bims-aukdir 2026-07-19 papers

bims-aukdir

Biomed News

on Automated knowledge discovery in diabetes research

Issue of 2026–07–19
twenty papers selected by
Mott Given

A hybrid CNN-GNN and multitask learning pipeline for improving mild diabetic retinopathy sensitivity.
Prediction of Factors Influencing the Incidence of Diabetic Foot Ulcers Using Classical Statistical and Machine Learning Approaches: A Systematic Review.
Dynamic evolution of readmission risk factors across short-, medium-, and long-term horizons in type 2 diabetes: a machine learning-based predictive modeling study with SHAP interpretability.
Machine Learning Integration of Clinical and Molecular Biomarkers to Predict Vascular Complications in Type 2 Diabetes.
Development and validation of a multimodal interpretable machine learning model for the identification of osteoporosis in patients with type 2 diabetes mellitus: a multicenter retrospective study.
Machine learning evaluation of clinical, social and behavioural factors influencing progression from pre-diabetes to type 2 diabetes: a retrospective cohort study in southeast Michigan.
Interpretable diabetes risk prediction: a comparative study of tree-based algorithms using SHAP and LIME.
Interpretable machine learning for predicting mild cognitive impairment in elderly patients with type 2 diabetes mellitus: model development and performance assessment.
Deep learning-derived retinal biomarker associated with diabetes-related amputation in type 2 diabetes.
MS-EGT-Net: a multi-scale enhanced graph-transformer network for diabetic foot ulcer classification.
Prism-refraction K-means polar lights optimizer-driven feature selection for accurate diabetes diagnosis.
Interpretable single-cell machine learning identifies PINK1-centred mitophagy as a determinant of metabolic fitness in type 2 diabetes.
Association of Variability of Monthly Continuous Glucose Monitoring-Derived Metrics With Diabetic Kidney Disease in Type 1 Diabetes.
Early Prediction of Diabetic Macular Edema via Machine Learning Survival Analysis on Checkup Data.
A Hybrid CNN-MLLM Architecture for Image-Based Nutrition Estimation and Advisory Insulin Decision Support in Type 1 Diabetes.
Quantitative pancreatic MRI in diabetes mellitus: current advances and future directions.
Narrative review of the development of an ischaemic heart disease prognostic scoring tool (i-IHD score) among patients with type 2 diabetes mellitus in Malaysia.
Assessing Quality, Readability, and Transparency of Online and AI-Generated Information on Type 2 Diabetes Mellitus: Cross-Sectional Exploratory Study.
Plasma proteomics based on machine learning predict the early risk of kidney outcomes in patients with DKD: a prospective cohort study from UK Biobank.
Inflammation and nutrition-derived indicators for predicting amputation risk in patients with type 2 diabetic foot ulcers.

MethodsX. 2026 Dec;17 104023

A hybrid CNN-GNN and multitask learning pipeline for improving mild diabetic retinopathy sensitivity.

Salsabila Amalia Harjanto, Rizka Wakhidatus Sholikah, Irzal Ahmad Sabilla, Gagatsatya Adiatmaja.

  Diabetic Retinopathy (DR) is a progressive microvascular complication of diabetes that can lead to irreversible vision loss if early-stage abnormalities remain undetected. Mild DR is particularly challenging to identify due to subtle micro-lesions, severe class imbalance, and limited robustness of conventional deep learning models under domain shift. To address these challenges, this article introduces a hybrid CNN-GNN multitask pipeline designed to improve sensitivity toward Mild DR. The proposed method integrates CNN for deep feature extraction, followed by graph-based representations that explicitly model spatial relationships between retinal regions. Both grid-based and adaptive k-nearest neighbor graph construction strategies are supported. A graph neural network performs message passing on these representations, and multitask learning is applied using two prediction heads: multiclass DR grading and binary Mild vs Non-Mild classification.•The proposed Hybrid CNN-GNN Multitask Pipeline improves sensitivity toward Mild Diabetic Retinopathy by integrating convolutional feature extraction with graph-based spatial modelling.•Multitask supervision enables enhanced Mild DR detection without substantial trade-off in overall classification accuracy, supporting balanced screening performance.•Validation across in-domain and cross-domain scenarios demonstrates the method's ability to consistently prioritize Mild DR detection, with evaluation protocol provided.

Keywords:  Convolutional neural network; Diabetic retinopathy; Graph neural network; Medical image analysis; Mild diabetic retinopathy; Multitask learning; Retinal fundus images

DOI:  https://doi.org/10.1016/j.mex.2026.104023
Int Wound J. 2026 Jul;23(7): e70982

Prediction of Factors Influencing the Incidence of Diabetic Foot Ulcers Using Classical Statistical and Machine Learning Approaches: A Systematic Review.

Alireza Jafarkhani, Mehdi Jafari Oori, Masoud Arabfard, Kiavash Hushmandi, Mohammad Pourebrahimi.

  Diabetic foot ulcers are among the most challenging complications of diabetes. Diabetes heightens patients' risk of severe complications, including amputations and death, while also driving up healthcare system costs. Given the significance of this problem, the present study conducted a systematic review of studies that used Classical Statistical and Machine Learning Approaches to identify factors influencing the development of diabetic foot ulcers in individuals with diabetes. A thorough literature search was conducted in PubMed, Scopus and Web of Science, covering their inception through 7 September 2025. This systematic review adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMAs) guidelines. Data were analysed narratively using content analysis. Eligible studies used predictive methods to identify factors associated with the development of diabetic foot ulcers. A total of 4396 articles were screened, and 66 studies were selected for full-text review after application of the inclusion and exclusion criteria. The review of these studies identified 95 factors associated with predicting diabetic foot ulcers, among which neuropathy, diabetes duration, age, body mass index and peripheral vascular disease were the most frequently reported. In addition, among the predictive models used in the studies, logistic and Cox regression models were the most useful for predicting factors associated with diabetic foot ulcers. This study identifies key predictive factors for diabetic foot ulcers, enabling healthcare systems to target high-risk patients through early screening. Proactive identification of vulnerable diabetic patients can prevent severe complications, such as amputation.

Keywords:  diabetes mellitus; diabetic foot; machine learning; systematic review

DOI:  https://doi.org/10.1111/iwj.70982
Front Endocrinol (Lausanne). 2026 ;17 1827506

Dynamic evolution of readmission risk factors across short-, medium-, and long-term horizons in type 2 diabetes: a machine learning-based predictive modeling study with SHAP interpretability.

Lei Li, Sheng Jiang.

   Background: T2DM readmission risk factors may evolve across time windows, but this dynamic remains poorly understood.
Methods: This retrospective cohort study developed nine machine learning models to predict 30-day, 60-day, and 365-day readmission in 12,041 T2DM patients (with an additional 2,007 patients used for temporal validation of the 30-day and 60-day models). Feature selection was performed using LASSO and Boruta. SHAP analysis was used for interpretability, with temporal validation performed for short- and medium-term models.
Results: ANN achieved the highest AUROC for 30-day and 60-day predictions. Random forest showed competitive performance for 365-day prediction. SHAP analysis revealed a dynamic evolution: age dominated the 30-day window; length of hospital stay and inflammatory markers (SII, SIRI) emerged as key predictors in the 60-day window; and diabetes-specific chronic complications dominated the 365-day window.
Conclusion: Model selection should be time window-specific: ANN for short/medium-term, random forest for long-term prediction. Risk factors shift from acute vulnerability to inflammatory burden and then to chronic complications, supporting dynamic risk monitoring in T2DM patients.

Keywords:  SHAP; artificial neural network; dynamic risk factors; machine learning; readmission prediction; temporal validation; type 2 diabetes

DOI:  https://doi.org/10.3389/fendo.2026.1827506
Diagnostics (Basel). 2026 Jun 30. pii: 2040. [Epub ahead of print]16(13):

Machine Learning Integration of Clinical and Molecular Biomarkers to Predict Vascular Complications in Type 2 Diabetes.

Gerardo García-Gil, Víctor Manuel Medina-Pérez, Joaquín Becerra-Contreras, José Alfonso Cruz-Ramos, Esteban González-Díaz, Héctor Raúl Pérez-Gómez, Kevin Javier Arellano-Arteaga, Arailym Yessenbekova, Botagoz Ussipbek, Nurzhanyat Ablaikhanova, Iryna Rusanova, Gabriela Del C López-Armas.

  Background/Objectives: Type 2 diabetes mellitus (T2DM) is a major global health challenge due to its high prevalence and association with chronic complications, highlighting the need for reliable predictive tools to support clinical decision-making. Methods: This study proposes a two-stage hierarchical prediction system based on a Random Forest (RF) classifier. In Stage 1, the model performs multiclass classification into healthy (H), T2DM without complications (D), and T2DM with complications (C). In Stage 2, patients classified as C are further stratified into microvascular or macrovascular complications. The dataset included 31 biochemical, molecular, inflammatory, and oxidative stress variables from Mexican and Spanish cohorts. Feature selection was performed using Pearson correlation, and feature relevance was further assessed using RF importance measures. Model training used stratified cross-validation, with additional evaluation on a hold-out set to approximate real-world performance. Results: The optimized RF achieved an accuracy of 92% and a macro F1-score of 0.92, outperforming baseline models, with an AUC-ROC of 0.89 for complication prediction. Key predictive features included IL-18, miR-126, duration of T2DM, HbA1c, and IL-10. Conclusions: The novelty of this study lies in integrating heterogeneous biomarkers within a hierarchical predictive framework, rather than in the machine learning algorithm itself. This multimodal approach, combined with interpretable machine learning techniques, is designed to deliver clinically meaningful insights for patient stratification and personalized management in T2DM.

Keywords:  Random Forest; clinical decision support; diabetes mellitus Type 2; machine learning; macrovascular; microvascular; predictive modeling

DOI:  https://doi.org/10.3390/diagnostics16132040
Front Endocrinol (Lausanne). 2026 ;17 1871923

Development and validation of a multimodal interpretable machine learning model for the identification of osteoporosis in patients with type 2 diabetes mellitus: a multicenter retrospective study.

Jihao Cheng, Chuanjiang Liu, Mengyin Gu, Dongying Su, Jingyun Liao.

   Background: Osteoporosis is a prevalent yet underdiagnosed complication in patients with type 2 diabetes mellitus (T2DM), significantly increasing the risk of fractures and mortality. However, current screening tools are limited by low accuracy and lack of generalizability. This multicenter cross-sectional study aimed to develop and validate a multimodal interpretable machine learning (ML) model integrating multimodal data to identify and classify osteoporosis risk in T2DM patients, with a focus on clinical translatability.
Methods: We retrospectively enrolled 1,002 T2DM patients from two tertiary hospitals in China. After exclusions, Cohort 1 (n = 852) was used for model development and internal validation, and Cohort 2 (n = 126) served as an independent external validation set. Multimodal data included demographics, laboratory tests, abdominal CT-derived parameters (e.g., skeletal muscle index at L3, SMI-L3), and composite metabolic indices (e.g., lymphocyte-to-HDL ratio [LHR], Metabolic Score for Visceral Fat [METS-VF]). Core predictors were selected using univariate analysis, Least Absolute Shrinkage and Selection Operator (LASSO) regression, and the Boruta algorithm. Seven ML algorithms were compared, with model performance evaluated by area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, F1-score, and area under the precision-recall curve (AUPRC). The SHapley Additive exPlanations (SHAP) method was applied for model interpretability.
Results: Seven predictors were identified: age, hemoglobin, neutrophil count, uric acid, LHR, SMI-L3, and METS-VF. The eXtreme Gradient Boosting (XGBoost) model demonstrated superior performance, achieving an AUC of 0.877 (95% confidence interval [CI]: 0.830-0.923) in the training set and 0.911 (95% CI: 0.879-0.943) in the external validation set. SHAP analysis revealed age, hemoglobin, and uric acid as the top contributors. Calibration and decision curve analyses confirmed the model's clinical utility.
Conclusion: This study presents a robust, multimodal interpretable ML model for osteoporosis risk prediction in T2DM using routinely available clinical data. By integrating multimodal features and providing transparent predictions via SHAP, the model offers a proof-of-concept tool for opportunistic screening to facilitate targeted dual-energy X-ray absorptiometry (DXA) referral, although prospective clinical validation is needed before its clinical utility can be established.

Keywords:  SHAP; machine learning; multicenter study; osteoporosis; type 2 diabetes mellitus

DOI:  https://doi.org/10.3389/fendo.2026.1871923
BMJ Open. 2026 Jul 16. 16(7): e115135

Machine learning evaluation of clinical, social and behavioural factors influencing progression from pre-diabetes to type 2 diabetes: a retrospective cohort study in southeast Michigan.

Ewelina Niedzialkowska, Sujoy Roy, Mary Townsend, Kevin E Heinrich, Berk Celik, Ramin Homayouni, Alexandra Halalau.

   OBJECTIVE: We aimed to use machine learning (ML) models to investigate the impact of clinical, social and behavioural factors on 1-year progression from pre-diabetes to type 2 diabetes mellitus (DM).
DESIGN: A retrospective cohort study.
SETTING: A large health system including eight sites in southeast Michigan.
PARTICIPANTS: Adults with haemoglobin A1c (HbA1c) between 5.7% and 6.4% for two consecutive years between 1 January 2008 and 31 December 2023, and no prior history of type 2 DM or metformin use.
PRIMARY OUTCOME MEASURE: New-onset type 2 DM (HbA1c ≥6.5%) in 1 year.
RESULTS: Among 11 809 individuals, 815 (6.9%) progressed to type 2 DM within 1 year. CatBoost demonstrated the best performance (average area under the curve 0.78). Prior-year HbA1c was the most influential covariate (SHapley Additive exPlanations 1.02). Traditional metabolic factors (high-density lipoprotein, body mass index (BMI), white blood cell, age, gender, triglycerides) also contributed. Lastly, while inclusion of social and behavioural determinants of health (SBDH) did not significantly improve the overall model performance, depression emerged as the prominent SBDH covariate. Depression was a stronger predictor of diabetes progression in individuals with higher baseline BMI and lower baseline HbA1c.
CONCLUSIONS: Inclusion of social and behavioural covariates provided no incremental value for prediction of diabetes progression from pre-diabetes. However, machine learning revealed that depression may play a role in progression to type 2 DM.

Keywords:  Depression & mood disorders; Diabetes Mellitus, Type 2; Machine Learning; PREVENTIVE MEDICINE; Preventive Health Services

DOI:  https://doi.org/10.1136/bmjopen-2025-115135
Sci Rep. 2026 Jul 17.

Interpretable diabetes risk prediction: a comparative study of tree-based algorithms using SHAP and LIME.

Ruth Reátegui, Lourdes Ramírez-Cerna, Estefanía Bautista-Valarezo.

  The global burden of diabetes mellitus continues to increase, particularly in low- and middle-income countries. Prior studies have explored the use of machine learning algorithms to predict the risk of this disease; however, evidence from Latin American populations remains limited. This study aims to predict the risk of diabetes mellitus using data from Ecuador and tree-based algorithms, including Gradient Boosting, Extreme Gradient Boosting, Light Gradient Boosting Machine, and CatBoost. The dataset comprises sociodemographic variables, anthropometric measurements, dietary and physical activity habits, clinical indicators, and family history of diabetes, among others. Extreme Gradient Boosting demonstrated the best balance between training and testing performance, achieving an area under the curve (AUC) of 0.99 and 0.96, respectively. Additionally, explainable artificial intelligence techniques, specifically SHAP and LIME, were applied to assess the contribution of variables to the prediction task. The results indicate that elevated blood glucose levels, family history of diabetes, abdominal circumference, age, visceral fat, low physical activity, waist-to-hip ratio, and muscle mass index are key factors contributing to diabetes risk prediction. These findings highlight the potential of interpretable machine learning models for early detection of diabetes in low- and middle-income contexts, including Ecuador.

Keywords:  Diabetes; Machine learning; XAI

DOI:  https://doi.org/10.1038/s41598-026-62441-2
Sci Rep. 2026 Jul 15.

Interpretable machine learning for predicting mild cognitive impairment in elderly patients with type 2 diabetes mellitus: model development and performance assessment.

Beibei Dong, Le Wang, Wenwen Shi, Lujia Li.

  Type 2 diabetes mellitus (T2DM) is a prevalent chronic condition, particularly in the elderly, and is associated with an increased risk of cognitive decline, including mild cognitive impairment (MCI). This study aimed to develop and validate an interpretable machine learning (IML) model to predict MCI in elderly T2DM patients using routine clinical data. A retrospective cohort of 923 elderly T2DM patients (≥ 60 years) was analyzed, with data collected from January 2021 to January 2025. Key MCI predictors were selected using a two-stage feature selection process involving Boruta and least absolute shrinkage and selection operator (LASSO). Six machine learning (ML) algorithms-logistic regression (LR), extreme gradient boosting (XGBoost), support vector machine (SVM), k-nearest neighbor (KNN), random forest (RF), and decision tree (DT)-were trained and evaluated. SHapley Additive exPlanations (SHAP) were employed to interpret model predictions and provide insights into feature importance. Among 923 participants, 424 (45.9%) had MCI. Baseline characteristics were comparable between the training and validation sets, with similar MCI prevalence (46.1% vs. 45.5%). Seven predictors were consistently selected: age, years of education, duration of diabetes, regular physical activity, cerebrovascular disease, glycated hemoglobin (HbA1c), and fasting plasma glucose (FPG). Among the six models, the RF model demonstrated the best overall performance, achieving an AUC of 0.842 in the validation set, with favorable discrimination, calibration, and net clinical benefit. SHAP analysis identified duration of diabetes as the most influential predictor, followed by HbA1c and FPG, emphasizing the role of both cumulative and current glycemic burden. An interpretable RF-based model using routine clinical data effectively predicted MCI in elderly patients with T2DM and provided clinically intuitive explanations of risk drivers. This approach may support risk-stratified cognitive screening and individualized management; external multicenter validation and prospective evaluation are warranted.

Keywords:  Elderly; Interpretable machine learning; Mild cognitive impairment; SHapley Additive exPlanations; Type 2 diabetes mellitus

DOI:  https://doi.org/10.1038/s41598-026-60337-9
Front Endocrinol (Lausanne). 2026 ;17 1866694

Deep learning-derived retinal biomarker associated with diabetes-related amputation in type 2 diabetes.

Junseok Park, Jung Soo Yoon, Sahil Thakur, Dongjin Nam, Sunjin Hwang.

  Diabetic foot (DF)-related amputation remains a major cause of morbidity in patients with type 2 diabetes, yet biomarkers associated with amputation risk remain limited. We aimed to assess the independent association between a deep learning (DL)-derived retinal biomarker using coronary artery calcification (Dr. Noon CVD) and DF-related amputation in patients with type 2 diabetes. This retrospective observational study conducted in a university hospital in South Korea included 392 individuals with type 2 diabetes receiving ophthalmic care (79 with DF-related amputation and 313 without). Participants were randomly split into training (70%) and validation (30%) sets. Model performance in relation to DF-related amputation was assessed using area under the receiver operating characteristic curve (AUC), continuous net reclassification index (cNRI), and integrated discrimination improvement (IDI). Prespecified rule-out (sensitivity ≥0.85) and rule-in (specificity ≥0.90) thresholds were also applied. Adding the retinal biomarker to a basic clinical model showed incremental association with DF-related amputation in the validation set (AUC 0.146, 95% CI: 0.046-0.249; cNRI 0.629, 95% CI: 0.184-1.027; IDI 0.062, 95% CI: 0.012-0.110). The full model achieved an AUC of 0.791, and the association remained consistent across sensitivity analyses using DF risk score derived from the external dataset. Under a 27% amputation prevalence (based on the reported prevalence of amputation among patients with diabetic foot), sensitivity was 87.5% with a negative predictive value of 92.6% at the rule-out cutoff; specificity was 90.4% with a positive predictive value of 56.3% at the rule-in cutoff. A DL-derived retinal biomarker shows meaningful association with DF-related amputation beyond conventional diabetes variables.

Keywords:  amputation; artificial intelligence; deep learning; diabetes; diabetic foot; retina

DOI:  https://doi.org/10.3389/fendo.2026.1866694
Sci Rep. 2026 Jul 13.

MS-EGT-Net: a multi-scale enhanced graph-transformer network for diabetic foot ulcer classification.

Minghui Cong, Zhuli Xiu, Jing Zhou, Hongqiao Sun.

  Diabetic foot ulcers (DFUs) are a serious complication of diabetes, and accurate, timely classification is crucial for supporting clinical decision-making. However, many existing approaches struggle to adequately capture the multi-scale and spatially heterogeneous characteristics of DFU morphology, which often limits their ability to integrate fine grained tissue details with broader contextual patterns. To address these limitations, we propose the Multi-Scale Enhanced Graph-Transformer Network (MS-EGT-Net), which uses a dual-scale feature extraction framework to process both high-resolution and downsampled region-of-interest (ROI) patches with a shared backbone. This design maintains semantic consistency while simultaneously capturing microscopic textures and global wound structures. In addition, a selective token refinement mechanism prunes less informative regions based on attention weight analysis, thereby retaining diagnostically relevant areas and enriching them with contextual information. The model further incorporates an adaptive graph encoding strategy that combines semantic affinity with spatial proximity to represent histopathological relationships and local structural coherence, and a bidirectional cross-scale attention module that promotes reciprocal integration between local and global features to form a more comprehensive diagnostic representation. Experimental results demonstrate that MS-EGT-Net consistently outperforms state-of-the-art methods across multiple evaluation metrics, indicating that it provides an effective solution for DFU classification with strong potential for clinical application.

Keywords:  Cross-attention fusion; Diabetic foot ulcer; Graph-transformer; Multi-scale extraction

DOI:  https://doi.org/10.1038/s41598-026-60024-9
Comput Methods Biomech Biomed Engin. 2026 Jul 13. 1-27

Prism-refraction K-means polar lights optimizer-driven feature selection for accurate diabetes diagnosis.

Yufeng Chen, Yanbo Ding, Ibrohimov Alisherjon, Xumin Chen, Huiling Chen, Chunhui Zhang.

  This study developed a machine learning framework for stratifying abnormal glucose tolerance risk among outpatients presenting with foamy urine. It retrospectively analyzed 64 features from 511 patients with foamy urine in the outpatient medical record system. Patients were divided into two groups based on oral glucose tolerance tests (OGTT) results. This study proposed a novel predictive framework, bPKPLO-SVM, which integrates the Prism-refraction K-means Polar Lights Optimizer (PKPLO) with the Support Vector Machine (SVM) for feature selection and early diabetes diagnosis. The proposed PKPLO algorithm enhances the Polar Lights Optimizer (PLO) by incorporating mechanisms for improved search diversity and solution refinement, inspired by Prism refraction and K-means clustering. Results showed that bPKPLO-SVM performed well in early diabetes prediction, achieving an accuracy of 81.639% and a recall of 85.927%, outperforming many common prediction models and similar algorithms. Model analysis further identified fasting plasma glucose (FPG), glycosylated hemoglobin (HbA1c), platelet distribution width-standard deviation (PDW-SD), urine specific gravity (USG), and age as frequently selected predictors. These findings suggest that bPKPLO-SVM may serve as a supportive risk-stratification tool for patients presenting with foamy urine, rather than a universal screening model for the general population.

Keywords:  Diabetes diagnosis; feature selection; machine learning; optimization algorithms; polar lights optimizer; support vector machine

DOI:  https://doi.org/10.1080/10255842.2026.2698059
Acta Diabetol. 2026 Jul 13.

Interpretable single-cell machine learning identifies PINK1-centred mitophagy as a determinant of metabolic fitness in type 2 diabetes.

Bingjun Zeng, Da Qin, Xiaowei Xu, Dayun Tao, Ying Jiang, Hao Zhou, Yanli Wang.

   BACKGROUND: Progressive pancreatic β-cell dysfunction constitutes a hallmark of type 2 diabetes (T2D), yet the molecular programmes governing metabolic fitness deterioration remain incompletely characterised at single-cell resolution.
METHODS: Single-cell transcriptomic data (GSE221156) encompassing pancreatic islet cells from 48 donors (non-diabetic [ND], n = 17; pre-diabetic [PD], n = 14; T2D, n = 17) were analysed using an interpretable machine learning framework that integrated sparse rule-based classification, pathway-constrained modelling, and mitochondrial fitness indexing.
RESULTS: Five transcriptionally distinct β-cell subtypes (β1-β5) were resolved. The β1 subtype exhibited peak proportional representation in PD donors (35.0%) compared with ND (16.7%) and T2D (21.1%), consistent with stress-induced adaptive expansion. Mitochondrial biogenesis and ER stress emerged as the top-ranked programmes discriminating T2D from ND β-cells (importance scores = 0.329 and 0.329). Within the mitophagy gene set, MFN2 exhibited the highest feature importance (|logFC| = 0.472), followed by PINK1 (|logFC| = 0.199). PINK1 expression progressively declined across β1→β5 subtypes. A mitochondrial fitness index integrating mitophagy, proteostasis, biogenesis, and oxidative phosphorylation achieved R² = 0.65 against module-based quality scores. Hypergeometric enrichment analysis of differentially expressed genes between the adaptive β1 subtype and dysfunctional β2-β5 subtypes revealed significant enrichment of insulin secretion (p = 0.0019), endoplasmic reticulum stress (p = 0.0013), and oxidative phosphorylation (p = 0.045) pathways.
CONCLUSIONS: PINK1-centred mitophagy represents a critical determinant of β-cell metabolic fitness in T2D, with implications for therapeutic strategies aimed at preserving islet function during disease progression.

Keywords:  Interpretable machine learning; Mitophagy; PINK1; Single-cell transcriptomics; Type 2 diabetes; β-cell heterogeneity

DOI:  https://doi.org/10.1007/s00592-026-02743-z
Diabetes Obes Metab. 2026 Jul 14.

Association of Variability of Monthly Continuous Glucose Monitoring-Derived Metrics With Diabetic Kidney Disease in Type 1 Diabetes.

Byeongjae Kang, Rosa Oh, Taeyoung Kim, Jee Hee Yoo, Danbee Kang, Gyuri Kim, Sang-Man Jin, Myung Jin Chung, Jae Hyeon Kim.

   BACKGROUND: While HbA1c is the standard for monitoring long-term glycaemic control, it fails to capture glycaemic variability. We investigated the discriminatory capacity of longitudinal continuous glucose monitoring (CGM) metrics and identified CGM metric patterns associated with diabetic kidney disease (DKD) in individuals with type 1 diabetes (T1D) using machine learning (ML).
METHODS: We analysed cross-sectional data from 282 T1D patients with 1-year consecutive CGM data. DKD was defined by persistent laboratory abnormalities (urine albumin-creatinine ratio ≥ 30 mg/g or estimated glomerular filtration rate < 60 mL/min/1.73 m2) confirmed by at least two measurements within the 1-year period. LightGBM, XGBoost, Random Forest, and Logistic Regression (LR) were developed. Feature importance was assessed using SHAP analysis.
RESULTS: The LightGBM model achieved the highest performance (AUROC = 0.91 [95% CI, 0.88-0.93], F1 score = 0.65). All tree-based ML models outperformed the LR model. SHAP analysis identified the standard deviation (SD) of monthly time in range (TIR) and time in tight range (TITR) as the most influential features. In contrast, the CV of sensor glucose did not differ significantly between groups (p = 0.416). Even in the early DKD subgroup, the SD of monthly TIR and TITR remained significantly elevated.
CONCLUSION: The SD of monthly TIR and TITR is strongly associated with DKD in T1D, whereas the CV of sensor glucose is not. ML-based integration of these longitudinal metrics offers improved discrimination of concurrent DKD status beyond conventional glycaemic markers.

Keywords:  continuous glucose monitoring; diabetes mellitus type 1; diabetic kidney disease; machine learning

DOI:  https://doi.org/10.1111/dom.71117
Ophthalmol Sci. 2026 Aug;6(8): 101262

Early Prediction of Diabetic Macular Edema via Machine Learning Survival Analysis on Checkup Data.

Yusuke Kashiwagi, Katsuyuki Chida, Ayaka Hananoe, Ryoichi Ishibashi, Masaya Koshizaka, Yoshiro Maezawa, Yosuke Inaba, Yoko Takatsuna, Tomoaki Tatsumi, Hiroko Inoue, Hanae Wakabayashi, Tetsuo Ishikawa, Akiko Hanai, Koutaro Yokote, Katsuhiko Asanuma, Eiryo Kawakami.

   Purpose: To predict the risk of diabetic macular edema (DME) onset and to identify features of the risk subgroups.
Design: Population-based observational study with a case-control design.
Subjects and Controls: Health checkup and diagnosis data from the JMDC claims database (January 2005-July 2020), one of the largest Japanese epidemiological databases, were used. From 272 337 individuals diagnosed with type 2 diabetes (International Classification of Diseases, 10th Revision E11), we analyzed 2368 pairs of DME and non-DME individuals, which matched 1:1 by the balancing score calculated from regression analysis of DME with sex, age, months of observation, number of checkups, duration of diabetes, and months until the first checkup.
Methods: We employed a multivariate Cox proportional hazards model, regularized Cox models, and a random survival forest (RSF). These models were trained via ID-level bootstrap resampling using 43 health checkup variables (missing ratio <50%) and 404 high-incidence diseases within 6 months, along with age, sex, and duration of diabetes mellitus, to assess DME risk. Temporal changes in RSF-predicted risk scores were analyzed using nonlinear modeling techniques.
Main Outcome Measures: Concordance index (C-index), integrated Brier score (IBS), and cumulative/dynamic mean area under the receiver operating characteristic curve (AUC).
Results: Thirteen checkup items and 44 disease history variables were significantly associated with the onset of DME. The RSF identified 43.8% of DME cases >5 years prior to onset, with a specificity of 85.5%. The RSF achieved median C-index, IBS, and mean AUC values of 0.694 (95% confidence interval, 0.688-0.697), 0.181 (0.179-0.184), and 0.750 (0.739-0.756), respectively, outperforming both multivariate and univariate Cox models. Three distinct DME risk subgroups were suggested by temporal changes in the RSF-predicted risk score. Predictors of DME onset varied markedly among these subgroups. In the explicit high-risk subgroup, urinary protein and urinary sugar were highly important, and liver function-related blood tests, such as alanine transaminase and γ-gultamyltransferase, were also ranked high in variable importance metrics. Anemia-related laboratory tests were associated with DME development only in this subgroup.
Conclusions: Random survival forest demonstrated superior performance in relative risk prediction of DME using health checkup data. External validation remains an essential prerequisite before any clinical application.
Financial Disclosures: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Keywords:  Claims data; Diabetic macular edema; Health checkups; Machine learning; Survival analysis

DOI:  https://doi.org/10.1016/j.xops.2026.101262
Nutrients. 2026 Jul 07. pii: 2205. [Epub ahead of print]18(13):

A Hybrid CNN-MLLM Architecture for Image-Based Nutrition Estimation and Advisory Insulin Decision Support in Type 1 Diabetes.

Jean Chrinot Velombe, Sema Bayraktar, Adnan Kavak, Muhammad Jamil, Alpaslan Burak İnner, Gautam Srivastava, Hossein Fotouhi.

   BACKGROUND/OBJECTIVES: Accurate estimation of meal composition from food images can support safer and more reliable insulin bolus decision-making for individuals with Type 1 diabetes. Existing food recognition and nutrition estimation systems are often designed for general dietary logging and do not directly integrate food analysis with personalized insulin therapy parameters.
METHODS: This study presents an image-based nutrition estimation and insulin decision-support module developed within the AI-assisted Diabetes Care (AIDCARE) platform. The proposed system uses a convolutional neural network (CNN) to classify food items from a single meal image, and retrieves reference nutritional values from a food composition database. A separate multimodal large language model (MLLM)-based estimation component is then used to estimate portion size, allowing carbohydrate and nutrient values to be scaled according to the observed serving.
RESULTS: A curated food image dataset containing 40 food categories was used to evaluate three CNN architectures: ResNet50, Inception V3, and EfficientNet-B0. EfficientNet-B0 achieved the best classification performance, with 94.91% validation accuracy, 95.55% precision, 94.87% recall, and 94.90% F1-score. The portion-estimation component achieved an MAE of 12.27 g and an RMSE of 15.11 g. The estimated carbohydrate value is combined with user-specific clinical parameters, including the insulin-to-carbohydrate ratio and insulin sensitivity factor, to generate advisory bolus guidance. To support safety, the system requires user confirmation or correction of the recognized food category and estimated portion before insulin guidance is displayed.
CONCLUSIONS: The proposed system is intended for advisory decision support only and is not designed to replace clinical judgment or autonomous insulin delivery systems.

Keywords:  Type 1 diabetes; food image recognition; insulin bolus decision support; mobile health; nutrition estimation

DOI:  https://doi.org/10.3390/nu18132205
Front Endocrinol (Lausanne). 2026 ;17 1887007

Quantitative pancreatic MRI in diabetes mellitus: current advances and future directions.

Wenqi He, Chun Cao, Ping Jiang.

  Currently, the diagnosis and monitoring of diabetes mellitus (DM) primarily rely on clinical symptoms, blood glucose measurements, and laboratory blood tests. With the continuous advancement of medical imaging, magnetic resonance imaging (MRI) is increasingly being applied to the study of diabetes and its associated complications. Given that the pancreas plays a pivotal role in the pathogenesis and progression of DM, quantitative MRI techniques have emerged as powerful tools; they provide not only fundamental structural information but also visualize the pathophysiological alterations of the pancreas throughout the disease course. This literature review suggests that quantitative pancreatic MRI holds great promise as a novel, non-invasive biomarker for evaluating diabetic pathophysiology, facilitating early diagnosis, and monitoring therapeutic efficacy. Advancing this field further, the integration of artificial intelligence and radiomics offers powerful tools for automated high-throughput mining of these datasets, though its clinical translation remains preliminary. Current implementation gaps-most notably the critical need for multi-center external validation and standardized models-must be recognized before these deep imaging phenotypes can be reliably utilized in clinical settings.

Keywords:  MRI; artificial intelligence; pancreatic fat; quantitative imaging; type 1 diabetes; type 2 diabetes

DOI:  https://doi.org/10.3389/fendo.2026.1887007
Malays Fam Physician. 2026 ;21 26

Narrative review of the development of an ischaemic heart disease prognostic scoring tool (i-IHD score) among patients with type 2 diabetes mellitus in Malaysia.

Muhammad Muzzammil Mohamad Salleh, Sazzli Shahlan Kasim, Tajul Rosli Razak, Nazar Mohd Azahar, Mohamad Rodi Isa.

   Introduction: Ischaemic heart disease (IHD) remains a major cause of mortality among individuals with type 2 diabetes mellitus (T2DM) in Malaysia. Conventional cardiovascular risk models, such as the Framingham risk score, often show limited calibration in Asian populations. Artificial intelligence (Al)-calibrated models have emerged as potential alternatives, yet their generalisability and clinical utility across different populations remain uncertain. This narrative review aimed to summarise existing prognostic models for IHD in patients with T2DM and identify methodological gaps relevant to the development of a locally calibrated model.
Methods: This narrative review employed a structured search strategy guided by PRISMA principles but was not conducted as a full systematic review. We synthesised evidence from epidemiological and prognostic research. Studies comparing conventional statistical approaches (e.g. logistic regression and Cox models) with AI-calibrated models such as extreme gradient boosting, random forest and support vector machines were reviewed.
Results: Eleven studies met the inclusion criteria; four used conventional statistical methods, and seven applied AI or machine learning algorithms. The reported discrimination (area under the curve=0.66-0.94) varied widely. Conventional models commonly lacked external validation and demonstrated restricted applicability beyond their original cohorts. AI-calibrated models showed promising discrimination in some datasets but similarly experienced limited validation and lacked benchmarking against traditional statistical methods. Across the studies, limited calibration and validation reduced generalisability to heterogeneous Malaysian populations.
Conclusion: Developing a locally AI-calibrated i-IHD score could enable early risk identification, guide targeted interventions and support national health initiatives, including the Health White Paper 2023 and 13th Malaysia Plan.

Keywords:  Artificial intelligence; Diabetes mellitus; Malaysia; Myocardial ischaemia; Prognosis; Type 2

DOI:  https://doi.org/10.51866/rv.1027
JMIR Diabetes. 2026 Jul 15. 11 e93822

Assessing Quality, Readability, and Transparency of Online and AI-Generated Information on Type 2 Diabetes Mellitus: Cross-Sectional Exploratory Study.

Sripradha Srinivasan, Ravanth Baskaran, Srinjay Mukhopadhyay, Mayank Patel.

   Background: Type 2 diabetes mellitus (T2DM) affects approximately 590 million people worldwide, and its management relies heavily on patient education. With the emergence of online health information and artificial intelligence (AI) large language models, patients are increasingly sourcing medical information independently.
Objective: This study compared the quality, readability, and transparency of websites and AI-generated leaflets (AIGLs) related to T2DM.
Methods: Four predefined search terms ("type 2 diabetes," "type 2 diabetes mellitus," "T2DM," and "adult diabetes") were entered into 3 major search engines (Google, Yahoo, and Bing), and the top 20 search results were retrieved. AIGLs with patient information about T2DM were produced using a standardized prompt in 4 AI large language models (ChatGPT, Gemini, DeepSeek, and Grok). Information quality was assessed using the DISCERN score, calculated by 3 independent raters and ChatGPT. The Journal of the American Medical Association (JAMA) benchmarks were used to measure reliability and transparency. The Flesch-Kincaid Grade Level was used to determine readability.
Results: Seventy-five websites and 4 AIGLs were evaluated. Mean author-rated DISCERN scores were 42.6 (SD 11.3) for websites and 43.9 (SD 1.74) for AIGLs, corresponding to fair quality (DISCERN 41-51). In contrast, ChatGPT-rated mean DISCERN scores were higher, with 58.5 (SD 11.5) for websites and 61.0 (SD 2.94) for AIGLs, corresponding to good quality (DISCERN 52-63). Mean JAMA benchmark scores were 2.74 (SD 0.965) for websites, indicating moderate reliability (2-3 out of 4 points), whereas all AIGLs scored 0 out of 4 points. Mean Flesch-Kincaid Grade Level scores for websites were 8.67 (SD 2.23) and 8.30 (SD 1.92) for AIGLs, corresponding to an eighth- to ninth-grade comprehension level. Spearman rank correlation demonstrated minimal variability among the 3 independent raters but showed a significant difference between ChatGPT and the 3 independent raters.
Conclusions: Given the high prevalence of T2DM, both websites and AIGLs demonstrated suboptimal quality, readability, and transparency. Increasing patient reliance on digital health information calls for improved readability standards and stronger safeguards for AI-generated content. Both websites and AIGLs require an eighth- to ninth-grade comprehension level, far above the average reading age of 9 years in the United Kingdom (fourth- to fifth-grade level). This reduces the accessibility of online health information. The landscape of medical consultations is evolving, with patients increasingly presenting with preconceived notions based on online health information; hence, health care professionals should adapt to this shift.

Keywords:  AI; ChatGPT; artificial intelligence; digital health; online patient information; type 2 diabetes mellitus

DOI:  https://doi.org/10.2196/93822
Front Nephrol. 2026 ;6 1862772

Plasma proteomics based on machine learning predict the early risk of kidney outcomes in patients with DKD: a prospective cohort study from UK Biobank.

Li Jiang, Tingting Li, Haojun Zhang, Xiai Wu, Tingting Zhao.

   Background: Diabetic kidney disease (DKD) is the predominant cause of end-stage kidney disease worldwide. Early identification of patients at heightened risk for adverse kidney outcomes remains a critical unmet clinical need.
Methods: We conducted a prospective cohort study of 918 DKD participants from the UK Biobank. Baseline plasma proteomic profiling quantified 1,463 proteins using Olink proximity extension assays. Three nested Cox proportional hazards models with incremental covariate adjustment identified proteins associated with composite kidney outcomes. To prevent information leakage, all feature selection, hyperparameter optimization, and model development were performed exclusively within the training cohort (70% random split) via a multi-stage pipeline integrating LASSO-Cox regression, Random Survival Forest, Boruta algorithm, and Sequential Forward Selection. XGBoost Cox modeling with SHAP interpretability analysis quantified variable contributions. Predictive performance was validated through Kaplan-Meier survival analysis, 15-year longitudinal trajectory modeling, ROC benchmarking, 10-fold nested cross-validation, and sensitivity analyses restricted to KDIGO-defined DKD. An interactive web application was developed for clinical translation.
Results: Of 1,463 proteins examined, 633 demonstrated significant associations with kidney outcomes across all three Cox models. Functional enrichment highlighted immune-inflammatory pathways, PI3K-Akt signaling, and chemokine cascades as central to DKD progression. The machine learning framework identified an 11-protein signature, which was refined to 10 core biomarkers (HLA-E, EFNA1, GPR158, FSTL3, ART3, GM2A, CLEC1A, CKAP4, IFNGR1, EPHA2) following survival and longitudinal trajectory validation. The protein-only model achieved robust discrimination (AUC = 0.808 [95% CI 0.767-0.848] for composite outcomes; AUC = 0.807 [95% CI 0.765-0.848] for renal death), comparable to fully integrated models incorporating demographics and metabolic variables. Systematic benchmarking across 101 algorithm combinations identified LASSO-RSF as optimal (C-index = 0.94 in training; 0.73 in independent testing), with reliable calibration through mid-term follow-up (Integrated Calibration Index = 0.019-0.022). The web-based tool enables real-time, personalized risk stratification.
Conclusions: This study establishes a validated 10-protein signature for early prediction of kidney outcomes in DKD. The systematic machine learning framework and deployable web application provide accessible, interpretable risk assessment to support precision nephrology and preemptive clinical intervention.

Keywords:  diabetic kidney disease; early risk stratification; kidney outcomes; machine learning; plasma proteomics; predictive biomarkers

DOI:  https://doi.org/10.3389/fneph.2026.1862772
Front Nutr. 2026 ;13 1875176

Inflammation and nutrition-derived indicators for predicting amputation risk in patients with type 2 diabetic foot ulcers.

Junyu He, Jingxia Sun, Jinhua Wu, Jingming Yu, Lili Zhu, Wensheng Lu.

   Background: Inflammation and malnutrition affect the prognosis of diabetic foot ulcers (DFU). This study systematically evaluated the impact and predictive value of inflammation and nutrition-derived indicators for amputation in DFU using machine learning models.
Methods: The study enrolled 1,052 patients with type 2 diabetic foot ulcers. During hospitalization, 739 patients received conservative treatment and 313 patients underwent amputation, including 259 patients who received minor amputation and 54 patients who received major amputation. Multivariate logistic regression analysis was used to investigate the correlation between the inflammation and nutrition-derived indicators and amputation in DFU. Restricted cubic splines (RCS) and receiver operating characteristic (ROC) curves were utilized to evaluate the relationship and predictive value of derived indicators with the risk of amputation. Five machine learning algorithms (logistic regression, random forest, support vector machine (SVM), extreme gradient boosting (XGBoost), and light gradient boosting machine (LightGBM)) were used to construct predictive models for amputation in DFU. The SHapley Additive exPlanations (SHAP) method was applied to analyze the contribution of each feature in the model to the predictive outcomes.
Results: Multivariate logistic regression and RCS analysis demonstrated that systemic inflammatory response index (SIRI), neutrophil-lymphocyte ratio (NLR), platelet-lymphocyte ratio (PLR), neutrophil-high-density lipoprotein (HDL) ratio (NHR), monocyte-HDL ratio (MHR), platelet-HDL ratio (PHR), and neutrophil-albumin ratio (NAR) were positively correlated with amputation in DFU. The platelet-neutrophil ratio (PNR), lymphocyte-white blood cell ratio (LWR), and the hemoglobin, albumin, lymphocyte, and platelet (HALP) score were negatively associated with amputation. The ROC curve displayed that NHR (AUC = 0.755) and NAR (AUC = 0.756) had moderate predictive performance for amputation in DFU, which was not observed in previous study. XGBoost was selected as the optimal model to predict the risk of amputation in DFU (AUC = 0.894). SHAP analysis revealed that the Wagner degree, LWR, HALP, monocyte, white blood cells and NHR were the primary predictors in the model.
Conclusion: Ten inflammation and nutrition-derived indicators demonstrated significant correlations with amputation. The XGBoost model could predict the risk of amputation in DFU, based on inflammation and nutrition-derived biomarkers (NHR, LWR, and HALP). It may help identify patients at high risk of amputation at an early stage, which could facilitate timely interventions and potentially lower the risk of amputation.

Keywords:  amputation; diabetic foot ulcer; indicators; inflammation; machine learning; nutrition

DOI:  https://doi.org/10.3389/fnut.2026.1875176