic nephropathy in individuals with type 2 diabetes mellitus.

Tingting Li, Jinbo Chen, Xin Zhang, Kaiwen Wang, Xuesen Zhao, Yi Cao, Zhen Xu, Shiyue Wang, Peng Su, Xiaoyan He, Yang Yang, Xiaolu Cao, Xiaohua Liang, Dong Ma.

   Introduction: Diabetic kidney disease (DKD) represents the predominant form of chronic kidney disease (CKD) linked with diabetes mellitus. The application of artificial intelligence holds promise for delaying renal deterioration and decreasing treatment expenses by facilitating early detection and intervention. This is contingent upon the development of an efficient and user-friendly model for predicting DKD risk in diabetic individuals. In this study, leveraging extensive clinical datasets, we sought to develop and validate a predictive model employing machine learning techniques to assess the risk of DKD in patients with type 2 diabetes mellitus (T2DM).
Research design and methods: We conducted a retrospective collection of clinical data from 10,057 patients diagnosed with T2DM at Shijiazhuang Second Hospital. A random selection of 15% of these patients (n=1,508) was utilized for external validation. The remaining 8,549 patients were divided into a training set (n = 5,985) and a validation set (n = 2,564) using a simple random sampling method in a 7:3 ratio. Subsequently, we employed LASSO regression to identify variables significantly associated with DKD in T2DM patients. These variables were incorporated into eight distinct predictive models: Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), Gaussian Naive Bayes (GNB), KNeighbors Classifier (KNN), Gradient Boosting Classifier (GBM), AdaBoost Classifier (AdaBoost), and Extreme Gradient Boosting (XGBoost). The models' predictive performance was assessed using metrics such as the area under the curve (AUC), accuracy, F1 score, and Brier score. Finally, we developed an online calculator to estimate DKD risk in T2DM patients.
Results: Fifteen features-namely gender, age, systolic blood pressure (SBP), blood urea nitrogen (BUN), creatinine (Cr), BUN/Cr ratio, uric acid (UA), hemoglobin A1c (HbA1c), microalbuminuria, presence of diabetic retinopathy (DR), hypertension, coronary heart disease (CHD), history of cerebral infarction, family history of diabetes, and family history of CHD-associated with DKD were selected using LASSO regression. Among eight evaluated models, the XGBoost algorithm demonstrated superior performance on both training and validation datasets, with an AUCof 0.932 (95%CI: 0.926-0.938) and 0.930, (95%CI: 0.920-0.939), respectively. The model achieved an accuracy of 0.845 and 0.844, sensitivity of 0.834 and 0.850, specificity of 0.857 and 0.837, F1 score of 0.847 and 0.848, and a Brier score of 0.167 and 0.166, respectively. Decision curve analysis (DCA) further validated the superiority of the XGBoost model over other models across a range of clinically relevant risk thresholds, yielding the highest net benefits. Finally, an online predictive calculator for the occurrence of DKD was developed based on the XGBoost model, utilizing a cut-off value of 50.7%.
Conclusions: The developed XGBoost model demonstrated optimal predictive accuracy for the occurrence of DKD in patients with T2DM. This model facilitated the construction of an online prediction calculator, offering an accessible and practical tool for both patients and clinicians.

Keywords:  diabetic kidney disease; machine learning; prediction model; predictive value; type 2 diabetes mellitus

DOI:  https://doi.org/10.3389/fendo.2025.1587932
Healthcare (Basel). 2025 Oct 14. pii: 2588. [Epub ahead of print]13(20):

Interpretable Machine Learning Framework for Diabetes Prediction: Integrating SMOTE Balancing with SHAP Explainability for Clinical Decision Support.

Pathamakorn Netayawijit, Wirapong Chansanam, Kanda Sorn-In.

  Background: Class imbalance and limited interpretability remain major barriers to the clinical adoption of machine learning in diabetes prediction. These challenges often result in poor sensitivity to high-risk cases and reduced trust in AI-based decision support. This study addresses these limitations by integrating SMOTE-based resampling with SHAP-driven explainability, aiming to enhance both predictive performance and clinical transparency for real-world deployment. Objective: To develop and validate an interpretable machine learning framework that addresses class imbalance through advanced resampling techniques while providing clinically meaningful explanations for enhanced decision support. This study serves as a methodologically rigorous proof-of-concept, prioritizing analytical integrity over scale. While based on a computationally feasible subset of 1500 records, future work will extend to the full 100,000-patient dataset to evaluate scalability and external validity. We used the publicly available, de-identified Diabetes Prediction Dataset hosted on Kaggle, which is synthetic/derivative and not a clinically curated cohort. Accordingly, this study is framed as a methodological proof-of-concept rather than a clinically generalizable evaluation. Methods: We implemented a robust seven-stage pipeline integrating the Synthetic Minority Oversampling Technique (SMOTE) with SHapley Additive exPlanations (SHAP) to enhance model interpretability and address class imbalance. Five machine learning algorithms-Random Forest, Gradient Boosting, Support Vector Machine (SVM), Logistic Regression, and XGBoost-were comparatively evaluated on a stratified random sample of 1500 patient records drawn from the publicly available Diabetes Prediction Dataset (n = 100,000) hosted on Kaggle. To ensure methodological rigor and prevent data leakage, all preprocessing steps-including SMOTE application-were performed within the training folds of a 5-fold stratified cross-validation framework, preserving the original class distribution in each fold. Model performance was assessed using accuracy, area under the receiver operating characteristic curve (AUC), sensitivity, specificity, F1-score, and precision. Statistical significance was determined using McNemar's test, with p-values adjusted via the Bonferroni correction to control for multiple comparisons. Results: The Random Forest-SMOTE model achieved superior performance with 96.91% accuracy (95% CI: 95.4-98.2%), AUC of 0.998, sensitivity of 99.5%, and specificity of 97.3%, significantly outperforming recent benchmarks (p < 0.001). SHAP analysis identified glucose (SHAP value: 2.34) and BMI (SHAP value: 1.87) as primary predictors, demonstrating strong clinical concordance. Feature interaction analysis revealed synergistic effects between glucose and BMI, providing actionable insights for personalized intervention strategies. Conclusions: Despite promising results, further validation of the proposed framework is required prior to any clinical deployment. At this stage, the study should be regarded as a methodological proof-of-concept rather than a clinically generalizable evaluation. Our framework successfully bridges algorithmic performance and clinical applicability. It achieved high cross-validated performance on a publicly available Kaggle dataset, with Random Forest reaching 96.9% accuracy and 0.998 AUC. These results are dataset-specific and should not be interpreted as clinical performance. External, prospective validation in real-world cohorts is required prior to any consideration of clinical deployment, particularly for personalized risk assessment in healthcare systems.

Keywords:  Random Forest; SHAP; SMOTE; clinical decision support; diabetes prediction; explainable AI; healthcare AI; machine learning; model interpretability

DOI:  https://doi.org/10.3390/healthcare13202588
Ophthalmol Sci. 2026 Jan;6(1): 100934

In-Context Learning for Data-Efficient Diabetic Retinopathy Detection via Multimodal Foundation Models.

Murat S Ayhan, Ariel Y Ong, Eden Ruffell, Siegfried K Wagner, David A Merle, Pearse A Keane.

   Objective: This study aims to evaluate whether in-context learning (ICL), a prompt-based learning mechanism enabling multimodal foundation models to rapidly adapt to new tasks without retraining or large annotated datasets, can achieve comparable diagnostic performance to domain-specific foundation models. Specifically, we use diabetic retinopathy (DR) detection as an exemplar task to probe if a multimodal foundation model (Google Gemini 1.5 Pro), employing ICL, can match the performance of a domain-specific model (RETFound) fine-tuned explicitly for DR detection from color fundus photographs (CFPs).
Design: A cross-sectional study.
Subjects: A retrospective, publicly available dataset (Indian Diabetic Retinopathy Image Dataset) comprising 516 CFPs collected at an eye clinic in India, featuring both healthy individuals and patients with DR.
Methods: The images were dichotomized into 2 groups based on the presence or absence of any signs of DR. RETFound was fine-tuned for this binary classification task, while Gemini 1.5 Pro was assessed for it under zero-shot and few-shot prompting scenarios, with the latter incorporating random or k-nearest-neighbors-based sampling of a varying number of example images. For experiments, data were partitioned into training, validation, and test sets in a stratified manner, with the process repeated for 10-fold cross-validation.
Main Outcome Measures: Performance was assessed via accuracy, F1 score, and expected calibration error of predictive probabilities. Statistical significance was evaluated using Wilcoxon tests.
Results: The best ICL performance with Gemini 1.5 Pro yielded an average accuracy of 0.841 (95% confidence interval [CI]: 0.803-0.879), an F1 score of 0.876 (95% CI: 0.844-0.909), and a calibration error of 0.129 (95% CI: 0.107-0.152). RETFound achieved an average accuracy of 0.849 (95% CI: 0.813-0.885), an F1 score of 0.883 (95% CI: 0.852-0.915), and a calibration error of 0.081 (95% CI: 0.066-0.097). While accuracy and F1 scores were comparable (P > 0.3), RETFound's calibration was superior (P = 0.004).
Conclusions: Gemini 1.5 Pro with ICL demonstrated performance comparable to RETFound for binary DR detection, illustrating how future medical artificial intelligence systems may build upon such frontier models rather than being bespoke solutions.
Financial Disclosures: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Keywords:  AI; Artificial intelligence; Diabetic retinopathy; LLM; Large language model

DOI:  https://doi.org/10.1016/j.xops.2025.100934
Sci Rep. 2025 Oct 30. 15(1): 37954

DRCNN-Lesion Proxy: a hybrid CNN with lesion-inspired feature simulation for diabetic retinopathy severity classification.

Priyadharshini Sekar, Kanaga S Suba Raja, Ramaswamy Krishnaraj.

  Diabetic Retinopathy (DR) remains a leading cause of vision loss globally, necessitating accurate and scalable diagnostic solutions. Existing Deep Learning (DL) models often underutilize lesion-specific cues that are critical for early DR grading, while detection based models require costly lesion annotations. To address these limitations, we propose DRCNN-Lesion Proxy, a hybrid architecture that integrates a ResNet34 based CNN backbone for extracting global image level features with a Lesion Proxy Module, which simulates lesion-inspired cues without explicit lesion bounding box annotations. These heterogeneous features are fused through a late fusion classification head to enable robust multiclass DR severity prediction. The model was trained on a composite dataset and rigorously evaluated across six publicly available benchmarks namely EyePACS, Messidor-2, APTOS 2019, DDR, DIARETDB1, and IDRiD. Experimental findings show that the proposed framework consistently outperforms baseline CNNs and recent hybrid methods, achieving up to 98.37% accuracy, 97.28% F1-score, and 98.14% AUC. Statistical significance testing confirmed that these improvements were not due to chance. Furthermore, Grad-CAM visualizations highlighted clinically relevant retinal regions, and a pilot validation with three ophthalmologists on 20 cases reported mean scores above 3.5 out of 5, confirming that the explanations were perceived as clinically meaningful and useful for grading. The proposed framework provides an annotation light solution with strong generalizability, diagnostic precision, and clinically validated interpretability, advancing the state of the art in automated DR screening and offering a practical pathway for real world deployment.

Keywords:  Automated screening; Clinical validation; Diabetic retinopathy classification; Explainable artificial intelligence; Hybrid deep learning; Lesion proxy module

DOI:  https://doi.org/10.1038/s41598-025-21337-3
JAMA. 2025 Oct 27.

Artificial Intelligence and Diabetes Prevention.

Leigh Perreault.

DOI: https://doi.org/10.1001/jama.2025.20099
JAMIA Open. 2025 Oct;8(5): ooaf133

Deploying machine learning models in clinical settings: a real-world feasibility analysis for a model identifying adult-onset type 1 diabetes initially classified as type 2.

Irene Brusini, Suyin Lee, Jacob Hollingsworth, Amanda Sees, Matthew Hackenberg, Harm Scherpbier, Raquel López-Díez, Nadejda Leavitt.

   Objective: This study evaluates the performance and deployment feasibility of a machine learning (ML) model to identify adult-onset type 1 diabetes (T1D) initially coded as type 2 on electronic medical records (EMRs) from a health information exchange (HIE). To our knowledge, this is the first evaluation of such a model on real-world HIE data.
Materials and Methods: An existing ML model, trained on national US EMR data, was tested on a regional HIE dataset, after several adjustments for compatibility. A localized model retrained on the regional dataset was compared to the national model. Discrepancies between the 2 datasets' features and cohorts were also investigated.
Results: The national model performed well on HIE data (AUROC = 0.751; precision at 5% recall [PR5] = 25.5%), and localization further improved performance (AUROC = 0.774; PR5 = 35.4%). Differences in the 2 models' top predictors reflected the discrepancies between the datasets and gaps in HIE data capture.
Discussion: The adjustments needed for testing on HIE data highlight the importance of aligning algorithm design with deployment needs. Moreover, localization increased precision, making it more appealing for patient screening, but added complexity and may impact scalability. Additionally, while HIEs offer opportunities for large-scale deployment, data inconsistencies across member organizations could undermine accuracy and providers' trust in ML-based tools.
Conclusion: Our findings offer valuable insights into the feasibility of at-scale deployment of ML models for high-risk patient identification. Although this work focuses on detecting potentially misclassified T1D, our learnings can also inform other applications.

Keywords:  Health Information Exchange; clinical decision support systems; electronic health records; machine learning; type 1 diabetes mellitus

DOI:  https://doi.org/10.1093/jamiaopen/ooaf133
Comput Med Imaging Graph. 2025 Oct 21. pii: S0895-6111(25)00164-8. [Epub ahead of print]126 102655

A CNN-Transformer fusion network for Diabetic retinopathy image classification.

Xuan Huang, Zhuang Ai, Chongyang She, Qi Li, Qihao Wei, Sha Xu, Yaping Lu, Fanxin Zeng.

  Diabetic retinopathy (DR) is a leading cause of blindness worldwide, yet current diagnosis relies on labor-intensive and subjective fundus image interpretation. Here we present a convolutional neural network-transformer fusion model (DR-CTFN) that integrates ConvNeXt and Swin Transformer algorithms with a lightweight attention block (LAB) to enhance feature extraction. To address dataset imbalance, we applied standardized preprocessing and extensive image augmentation. On the Kaggle EyePACS dataset, DR-CTFN outperformed ConvNeXt and Swin Transformer in accuracy by 3.14% and 8.39%, while also achieving a superior area under the curve (AUC) by 1% and 26.08%. External validation on APTOS 2019 Blindness Detection and a clinical DR dataset yielded accuracies of 84.45% and 85.31%, with AUC values of 95.22% and 95.79%, respectively. These results demonstrate that DR-CTFN enables rapid, robust, and precise DR detection, offering a scalable approach for early diagnosis and prevention of vision loss, thereby enhancing the quality of life for DR patients.

Keywords:  Attention mechanism; Diabetic retinopathy; Feature fusion; Image classification; Transformer

DOI:  https://doi.org/10.1016/j.compmedimag.2025.102655
Sheng Wu Yi Xue Gong Cheng Xue Za Zhi. 2025 Oct 25. 42(5): 892-900

[An adaptive multi-label classification model for diabetic retinopathy lesion recognition].

Xina Liu, Jun Xie, Junjun Hou, Xinying Xu, Yan Guo.

  Diabetic retinopathy is a common blinding complication in diabetic patients. Compared with conventional fundus color photography, fundus fluorescein angiography can dynamically display retinal vessel permeability changes, offering unique advantages in detecting early small lesions such as microaneurysms. However, existing intelligent diagnostic research on diabetic retinopathy images primarily focuses on fundus color photography, with relatively insufficient research on complex lesion recognition in fluorescein angiography images. This study proposed an adaptive multi-label classification model (D-LAM) to improve the recognition accuracy of small lesions by constructing a category-adaptive mapping module, a label-specific decoding module, and an innovative loss function. Experimental results on a self-built dataset demonstrated that the model achieved a mean average precision of 96.27%, a category F1-score of 91.21%, and an overall F1-score of 94.58%, with particularly outstanding performance in recognizing small lesions such as microaneurysms (AP = 1.00), significantly outperforming existing methods. The research provides reliable technical support for clinical diagnosis of diabetic retinopathy based on fluorescein angiography.

Keywords:  Category-adaptive mapping; Diabetic retinopathy; Fundus fluorescein angiography; Label-specific decoding; Lesion recognition; Multi-label classification; Smooth focal loss

DOI:  https://doi.org/10.7507/1001-5515.202503056
Turk J Ophthalmol. 2025 Oct 27. 55(5): 260-275

Diabetic Retinopathy Screening Approaches in Developing Countries: A Systematic Review and Meta-Analysis.

Yudistira Yudistira, Kevin Anggakusuma Hendrawan, Ari Andayani, Ni Made Ari Suryathi, Titiek Ernawati, Alyssa Claudia Valerie Gunawan, Ni Putu Kostarika Melia Daradila.

   Objectives: Diabetic retinopathy (DR) is one of the primary causes of vision loss among people living with diabetes and is expected to rise globally in the coming years. Effective screening strategies are essential, particularly in developing countries where resources and access to specialized care are limited. Our objective was to assess how accurately different screening methods detect DR, specifically artificial intelligence (AI)-based tools, portable fundus cameras, and trained non-ophthalmologist personnel, implemented in a developing country.
Materials and Methods: A literature search was conducted in ScienceDirect, PubMed, and the Cochrane Library. Study quality was assessed using the Quality Assessment of Diagnostic Accuracy Studies 2 tool. While all included studies were reviewed qualitatively, only those evaluating AI-based screening tools were included in the meta-analysis. Meta-analysis was performed using MetaDisc 2.0 to calculate pooled sensitivity, specificity, diagnostic odds ratio, and likelihood ratios for any DR, referable DR, and vision-threatening DR.
Results: A total of 25 studies were included, with 21 AI-based studies eligible for the meta-analysis. The pooled sensitivity and specificity respectively were 0.890 (95% confidence interval [CI]: 0.845-0.924) and 0.900 (95% CI: 0.832-0.942) for any DR, 0.933 (95% CI: 0.890-0.960) and 0.903 (95% CI: 0.871-0.928) for referable DR, and 0.891 (95% CI: 0.393-0.990) and 0.936 (95% CI: 0.837-0.977) for vision-threatening DR. Meta-regression identified camera type as a significant factor. Portable fundus cameras and general physicians showed good agreement with the gold standards.
Conclusion: These findings support the potential of AI-assisted DR screening in low-resource settings and highlight the complementary roles of portable imaging and task-shifting to trained non-specialists.

Keywords:  Diabetic retinopathy screening; artificial intelligence; developing countries; non-specialist; portable fundus camera

DOI:  https://doi.org/10.4274/tjo.galenos.2025.32916
Diagnostics (Basel). 2025 Oct 17. pii: 2619. [Epub ahead of print]15(20):

Pixel-Attention W-Shaped Network for Joint Lesion Segmentation and Diabetic Retinopathy Severity Staging.

Archana Singh, Sushma Jain, Vinay Arora.

  Background: Visual impairment remains a critical public health challenge, and diabetic retinopathy (DR) is a leading cause of preventable blindness worldwide. Early stages of the disease are particularly difficult to identify, as lesions are subtle, expert review is time-consuming, and conventional diagnostic workflows remain subjective. Methods: To address these challenges, we propose a novel Pixel-Attention W-shaped (PAW-Net) deep learning framework that integrates a Lesion-Prior Cross Attention (LPCA) module with a W-shaped encoder-decoder architecture. The LPCA module enhances pixel-level representation of microaneurysms, hemorrhages, and exudates, while the dual-branch W-shaped design jointly performs lesion segmentation and disease severity grading in a single, clinically interpretable pass. The framework has been trained and validated using DDR and a preprocessed Messidor + EyePACS dataset, with APTOS-2019 reserved for external, out-of-distribution evaluation. Results: The proposed PAW-Net framework achieved robust performance across severity levels, with an accuracy of 98.65%, precision of 98.42%, recall (sensitivity) of 98.83%, specificity of 99.12%, F1-score of 98.61%, and a Dice coefficient of 98.61%. Comparative analyses demonstrate consistent improvements over contemporary architectures, particularly in accuracy and F1-score. Conclusions: The PAW-Net framework generates interpretable lesion overlays that facilitate rapid triage and follow-up, exhibits resilience under domain shift, and maintains an efficient computational footprint suitable for telemedicine and mobile deployment.

Keywords:  W-shaped network; diabetic retinopathy; lesion segmentation; pixel-attention

DOI:  https://doi.org/10.3390/diagnostics15202619
Digit Health. 2025 Jan-Dec;11:11 20552076251390574

Dynamics of the atherogenic index of plasma and diabetes conversion: Insights from machine learning and longitudinal pattern analysis in middle-aged and older Chinese adults with prediabetes.

Zhan Chen, Yue Zhang, Liming Wang.

   Background: Atherogenic dyslipidemia significantly influences diabetes development, yet the longitudinal prognostic value of the atherogenic index of plasma (AIP) in prediabetic populations remains understudied. This study investigated whether dynamic AIP measures predict diabetes conversion in middle-aged and older Chinese adults with prediabetes while exploring interactions with other metabolic risk factors through interpretable machine learning.
Methods: We analyzed data from 1965 prediabetic adults (≥ 45 years) from the China Health and Retirement Longitudinal Study (2012-2015), assessing baseline AIP, cumulative AIP exposure (AIPcum), and change patterns via K-means clustering. Associations with incident diabetes were evaluated using logistic regression with progressive confounder adjustment. Five machine learning models were developed using predictors identified through the Boruta algorithm and recursive feature elimination, with the best model interpreted via SHAP (SHapley Additive exPlanations) analysis.
Results: During follow-up, 15.0% of participants progressed to diabetes. Elevated AIPcum showed the strongest association with diabetes risk (OR for highest vs. lowest quartile = 2.37; 95% CI: 1.60-3.57; P < 0.001), while "Persistently High" and "Increasing" AIP change patterns exhibited ∼2.5-fold higher risk versus "Persistently Low" patterns. Random forest achieved the best predictive performance (AUROC = 0.760), with glucose, hand grip strength, AIPcum, and waist circumference as key predictors. Interaction analyses revealed significant synergistic effects between AIPcum and other metabolic factors.
Conclusion: Cumulative AIP exposure and unfavorable AIP change patterns independently predict prediabetes-to-diabetes progression. Incorporating these measures into risk assessment may enhance early identification of high-risk individuals and inform targeted interventions.

Keywords:  Atherogenic index of plasma; SHAP analysis; cumulative exposure; diabetes; machine learning; prediabetes

DOI:  https://doi.org/10.1177/20552076251390574
BMC Public Health. 2025 Oct 31. 25(1): 3688

Leveraging XGBoost and explainable AI for accurate prediction of type 2 diabetes.

Zahra Rafie, Moslem Sedaghat Talab, Behrooz Ebrahim Zadeh Koor, Ali Garavand, Cirruse Salehnasab, Mustafa Ghaderzadeh.



Keywords:  CatBoost; Dena cohort; Explainable AI; Machine learning; SHAP; Type 2 diabetes; XGBoost

DOI:  https://doi.org/10.1186/s12889-025-24953-w
J Diabetes Res. 2025 ;2025 7304414

Predicting Mild Cognitive Impairment in Type 2 Diabetes: A Machine Learning Approach.

Fangyi Li, Shengyi Zhao, Tianyu Wu, Sijue Yang, Yanjie Duan, Jing Sun, Wenhui Zhu, Beibei Zhai, Congcong Yu, Shihua Chen, Zhou Zhang, Wei Tang, Yan Bi.

   Background: Diabetes significantly increases the risk of cognitive impairment, particularly mild cognitive impairment (MCI). Early identification of individuals at risk for MCI is crucial for timely intervention. This study was aimed at developing and validating a machine learning-based model to predict MCI in patients with Type 2 diabetes (T2DM).
Methods: Participants with T2DM and completed cognitive assessments were included. Feature selection was done using statistical methods and genetic programming to reduce collinearity. Six classification models were trained and evaluated using cross-validation and hyperparameter tuning. External validation was performed with cohorts from the Jiangsu DiabEtes COgnitive Dysfunction Early Diagnosis and Intervention (DECODE) study and the Third National Health and Nutrition Examination Survey (NHANES III). SHAP analysis identified key predictors, and a web interface was developed for practical application.
Results: A total of 2074 participants were included. Significant predictors were education, age, GCA index (glycolipid metabolism), systolic blood pressure, eGFR, BMI, and diabetes duration. The support vector classifier (SVC) model achieved the highest performance, with an AUC of 0.74 ± 0.04, an F1 score of 0.62 ± 0.06, and a recall of 0.74 ± 0.09 in internal validation. External validation with the DECODE cohort yielded an AUC of 0.80, an F1 score of 0.80, and a recall of 0.89. NHANES III validation confirmed the model's reliability in predicting MCI risk.
Conclusions: This study compared machine learning models for diagnosing MCI in T2DM patients. The SVC model demonstrated strong efficacy and accuracy, highlighting the potential of machine learning in diagnosing MCI in this population.

Keywords:  Type 2 diabetes mellitus; machine learning; mild cognitive impairment; prediction model

DOI:  https://doi.org/10.1155/jdr/7304414
J Biomed Inform. 2025 Oct 23. pii: S1532-0464(25)00167-4. [Epub ahead of print] 104938

Enhancing AI-based diabetic retinopathy screening in low- and middle-income countries with synthetic data.

Zitao Shuai, Chenwei Wu, Zhengxu Tang, David Restrepo, Michael Morley, Luis Filipe Nakayama.

   OBJECTIVE: AI-based DR screening is promising in low- and middle-income countries (LMICs), where limited human resources constrain access to specialist-led programs. However, current systems often degrade under real-world image-quality variations, especially with portable devices that are vital for low- and middle-income countries. This study aims to develop Retsyn, a synthetic-data augmentation framework that improves screening robustness across devices and imaging conditions.
METHODS: RetSyn leverages advanced diffusion models to generate synthetic retinal images with diverse device and imaging quality characteristics. To address the challenges of (1) portable device data scarcity, (2) disease and quality distribution imbalance, and (3) varying image quality, RetSyn uses class and quality-conditioned diffusion for controllable synthesis, a group-balanced loss to increase coverage of minority (quality, disease) pairs, and a Direct Preference Optimization alignment step with a small paired smartphone-tabletop set. The synthesized images are then used to augment classifier training.
RESULTS: The effectiveness of RetSyn-generated images was evaluated by training retinal diagnosis models on a combination of real and synthetic data. RetSyn yields consistent gains in-domain and out-of-domain. On low-quality tabletop images, F1 improves from 0.781 to 0.874 (binary) and 0.607 to 0.703 (three-class), while AUROC reaches 0.982 and 0.951, respectively. On out-of-domain portable images, RetSyn attains AUROC 0.813/F1 0.703 (binary) and AUROC 0.804/F1 0.609 (three-class), exceeding group-robustness baselines such as GroupDRO (binary: AUROC 0.786/F1 0.626; three-class: AUROC 0.789/F1 0.544).
CONCLUSION: RetSyn presents an effective and scalable synthetic data framework that significantly enhances the robustness and generalizability of AI-based DR screening models in LMICs. By addressing the critical challenges posed by varying image quality and device characteristics, RetSyn facilitates more reliable deployment of AI diagnostics in underserved regions. Additionally, the release of the first publicly available paired smartphone-tabletop retinal image dataset will support further research into cross-device DR screening solutions.

Keywords:  Cross-device generalization; Diabetic retinopathy; Diffusion models; Low-resource settings; Synthetic data

DOI:  https://doi.org/10.1016/j.jbi.2025.104938
Comput Methods Programs Biomed. 2025 Oct 18. pii: S0169-2607(25)00538-3. [Epub ahead of print]273 109122

Refine XGBoost with SHAP explainability for non-invasive early detection of diabetic kidney disease: Estimated cardiac output as a potential indicator.

Ngoc Minh To, Quoc Cuong Ngo, Barbara Polus, Minh Ngoc Dinh, Ahsan Khandoker, Anish Menon, Dinesh Kumar.

   INTRODUCTION: Moderately increased (micro) albuminuria serves as a critical early indicator of Diabetic Kidney Disease (DKD). However, traditional screening methods that rely on laboratory-based analyses face significant challenges in enabling timely and continuous monitoring. This study addresses these limitations by introducing a non-invasive approach for albuminuria risk detection, allowing real-time estimation of mild albuminuria increases using vital signs and body measurements.
METHODS: We developed a non-invasive model for albuminuria risk detection using vital signs and body measurements. Data were drawn from the NHANES cohort (USA) and a Bangladeshi cohort of people with diabetes (PwD). Feature selection identified four non-laboratory predictors - estimated cardiac output (eCO), body mass index, waist circumference, and diabetes duration - as the most informative inputs. The proposed models were benchmarked against baseline machine learning approaches and existing methods developed over the past decade, with model interpretability assessed via SHapely Additive exPlanation (SHAP) contributions.
RESULTS: Our best model, an XGBoost classifier, achieved an AUC of 0.75 [0.67-0.84], an accuracy of 0.70, and a macro F1 score of 0.68, outperforming other non-invasive risk scores (0.58) and machine learning baselines. Validation against an external reference risk score confirmed superior precision-recall balance for both positive (microalbuminuria) and negative classes.
CONCLUSION: This study demonstrates that a fine-tuned, non-invasive XGBoost model using simple clinical measures can support albuminuria monitoring and early DKD detection without laboratory tests. While the selected predictors may not represent the definitive or optimal feature set, their strong performance highlights the potential of leveraging easily obtainable, clinically relevant measures. In particular, the contribution of eCO underscores a promising direction for exploring heart-kidney-metabolism interactions in DKD risk assessment. Together, these findings highlight a scalable, non-invasive tool for resource-limited settings, an interpretable framework for clinical trust, and a pathway to refining feature sets for both accuracy and biological insight.

Keywords:  Albuminuria; DKD; Diabetic kidney disease; Estimated cardiac output (eCO); Machine learning; Non-invasive medical screening

DOI:  https://doi.org/10.1016/j.cmpb.2025.109122
Nanoscale Adv. 2025 Oct 28.

Clinical diagnosis of diabetes using machine learning and surface-enhanced Raman spectroscopy liquid biopsy: an exploratory study.

Allah Ditta, Peiying Wu, Rui Zhang, Haq Nawaz, Muhammad Irfan Majeed, Sima Rezvantalab, Sara Mihandoost, Eva Miriam Buhl, Stephan Rütten, Fabian Kiessling, Twan Lammers, Roger M Pallares.

The impact of diabetes on global health is increasing, underscoring the need for early and accurate diagnosis to prevent severe complications. Nevertheless, conventional diagnostic approaches, such as glycated hemoglobin testing and oral glucose tolerance tests, often lack sensitivity or specificity, particularly for detecting the disease at an early stage. In this exploratory clinical study, we present a promising alternative, label-free surface-enhanced Raman spectroscopy (SERS), which enables rapid, non-invasive biochemical analysis of liquid samples. Using gold nanoparticles as substrates, we applied label-free SERS to clinical serum samples for diabetes diagnosis. Because label-free SERS analysis of biological samples yields complex spectra, we developed a machine learning workflow tailored to clinical samples, exploring four different machine learning models in combination with synthetic data augmentation. This approach achieved classification accuracies of 96% and 94% for the healthy and diabetes groups, respectively. Our results demonstrate the benefits of integrating label-free SERS and machine learning models for efficient, accurate diabetes diagnosis via liquid biopsy, offering a powerful tool to enhance detection and potentially improve patient outcomes worldwide.

DOI: https://doi.org/10.1039/d5na00905g
BMC Ophthalmol. 2025 Oct 28. 25(1): 602

Automated FAZ segmentation and diabetic retinopathy classification using OCTA images.

Jamshid Saeidian, Hamid Riazi-Esfahani, Hossein Azimi, Hossein Farrokhpour, Ali Momeni, Mahdi Jamalitootakani, Ahmad Mirshahi, Hooshang Faghihi, Reza Sadeghi, Elias Khalili Pour.

   BACKGROUND: Accurate segmentation of the foveal avascular zone (FAZ) is valuable for retinal imaging, as FAZ alterations are key biomarkers for diabetic retinopathy (DR). This study presents an automated framework exploring the feasibility of FAZ segmentation and DR classification using optical coherence tomography angiography (OCTA) images.
METHODS: In this cross-sectional study conducted at Farabi Eye Hospital, Tehran, Iran, a two-step deep learning pipeline was developed. First, a neural network combining DeepLabv3+, EfficientNetB0, Squeeze-and-Excitation (SE) blocks, and Atrous Spatial Pyramid Pooling (ASPP) was trained to segment the FAZ from superficial capillary plexus (SCP) and deep capillary plexus (DCP) OCTA slabs. Second, a GoogLeNet-based convolutional neural network (CNN) classified segmented FAZ images into binary (normal vs. DR) and three-class (normal, non-proliferative DR [NPDR], proliferative DR [PDR]) categories to differentiate DR stages based on FAZ shape characteristics. For the classification task using the deep learning-generated segmented FAZ images as input, the data was split into 70% training, 10% validation, and 20% testing, with 5-fold cross-validation to mitigate overfitting. Data augmentation and Synthetic Minority Oversampling Technique (SMOTE) were applied to improve classification performance.
RESULTS: The final dataset comprised 253 OCTA scans (126 SCP, 127 DCP) from 161 eyes of 161 participants (one eye per participant), with 39 normal participants (24.2%), 78 patients with NPDR (48.4%), and 44 with PDR (27.3%). The mean age was 49.7 ± 11.8 years, and 54% were male. The FAZ segmentation network achieved a Dice similarity coefficient (DSC) of 97.5% across the dataset, achieving high precision even in challenging images. The classification model, using the deep learning generated segmented FAZ images as input, reached an area under the curve (AUC) of 100% for binary classification (normal vs. DR) and 87% for three-class classification (normal, NPDR, PDR) with oversampling.
CONCLUSION: This system, with its potential for integrating into clinical workflows, offers a promising assistive tool for clinicians, which could enable earlier and more accurate diagnosis of diabetic retinopathy from OCTA images.
CLINICAL TRIAL NUMBER: Not applicable.

Keywords:  Classification; Deep learning; Diabetic retinopathy; Foveal avascular zone; Image segmentation; OCTA

DOI:  https://doi.org/10.1186/s12886-025-04473-2
Front Endocrinol (Lausanne). 2025 ;16 1675152

Triglyceride-glucose index and mortality in congestive heart failure with diabetes: a machine learning predictive model.

Lin Yu, Haizhu Chen, Jiwen Zhang, Wei Han.

   Background: The triglyceride-glucose (TyG) index serves as a marker for insulin resistance. Research exploring the link between the TyG index and adverse outcomes among patients suffering from congestive heart failure (CHF) and diabetes mellitus (DM) is limited. This investigation endeavors to uncover the connection between the TyG index and mortality risk in subjects suffering from CHF and DM.
Methods: We obtained clinical data for patients with CHF and DM from the MIMIC-IV (3.1) database. The optimal cutoff value for the TyG index was determined using X-tile software, and patients were classified into three groups. The primary outcome was 28-day hospital mortality, and the secondary outcome was 28-day ICU mortality. We used restricted cubic splines (RCS), COX regression analysis, and Kaplan-Meier survival curves to examine the association between the TyG index and adverse outcomes. Subgroup analyses were conducted based on age, gender, chronic pulmonary disease, atrial fibrillation, hypertension, and mechanical ventilation to assess the robustness of our findings. Feature selection was performed using LASSO regression, and predictive modeling was carried out using machine learning algorithms.
Results: This study included 1046 patients with CHF and DM. Using a fully adjusted COX regression model, we identified a significant association between the TyG index and both 28-day hospital mortality (HR=1.31, 95% CI: 1.09-1.57, P=0.004) and 28-day ICU mortality (HR=1.29, 95% CI: 1.07-1.54, P=0.006). Using restricted cubic spline analysis, a linear link between the TyG index and mortality rates was found, indicating that a rise in TyG correlates with a heightened risk of unfavorable outcomes. The predictive performance was evaluated using seven machine learning algorithms, with the Random Survival Forest (RSF) algorithm achieving the best performance (AUC=0.817).
Conclusions: In patients with CHF and DM, TyG exhibited a linear correlation with both 28-day hospital mortality and 28-day ICU mortality. Elevated TyG values were significantly linked to a heightened risk of adverse events.

Keywords:  congestive heart failure; diabetes; machine learning; mortality; triglyceride glucose index

DOI:  https://doi.org/10.3389/fendo.2025.1675152
Ophthalmol Sci. 2026 Jan;6(1): 100935

Autonomous Artificial Intelligence in Diabetic Retinopathy Testing-Lessons Learned on Successful Health System Adoption.

Clare W Teng, Saawan D Patel, Andrew J Barkmeier, T Y Alvin Liu, David Myung, Jeffrey Henderer, James Liu, Eric Hansen, Lama A Al-Aswad.

   Purpose: Artificial intelligence (AI)-aided diabetic retinopathy (DR) testing systems have been commercialized for 5 years, but adoption is still relatively limited. This article aims to summarize the evidence in clinical settings, describe the current state of adoption, and share themes of successful implementation.
Design: Evaluation of diagnostic test or technology.
Participants: Ophthalmologists.
Methods: We performed literature review and conducted interviews with ophthalmologists leading implementation of AI-aided DR testing programs at several academic health systems. The study focused on the 3 currently US Food and Drug Administration-cleared AI systems: LumineticsCore, EyeArt, and AEYE Diagnostic Screening (AEYE-DS), assessing their performance and strategies utilized by health systems to effectively implement this technology in clinics.
Main Outcome Measures: Diagnostic accuracy data, ophthalmologist feedback.
Results: The literature review found 6 publications reporting diagnostic accuracy data of autonomous AI DR testing in primary care office settings, including 5 for LumineticsCore and 1 for EyeArt. Additional articles, of which 18 were selected for detailed review, addressed impact on patient adherence, health equity, and carbon footprint, as well as cost-effectiveness and workflow efficiency analyses. There were no studies comparing the systems on the same patients. In aggregate, adopters of the AI systems reported average nonmydriatic gradability of 49% to 75% (n = 5), sensitivity 87% to 100% (n = 3), and specificity 60% to 91% (n = 4). Based on public records at the time of writing, both LumineticsCore and EyeArt have >5 academic adopters in the United States. Limited information is available on AEYE-DS given recency of regulatory clearance. Elements of successful implementation include proper site selection, aligning AI tools with primary care clinic workflows, streamlining patient engagement and referrals, and ongoing training of staff. Health systems utilizing AI reported improved Healthcare Effectiveness Data and Information Set measures, health equity, productivity, and patient adherence to follow-up with ophthalmology.
Conclusions: Artificial intelligence-aided diabetic eye examinations present a promising solution to facilitate early detection of DR, promote equitable access, and drive down system-level cost of care. Its successful implementation requires addressing technological, operational, and stakeholder engagement challenges. Our study underscores the potential of AI to revolutionize care delivery provided its adoption is strategically managed.
Financial Disclosures: The author has no/the authors have no proprietary or commercial interest in any materials discussed in this article.

Keywords:  Autonomous artificial intelligence; Diabetic retinopathy testing; Health system adoption; Success factors; Value propositions

DOI:  https://doi.org/10.1016/j.xops.2025.100935
Clin Exp Med. 2025 Oct 31. 25(1): 342

Reliable biomarkers for diabetic nephropathy using machine learning-assisted contrast-enhanced ultrasonography and clinical characteristics.

Lin Lin, Lin Yan, Nan Li, Yiru Wang, Yukun Luo.

   OBJECTIVE: To utilize machine learning techniques to screen contrast-enhanced ultrasound (CEUS) parameters and clinical characteristics, aiming to differentiate diabetic nephropathy (DN) from non-diabetic renal disease (NDRD) in patients with diabetic kidney injury.
METHODS: Data from 120 diabetic patients (240 kidneys) with chronic kidney disease (CKD) were analyzed. The data included basic clinical features for each kidney and renal vascular data obtained through CEUS. Statistical analysis, tenfold cross-validation and random forest method were used for data processing. Receiver operating characteristic (ROC) curves were employed to depict the diagnostic performance of the indicators.
RESULTS: The random forest model integrating CEUS parameters and clinical characteristics achieved an average classification accuracy of 87.6% in differentiating kidney injury types. ROC curve analysis showed an AUC of 0.918.
CONCLUSION: Through machine learning, CEUS quantitative parameters and clinical features of the screened model can be used as important noninvasive biomarkers to identify kidney injury in T2DM patients with DN. Ai-assisted CEUS and specific clinical features are a fast and reliable tool for DN screening.

Keywords:  Biomarker; Clinical characteristics; Contrast-enhanced ultrasound; Diabetic nephropathy

DOI:  https://doi.org/10.1007/s10238-025-01837-2
Sheng Wu Yi Xue Gong Cheng Xue Za Zhi. 2025 Oct 25. 42(5): 1005-1011

[Research on type 2 diabetes prediction algorithm based on photoplethysmography].

Mingying Hu, Quanyu Wu, Yifan Cao, Jin Cao, Yifan Zhao, Lin Zhang, Xiaojie Liu.

  To address the current issues of data imbalance and scarcity in photoplethysmography (PPG) data for type 2 diabetes mellitus (T2DM) prediction, this study proposes an improved conditional Wasserstein generative adversarial network with gradient penalty (CWGAN-GP). The algorithm integrated gated recurrent unit (GRU) networks and self-attention mechanisms to construct a generator, aiming to produce high-quality PPG signals. Various data augmentation methods, including the improved CWGAN-GP, were employed to expand the PPG dataset, and multiple classifiers were applied for T2DM prediction analysis. Experimental results showed that the model trained on data generated by the improved CWGAN-GP achieved the optimal prediction performance. The highest accuracy reached 0.895 0, and compared with other data enhancement methods, this approach exhibited significant advantages in terms of precision and F1-score. The generated data notably enhances the accuracy and generalization ability of T2DM prediction models, providing a more reliable technical basis for non-invasive early T2DM screening based on PPG signals.

Keywords:  Data augmentation; Generative adversarial networks; Photoplethysmography; Self-attention; Type 2 diabetes

DOI:  https://doi.org/10.7507/1001-5515.202501006
J Neurol Sci. 2025 Oct 22. pii: S0022-510X(25)00352-1. [Epub ahead of print]479 123732

Response to Letter to the Editor: Commentary on "Multimodal MRI analysis selecting key brain features for machine learning based classification of diabetic neuropathic pain and phenotypes".

Søren Nf Hostrup, Jens B Frøkjær.

DOI: https://doi.org/10.1016/j.jns.2025.123732
Front Neurol. 2025 ;16 1599793

Automatic diagnosis of type 2 diabetes mellitus with mild cognitive impairment using artificial intelligence based on routine T1-weighted MRI.

Chang Li, Jun Zhang, Bo Xue, Yuwei Xia, Feng Shi, Xingyan Le, Junbang Feng, Peng Chen, Chuanming Li.

   Background: Patients with type 2 diabetes mellitus (T2DM) exhibit a heightened susceptibility to developing dementia, especially those who already present with mild cognitive impairment (MCI). Nevertheless, the fundamental etiology remains elusive, underscoring the pressing need for an objective and precise diagnostic approach in clinical settings. This study investigates the utilization of machine learning algorithms in conjunction with high-resolution sagittal T1-weighted structural imaging to facilitate automated diagnosis of T2DM patients with MCI, differentiating them from both T2DM patients without MCI and healthy controls (HCs).
Methods: Thirty patients with T2DM and MCI, thirty T2DM patients without MCI, and thirty matched healthy controls (HCs) were enrolled to identify independent biomarkers and develop a diagnostic model for early cognitive impairment in T2DM. Whole-brain structural features-including cortical surface area, volume, thickness, curvature index, folding index, Gaussian curvature, mean curvature, thickness standard deviation, nuclear volume, hippocampal volume, and white matter volume-were extracted from the images of brains using automated segmentation methods. The minimum redundancy maximum relevance (MRMR) method was employed to filter out irrelevant and redundant features, reducing the dimensionality of the dataset. Subsequently, the least absolute shrinkage and selection operator (LASSO) algorithm was applied for further feature selection, ensuring the retention of only the most diagnostic features. The Random Forest (RF) classifier, a powerful machine learning model within the realm of artificial intelligence, was meticulously trained utilizing a curated feature set that had undergone rigorous refinement. To ensure the robust diagnostic performance and generalizability of the established random forest model, a 5-fold cross-validation was employed, providing a dependable estimation of the model's effectiveness.
Results: The FreeSurfer software automatically segmented the cerebral imaging data into up to 70 regions. For model establishment, a comprehensive set of 705 structural features, a series of neuropsychological tests, and standard laboratory tests were utilized. Ultimately, 8 features were selected through two feature selection strategies aimed at refining the optimal features. These included bilateral brainstem volume, left hippocampus volume, left transverse temporal gyrus volume, bilateral posterior corpus callosum volume, left medial orbitofrontal cortex Gaussian curvature, glycosylated hemoglobin, blood sugar levels, and the Digit Span Test (DST) backward score. The Random Forest (RF) model, based on the combined features, exhibited the excellent performance, with mean AUCs of 0.959 (95% CI, 0.940-0.997, mean specificity = 94.2%, mean sensitivity = 88.3%, mean accuracy = 88.3% and mean precision = 88.3%) for the training dataset and mean AUCs of 0.887 (95% CI, 0.746-0.992, mean specificity = 85.0%, mean sensitivity = 70.0%, mean accuracy = 70.0% and mean precision = 69.6%) for the testing dataset, based on 5-fold cross-validation.
Conclusion: The RF model, leveraging a combination of features, demonstrates high accuracy in diagnosing T2DM with MCI. The exclusion of T2DM patients with complications may limit generalizability to the broader T2DM population, potentially inflating performance estimates. Among these features, 8 optimal indicators comprising 5 structural features, 1 neuropsychological test feature, and 2 standard laboratory test features emerge as the potential independent biomarkers for detecting early-stage cognitive impairment in T2DM patients. These features hold significant importance in understanding the pathophysiological mechanisms of T2DM-related cognitive impairment. Our fully automated model is capable of swiftly processing MRI data, enabling precise and objective differentiation of T2DM with MCI. This model significantly enhances diagnostic efficiency and holds considerable value in clinical practice, facilitating early diagnosis of T2DM with MCI.

Keywords:  MRI; artificial intelligence - AI; mild cognitive impairment - MCI; random forest model (RF); type 2 diabetes mellitus

DOI:  https://doi.org/10.3389/fneur.2025.1599793