bims-aukdir 2026-10-04 papers

bims-aukdir

Biomed News

on Automated knowledge discovery in diabetes research

Issue of 2026–10–04
fifteen papers selected by
Mott Given

Ultrasound radiomics-based machine learning models for differentiating diabetic kidney disease from non-diabetic kidney disease in type 2 diabetes.
Pattern and texton-based feature descriptors for Diabetic foot ulcer detection.
Artificial intelligence driven prognostic models for predicting diabetic retinopathy risk in type 2 diabetes mellitus: A systematic review.
Ocular Factors Affecting AI Diagnosis in Diabetic Retinopathy Screening for Resource-Limited Regions: Cross-Sectional Study.
Interpretable hybrid deep learning features of pectoralis on low-dose chest CT for detecting type 2 diabetes.
Uncovering Placental Endocrine Dysregulation in Gestational Diabetes Mellitus Using Bioinformatics and Machine Learning.
Development and validation of an AI-enhanced prediction model for 3-year visual decline in patients with diabetes using ophthalmic imaging: protocol for a real-world longitudinal cohort study.
GlyRAG: Context-Aware Retrieval-Augmented Framework for Blood Glucose Forecasting.
Explainable Artificial Intelligence for Early Diabetic Kidney Disease Risk Profiling in Type 2 Diabetes: Translational Readiness and Clinical Boundaries.
Resolution-Aware Cross-Modality Evaluation of CFP-Pretrained Retinal Vision-Language Models for Ultra-Widefield Diabetic Retinopathy Grading.
Artificial intelligence for clinical decision-making in retinal disorders: From screening and diagnosis to treatment and longitudinal management.
Information Quality Requirements for AI-Generated Patient Education Materials in Diabetic Foot Ulcer Care: A Directed Qualitative Content Analysis of Patient and Nurse Perspectives.
Integrative transcriptomic and machine learning analysis identifies candidate biomarkers and immune features in diabetic ulcers.
TransUNet-GradCAM: a hybrid transformer-U-Net with self-attention and explainable visualizations for foot ulcer segmentation.
From risk classification to clinical action: a prespecified paired pilot benchmark of public large language model interfaces for diabetes-related foot ulcer prevention.

BMC Nephrol. 2026 Sep 14. pii: 563. [Epub ahead of print]27(1):

Ultrasound radiomics-based machine learning models for differentiating diabetic kidney disease from non-diabetic kidney disease in type 2 diabetes.

Wenye Chen, Dan Zhao, Lehao Wu, Wenkun Bai, Xintong Wang, Ying Ying Zhang, Jian-Ying Niu, Chen Yu.

   BACKGROUND AND OBJECTIVE: Noninvasive differentiation of diabetic kidney disease (DKD) from non-diabetic kidney disease (NDKD) in patients with type 2 diabetes mellitus (T2DM) remains challenging, as conventional clinical markers and ultrasound imaging lack sufficient accuracy and reliability. This limitation often leads to unnecessary renal biopsies or delayed diagnosis. To address this gap, this study aimed to develop and validate a machine learning model integrating renal ultrasound radiomic features and clinical predictors to distinguish DKD from NDKD in T2DM patients.
METHODS: Patients with T2DM who underwent renal biopsy were retrospectively enrolled from three centers. Based on biopsy findings, patients were classified as DKD or NDKD, with overlapping lesions assigned to DKD. Renal ultrasound images were collected for ROI delineation and radiomic feature extraction. Clinical predictors were selected using logistic regression, correlation analysis, and LASSO. SVM, KNN, RF, and XGBoost models were developed using radiomic features alone or combined with clinical variables. Performance was evaluated by AUC, accuracy, precision, recall, and F1-score, and SHAP was used for model interpretation.
RESULTS: Univariate analysis identified serum creatinine, eGFR, albumin and LDL‑C as significant variables, and eGFR was selected as the final clinical predictor after LASSO and multivariable logistic regression. The radiomics‑only model performed well in the training cohort but showed lower performance in the validation cohorts. The integrated model combining radiomic features with eGFR achieved AUCs of 0.991, 0.895 and 0.721 in the training, internal validation and external validation cohorts, respectively, with corresponding F1‑scores of 0.939, 0.815 and 0.714. DeLong's test further showed that XGBoost performed better than several comparator algorithms in the training and internal validation cohorts, whereas no significant differences were observed among models in the external validation cohort.
CONCLUSION: The XGBoost model integrating renal ultrasound radiomic features with eGFR showed favourable performance for differentiating DKD from NDKD in patients with T2DM. By using routinely available ultrasound images and a readily accessible clinical indicator, this model may serve as a non-invasive auxiliary tool for preliminary screening in primary-level hospitals or nephrology departments, providing supportive evidence for individualized assessment and renal biopsy decision-making.

Keywords:  Diabetic kidney disease; Machine learning; Radiomics; Type 2 Diabetes Mellitus (T2DM); Ultrasound

DOI:  https://doi.org/10.1186/s12882-026-05353-7
Expert Rev Med Devices. 2026 Sep 28.

Pattern and texton-based feature descriptors for Diabetic foot ulcer detection.

Mastan Vali, N Hema Rajini, M Laxmaiah.

  Diabetic foot ulcers (DFU) are a dangerous side effect of diabetes mellitus that significantly impairs a patient's health and standard of living. For efficient treatment and to avoid serious outcomes like infections and amputations, DFUs must be detected promptly and precisely. This paper introduces a novel Deep Learning Strategy-based framework for Diabetic Foot Ulcer Detection (DLS-DFUD). The DLS-DFUD framework involves four key steps: preprocessing, segmentation, extraction of features, and classification. The detection process starts with acquiring the input image, which is then preprocessed using Wavelet Transform-based Wiener Filtering (WT-WF) to minimize noise and improve image quality. The cleaned image is subsequently segmented with a Middle Convolutional layer Assisted U-Net (MCA-U-Net) model, precisely distinguishing DFU regions from healthy tissue. From this segmentation, various features are extracted, including Median Binary Patterns (MBP), Modified Pixel Computation in Multi-Texton (MPC-MT), shape features, and statistical features to capture crucial ulcer characteristics.

Keywords:  And feature extraction; Diabetic foot ulcer detection; deep learning; segmentation; wavelet transform-wiener filter

DOI:  https://doi.org/10.1080/17434440.2026.2740306
Med J Malaysia. 2026 Sep;81(5): 759-768

Artificial intelligence driven prognostic models for predicting diabetic retinopathy risk in type 2 diabetes mellitus: A systematic review.

M M S Teruna, T R Razak, S M Yasin, A A R Ali, M M M Salleh, M Z M Yusoff, M H A Khebir, N A C Rameli, M I M Sallehhudin, M S M Asraff, M R Isa.

INTRODUCTION: Diabetic retinopathy remains a leading cause of preventable blindness among individuals with type 2 diabetes mellitus and is driven by the increasing global burden of diabetes. Early identification of individuals at high risk is essential for timely intervention. In recent years, artificial intelligence, including machine learning and deep learning, has emerged as a promising approach for prognostic modelling. However, evidence on artificial intelligence-driven models for predicting diabetic retinopathy risk remains fragmented. This systematic review aims to synthesise current evidence on artificial intelligence-driven prognostic models for predicting the risk of diabetic retinopathy among individuals with type 2 diabetes mellitus, focusing on model characteristics, predictor variables, performance, and validation strategies.
MATERIALS AND METHODS: A comprehensive search of PubMed, Scopus, Web of Science, and ScienceDirect was performed for studies published between January 2016 and December 2025. Eligible studies included those applying artificial intelligence-based models for diabetic retinopathy risk prediction in adult populations with type 2 diabetes mellitus and reporting model performance metrics. Data were extracted and synthesised narratively, and methodological quality was assessed using the Newcastle-Ottawa Scale.
RESULTS: A total of 1040 records were identified, with eight studies included after screening. A wide range of artificial intelligence algorithms was applied, with ensemble models such as extreme gradient boosting and random forest demonstrating superior performance. Key predictors consistently included glycated haemoglobin levels, duration of diabetes, blood pressure, lipid profile, and renal function markers, alongside emerging metabolomic biomarkers. Model performance ranged from moderate to excellent, with area under the receiver operating characteristic curve values between 0.68 and 0.97. Most studies employed internal validation techniques such as cross-validation or training and testing data splits, while external validation was largely absent. Overall methodological quality was high, although variability in study design, predictor selection, and reporting was observed.
CONCLUSION: Artificial intelligence-driven prognostic models show substantial potential in predicting diabetic retinopathy risk among individuals with type 2 diabetes mellitus, particularly through the integration of clinical and highdimensional data. Despite promising predictive performance, limitations in external validation, heterogeneity, and reporting of clinical utility hinder translation into practice. Future research should prioritise external validation, standardised reporting frameworks, and clinically interpretable models to enhance applicability in real-world healthcare settings.
J Med Internet Res. 2026 Sep 30. 28 e85181

Ocular Factors Affecting AI Diagnosis in Diabetic Retinopathy Screening for Resource-Limited Regions: Cross-Sectional Study.

Ying Xue, Xin Hu, Shuaijie Yuan, Junfang Zhang, Hongxia Hu, Shangbo Yang, Rongrong Zhu.

   Background: Diabetic retinopathy (DR) is the leading cause of vision loss among working-age adults worldwide. AI-assisted automated image reading has effectively alleviated the human resource challenges in large-scale remote screenings, yet there is limited analysis on the impact of complex ocular factors on the diagnostic efficacy of AI.
Objective: This study aimed to systematically analyze the impact of multiple ocular factors on the diagnostic efficacy of the EVisionAI system in detecting DR and vision-threatening diabetic retinopathy (VTDR) during large-scale screening of residents with diabetes in resource-limited regions.
Methods: This cross-sectional study used a multistage stratified random sampling method to screen residents with type 2 diabetes at primary health centers in resource-limited regions. Data collection involved structured questionnaires (for basic and disease information), hemoglobin A1c testing, and comprehensive ophthalmic examinations (visual acuity, intraocular pressure, slit-lamp examination, axial length measurement, and fundus photography). Following data collection, 2 ophthalmologists independently graded fundus photographs according to the American Academy of Ophthalmology standards. The influence of various ocular factors on the diagnostic performance of EVisionAI was subsequently evaluated.
Results: Between October 21 and November 12, 2024, 1847 participants with type 2 diabetes (aged 32-91 years) were enrolled, of whom 1748 (94.6%) completed the screening process and 3392 eyes were eligible for DR analysis. Participants had a mean age of 67.17 (SD 8.52) years and mean diabetes duration of 9.13 (SD 6.97) years, with 19.2% (335/1748) having DR and 7.4% (129/1748) having VTDR. Although EVisionAI's diagnostic efficacy was comparable to that of ophthalmologists (sensitivity: 90.61%, 95% CI 87.89%-92.79%; specificity: 98.99%, 95% CI 98.52%-99.31%), some ocular factors-including pupil size, refractive media opacity, and tessellated fundus (TF)-significantly impaired its efficiency. Severe refractive media opacity and TF reduced its sensitivity to 80.95% and 82.86%, respectively, and these factors interfered more with early-stage DR detection than VTDR (97.45% detected), particularly in eyes with severe TF changes (sensitivity decreased to 60.71%). Most notably, severe vitreous degeneration-induced opacity almost invariably led to VTDR misdiagnosis. Additionally, pupil dilation improved the sensitivity of EVisionAI for diagnosing DR (excluding early-stage DR) but had minimal impact on specificity.
Conclusions: EVisionAI achieved high diagnostic accuracy for large-scale DR screening in resource-limited regions, yet its performance for early-stage disease was diminished by severe ocular factors. Optimizing for these factors is therefore essential to maximize its clinical utility in primary care settings with limited specialist access.

Keywords:  AI; artificial intelligence; diabetic retinopathy; diagnostic performance; screening; telemedicine

DOI:  https://doi.org/10.2196/85181
Front Endocrinol (Lausanne). 2026 ;17 1926903

Interpretable hybrid deep learning features of pectoralis on low-dose chest CT for detecting type 2 diabetes.

Yuning Guo, Song Wang, Jiaxun Lai, Xueyan Liu, Shunyi Lyu, Huimei Chao, Dewei Kong, Xifu Wang, Qiufeng Zhao.

   Objective: To develop an explainable hybrid deep learning features framework using low-dose chest CT (LDCT)-derived pectoralis features for opportunistic type 2 diabetes mellitus (T2DM) opportunistic screening, evaluating its incremental value over conventional biomarkers.
Methods: This multi-center study analyzed LDCT images from 1,209 individuals. Following automated MedSAM segmentation, conventional metrics (e.g., Mean_HU, SMI) and deep learning (DL) features (ResNet152) were extracted. To prevent data leakage, PCA, LASSO selection, and ExtraTrees model optimization were strictly confined to the training set. A hybrid (COM) model integrating both feature types was compared against a baseline Muscle model. SHapley Additive exPlanations (SHAP) provided clinical interpretability.
Results: The COM model achieved an area under the curve (AUC) of 0.830 (training), 0.759 (internal validation), and 0.746 (external validation). While performing comparably to the DL model without significant difference across validation cohorts (DeLong p > 0.05), it significantly outperformed the baseline Muscle model (external AUC: 0.515; DeLong test, p < 0.001). SHAP analysis confirmed that lower Mean_HU correlated with higher T2DM risk, while DL signatures provided robust, independent diagnostic value.
Conclusion: This explainable hybrid model and DL model capture sub-visual pectoralis alterations, offering significant incremental diagnostic value over simple clinical and imaging metrics. They serves as an efficient, non-invasive tool for opportunistic T2DM screening during routine LDCT.

Keywords:  deep learning; explainable artificial intelligence; low-dose chest computed tomography; pectoralis; type 2 diabetes

DOI:  https://doi.org/10.3389/fendo.2026.1926903
Am J Reprod Immunol. 2026 Oct;96(4): e70325

Uncovering Placental Endocrine Dysregulation in Gestational Diabetes Mellitus Using Bioinformatics and Machine Learning.

Lijun Deng, Mo Yang, Xin Wang, Ning Gong, Ning Zhang, Qing Yang, Chengmin Gu.

   BACKGROUND: Gestational diabetes mellitus (GDM) is a common metabolic disorder, posing serious health risks to both mother and fetus. This study aims to explore the placental endocrine mechanisms involved in GDM.
METHODS: The training dataset GSE203346 and endocrine-related gene sets from the Molecular Signatures Database were used to identify endocrine-related differentially expressed genes (ER-DEGs). Support vector machine recursive feature elimination (SVM-RFE) and Boruta algorithms were employed to identify hub genes. Diagnostic performance was analyzed using the receiver operating characteristic (ROC) curves. Immune cell infiltration was evaluated using the CIBERSORT algorithm, followed by single-gene gene set enrichment analysis (GSEA).
RESULTS: A total of 57 ER-DEGs were identified in the GDM placental samples. Machine learning identified nine important genes, of which CACNA1C, CCDC102B, and KIF21A exhibited high diagnostic performance (AUC > 0.80) in two external validation datasets (GSE154414 and GSE255075). Immune infiltration analysis revealed potential differences in several immune cell proportions between high- and low-expression groups of the hub genes. High expression of CACNA1C, CCDC102B, and KIF21A was associated with the Notch pathway, mTOR pathway, and autophagy.
CONCLUSIONS: CACNA1C, CCDC102B, and KIF21A may be the potential biomarkers and therapeutic targets for GDM.

Keywords:  diagnostic biomarkers; endocrine‐related genes; gestational diabetes mellitus; machine learning

DOI:  https://doi.org/10.1111/aji.70325
BMJ Open. 2026 Sep 30. 16(9): e123521

Development and validation of an AI-enhanced prediction model for 3-year visual decline in patients with diabetes using ophthalmic imaging: protocol for a real-world longitudinal cohort study.

Yexian Yu, Jiarui Yang, Yanyan Zhang, Shaodong Ma, Jing Hong, Shan Nan, Feng Sun.

   BACKGROUND: Diabetic retinopathy (DR) is one of the leading causes of visual impairment among working-age adults worldwide. Existing artificial intelligence (AI) studies have mainly focused on automated diagnosis and grading of DR or diabetic macular oedema while relatively few studies have investigated long-term prediction of visual outcomes. In clinical practice, patients and clinicians are more concerned about whether future visual decline will occur rather than merely identifying the current presence of DR. This study aims to develop and validate a real-world prognostic model integrating AI-derived imaging scores, ophthalmic examinations and systemic clinical variables to predict 3-year visual decline in patients with diabetes.
METHODS AND ANALYSIS: This retrospective longitudinal cohort study will use routinely collected ophthalmic imaging and clinical data from Peking University Third Hospital for model development and internal validation, with independent external validation planned using data from Ningbo Eye Hospital, Wenzhou Medical University. The development cohort will cover the period from 1 January 2018 to 1 May 2026. Eligible eyes will have baseline colour fundus photography (CFP) and/or optical coherence tomography (OCT), baseline best-corrected visual acuity and at least one follow-up visual acuity record within 3 years after the index date. CFP-enhanced analyses will include eyes with eligible CFP, OCT-enhanced analyses will include eyes with eligible OCT and multimodal analyses will require eligible paired CFP and OCT. The primary outcome will be time to first visual decline within 3 years, defined as an increase in logarithm of the minimum angle of resolution (logMAR) best-corrected visual acuity of ≥0.2 from baseline in the same eye. Sustained visual decline, requiring confirmation of a ≥0.2-logMAR deterioration at a subsequent eligible visit, will be evaluated as a key secondary outcome. AI-derived imaging scores will be generated using prespecified modality-specific and multimodal retinal image models. Separate CFP and OCT models will generate continuous CFP-derived and OCT-derived imaging scores, while a prespecified multimodal model jointly integrating paired CFP and OCT information will generate a single continuous multimodal imaging score. These imaging scores will be incorporated separately into prespecified AI-enhanced prognostic models together with demographic, systemic and ophthalmic predictors. Cox regression, penalised Cox regression and exploratory machine learning survival models will be developed and internally validated and the final prognostic model will be externally evaluated in an independent cohort from Ningbo Eye Hospital. Performance will be evaluated using discrimination, calibration, Brier score and decision curve analysis.
ETHICS AND DISSEMINATION: This study has received ethical approval from the Ethics Committee of Peking University Third Hospital (approval number: IRB00006761-M20260399). All data will be de-identified before analysis. The findings of this study will be disseminated through peer-reviewed publications and conferences.

Keywords:  Artificial Intelligence; Diabetic retinopathy; Longitudinal studies; Prognosis

DOI:  https://doi.org/10.1136/bmjopen-2026-123521
IEEE J Biomed Health Inform. 2026 Sep 28. PP

GlyRAG: Context-Aware Retrieval-Augmented Framework for Blood Glucose Forecasting.

Shovito Barua Soumma, Hassan Ghasemzadeh.

Accurate blood glucose forecasting using continuous glucose monitoring (CGM) data can support early prediction of dysglycemic risk. However, current neural-network-based forecasting models treat CGM data as a purely numerical sequence without integrating con textual information that the CGM signal morphology contains. Recently, large language models (LLMs) have shown promise for time-series forecasting tasks, yet their role as agentic context extractors in diabetes care remains largely unexplored. In this study, we bridge glucose forecasting and LLM-based contextualization, and develop GlyRAG, a context-aware retrieval-augmented forecasting framework that uses an LLM as a contextualization agent to summarize glucose morphology directly from a timed CGM window. The generated CGM-only narrative is embedded and fused with patch-based glucose representations, and a retrieval module incorporates similar historical training episodes through cross-attention. We evaluate GlyRAG on OhioT1DM and AZT1D datasets for 5-, 30-, and 60-minute forecasting horizons. Against strong CGM-only baselines, GPT-4 GlyRAG significantly improves long-horizon RMSE (Root Mean Square Error) over PatchTST on both datasets. For example, RMSE decreases from 13.8 to 10.6 at 30 minutes and from 23.1 to 20.2 at 60 minutes on OhioT1DM. LLaMA 3.1 shows smaller but significant long-horizon gains, suggesting that the contextualization pipeline is not limited to GPT-4. Clinical error-grid analyses further show that approximately 85% of predictions fall in clinically acceptable Clarke Error Grid Zones A-B. These results suggest that CGM-derived linguistic context and case-based retrieval can improve long-horizon glucose forecasting without requiring additional sensing modalities.

DOI: https://doi.org/10.1109/JBHI.2026.3737468
Diabetes Metab Syndr Obes. 2026 ;19 624805

Explainable Artificial Intelligence for Early Diabetic Kidney Disease Risk Profiling in Type 2 Diabetes: Translational Readiness and Clinical Boundaries.

Tianqiang Wu, Wenpin Cai, Zihan Guan, Zhixiang Li.

   Purpose: Early kidney-risk profiling in type 2 diabetes is difficult because albuminuria, estimated glomerular filtration rate, comorbidity, and care access capture different aspects of diabetic kidney disease. This review examines where explainable artificial intelligence may support clinically defensible kidney-risk profiling.
Methods: This narrative review used structured searches of PubMed, Semantic Scholar, and OpenAlex using kidney, diabetes, and artificial intelligence/explainability terms, supplemented by citation chasing and full-text assessment of key records. Broad query returns were documented for PubMed (600) and Semantic Scholar (956). Records were screened for topic fit and prioritized for clinical relevance to kidney-risk tasks, endpoint validity, explainability, validation, calibration, missing-data handling, fairness, and workflow relevance; 18 clinically relevant evidence records were retained for detailed synthesis in Supplementary Table 1.
Results: Evidence was organized by clinical task: occult or early diabetic kidney disease detection, incident kidney disease prediction, progression and referral forecasting, retinal and multimodal profiling, and equity-aware risk assessment. Current studies show promising discrimination and increasing use of explanation methods, calibration, decision-curve analysis, web tools, and external or temporal validation. Persistent limits include heterogeneous endpoints, diagnostic circularity, uneven external validation, sparse workflow evidence, and limited fairness or uncertainty reporting.
Conclusion: Explainable artificial intelligence is best viewed as a risk-stratification, communication, and audit layer within guideline-based care, not as a standalone diagnostic or treatment-decision system.

Keywords:  clinical decision support; diabetic kidney disease; explainable artificial intelligence; risk profiling; translational readiness; type 2 diabetes

DOI:  https://doi.org/10.2147/DMSO.S624805
J Imaging Inform Med. 2026 Sep 28.

Resolution-Aware Cross-Modality Evaluation of CFP-Pretrained Retinal Vision-Language Models for Ultra-Widefield Diabetic Retinopathy Grading.

Siyun Lee, Baek Hwan Cho, Jihyeon Son, Yujin Nam, Junhyung Moon, Joonhyung Kim.

  Retinal vision-language models (VLMs) pretrained on conventional color fundus photography (CFP) may be applied to other imaging systems without adequate modality-specific validation. We hypothesized that direct zero-shot transfer to ultra-widefield (UWF) photographs would produce class-specific failure and assessed whether linear probing showed broader five-grade diabetic retinopathy (DR) separation in a held-out set. In this retrospective single-center study, 300 de-identified UWF images from 120 eyes of 61 patients with diabetes were graded as no DR, mild, moderate, or severe nonproliferative DR, or proliferative DR. FLAIR and CLIP-DR were assessed at 512 × 512, 768 × 768, and 1024 × 1024 pixels. Both strategies were evaluated on the same 151-image test set from 24 patients; linear probes were fitted using a separate 149-image training set from 37 patients. Balanced accuracy, macro-F1, and macro-area under the receiver operating characteristic curve (macro-AUROC) were reported for both models. Class-specific and ordinal-error measures were also reported. CLIP-DR linear probing achieved macro-F1 of 0.365 at 1024 × 1024, compared with 0.254 for zero-shot inference. FLAIR zero-shot failed to identify severe nonproliferative DR at every resolution. FLAIR linear-probing macro-F1 peaked at 0.338 at 768 × 768 and decreased to 0.295 at 1024 × 1024; moderate- and severe-stage recall remained low. Direct CFP-to-UWF reuse therefore produced marked class collapse. Linear probing yielded higher macro-F1 estimates, but patient-cluster bootstrap intervals for all four primary contrasts included zero. Increasing resolution did not consistently improve performance. Modality-specific, class-resolved validation is required before retinal VLMs are reused across fundus imaging systems.

Keywords:  Diabetic retinopathy; Domain shift; Model validation; Ultra-widefield imaging; Vision-language model

DOI:  https://doi.org/10.1007/s10278-026-02349-5
Prog Retin Eye Res. 2026 Sep 29. pii: S1350-9462(26)00099-6. [Epub ahead of print]115 101533

Artificial intelligence for clinical decision-making in retinal disorders: From screening and diagnosis to treatment and longitudinal management.

Kai Jin, Kaikai Zhao, Rupesh Agrawal, Gui-Shuang Ying, Andrzej Grzybowski.

  Artificial intelligence (AI) in retinal imaging has expanded from image classification to multimodal interpretation, longitudinal prediction, and clinically oriented decision support. This narrative review focuses on four complementary clinical settings: diabetic retinopathy (DR) screening and referral, diabetic macular edema (DME) treatment assessment, neovascular age-related macular degeneration (nAMD) retreatment and longitudinal monitoring, and inherited retinal disease (IRD) diagnosis, genotype-phenotype support, progression modeling, and trial enrichment. We organize the evidence around disease-specific decision points, relevant data modalities, and the requirements for multimodal and longitudinal integration. Evidence is strongest for DR screening, supported by prospective and real-world validation, whereas treatment-oriented applications in DME and nAMD and multimodal diagnostic or prognostic applications in IRDs remain less consistently validated. We highlight the gap between model performance and clinical utility and propose a cautious translational roadmap emphasizing external validation, calibration, uncertainty handling, workflow integration, prospective evaluation, and accountable deployment.

Keywords:  Artificial intelligence; Clinical decision-making; Diabetic macular edema; Diabetic retinopathy; Inherited retinal diseases; Longitudinal monitoring; Neovascular age-related macular degeneration; Retinal disorders

DOI:  https://doi.org/10.1016/j.preteyeres.2026.101533
Patient Prefer Adherence. 2026 ;20 626817

Information Quality Requirements for AI-Generated Patient Education Materials in Diabetic Foot Ulcer Care: A Directed Qualitative Content Analysis of Patient and Nurse Perspectives.

Jingwen Song, Norafisyah Makhdzir, Zarina Haron, Chunping Qin, Li Wang.

   Purpose: To explore the information quality requirements of artificial intelligence generated patient education materials for diabetic foot ulcers from the perspectives of patients and nurses.
Patients and Methods: This single-center qualitative study was conducted at the Fifth Affiliated Hospital of Guangxi Medical University, China, from November 2025 to June 2026. Maximum-variation purposive sampling recruited 15 hospitalized patients with diabetic foot ulcers and 10 nurses. Saturation was assessed separately for each group and defined as no new codes, categories, or relevant insights. It was reached after 13 patient and 8 nurse interviews, with two additional interviews per group for confirmation. Face-to-face semi-structured interviews used standardized artificial intelligence generated diabetic foot ulcer education material as a discussion stimulus. Data were analyzed using directed content analysis guided by Wang and Strong's information quality framework.
Results: Five themes and 17 subthemes were identified. Four themes mapped to Wang and Strong's framework: intrinsic information quality (accuracy, believability, objectivity, and reputation); contextual information quality (relevancy, completeness, timeliness, appropriate amount of data, and value-added); representational information quality (ease of understanding, interpretability, consistent representation, and concise representation); and accessibility information quality (accessibility and access security). An additional specific theme comprised clinical safety and actionability. Both groups valued reliable, relevant, understandable, and accessible information, but their priorities differed: nurses emphasized clinical accuracy, professional review, and the risks of unsafe recommendations, whereas patients prioritized relevance to current wound-care needs, ease of understanding and access, and clear guidance for self-care.
Conclusion: Patients and nurses showed different but complementary priorities. Clinical safety and actionability extend conventional information quality considerations. Artificial intelligence may support educational content generation. However, a professional review is necessary for clinically sensitive information, and patient input is essential to ensure usability. Future studies should refine and validate these requirements across settings and user groups.

Keywords:  artificial intelligence; diabetic foot ulcer; directed content analysis; health education; information quality

DOI:  https://doi.org/10.2147/PPA.S626817
Mol Biol Rep. 2026 Oct 01. pii: 1669. [Epub ahead of print]53(1):

Integrative transcriptomic and machine learning analysis identifies candidate biomarkers and immune features in diabetic ulcers.

Yue Jiang, Yunxi Cai, Qingkai Liu, Jingsi Jiang, Fang Shen, Ying Zhang, Xiaoya Fei, Le Kuai, Zhan Zhang, Ying Luo, Jiankun Song, Xin Ma, Bin Li, Xiaoxuan Ma, Yi Ru.

   BACKGROUND: Diabetic ulcers (DU) are a severe complication of diabetes mellitus and are associated with infection, recurrence, amputation, and excess mortality. Robust molecular markers that distinguish ulcer tissue from non-ulcer tissue and clarify the biological basis of impaired healing remain limited.
METHODS AND RESULTS: Public transcriptomic datasets were integrated to identify DU-associated genes initially selected from lipid metabolism-related gene sets. Differential expression analysis, random forest modeling, single-sample gene set enrichment analysis (ssGSEA), gene set enrichment analysis (GSEA), Gene Set Variation Analysis (GSVA), drug-gene interaction analysis, and single-cell RNA sequencing were used to prioritize candidate biomarkers. External transcriptomic cohorts were used for validation. A streptozotocin-induced diabetic wound model in C57BL/6 mice was evaluated by serial wound imaging, hematoxylin and eosin staining, quantitative real-time polymerase chain reaction (qRT-PCR), enzyme-linked immunosorbent assay (ELISA), and western blotting. A ten-gene panel comprising ANGPTL4, BNIP3, EEF2K, EIF4EBP1, KLHDC1, KLK10, KLK8, NFIX, QSOX1 and S100A8 showed strong discrimination and reproducible expression patterns across validation datasets. Immune analyses linked the panel to T helper 17 (Th17) cell infiltration and interleukin receptor activity. Single-cell analysis localized these genes to distinct wound-associated cell populations. Diabetic mice exhibited delayed wound closure and greater residual wound widths than control mice. Transcript and protein assays showed concordant changes in representative genes.
CONCLUSIONS: These findings identify a candidate biomarker panel for DU and connect its transcriptomic pattern with immune remodeling, cell-specific expression, and impaired wound repair. Further validation in larger human cohorts is required before clinical application.

Keywords:  diabetic ulcers; experimental validation; immune microenvironment; lipid metabolism; machine learning; single-cell RNA sequencing

DOI:  https://doi.org/10.1007/s11033-026-12862-z
BMC Med Imaging. 2026 Sep 17. pii: 474. [Epub ahead of print]26(1):

TransUNet-GradCAM: a hybrid transformer-U-Net with self-attention and explainable visualizations for foot ulcer segmentation.

Akwasi Asare, Mary Sagoe, Justice Williams Asare, Stephen E Moore.

  Automated segmentation of diabetic foot ulcers (DFUs) supports clinical diagnosis, treatment planning, and longitudinal wound monitoring, but remains difficult owing to the heterogeneous appearance, irregular morphology, and cluttered backgrounds of ulcers in clinical photographs. Convolutional networks such as U-Net localise well but model long-range context poorly, whereas Vision Transformers capture global dependencies. We employ a hybrid ViT-bottleneck U-Net that combines a convolutional encoder-decoder with a Transformer bottleneck and attention-gated skip connections, and we emphasise that the contribution is a rigorously validated and explainable application rather than a new architecture. The model was trained on the public Foot Ulcer Segmentation Challenge (FUSeg) dataset with a hybrid Dice and cross-entropy loss, and all results are reported over five seeds as mean ± 95% confidence interval at a single fixed threshold. On the internal validation set it achieved a Dice of 0.8035 ± 0.0053 and an IoU of 0.7149 ± 0.0073 (HD95 = 19.74 px, ASSD = 6.12 px). A component-wise ablation showed that only the hybrid loss produced a statistically significant change in Dice (- 0.038, p < 0.001); the Transformer bottleneck, attention gates, and augmentation each had small, non-significant in-domain effects. External validation without retraining retained about 92% of internal Dice on the Advancing the Zenith of Healthcare (AZH) Wound Care Center cohort (Dice 0.7460, n = 278), while a small Medetec subset (n = 8) served only as a qualitative check, indicating partial rather than robust generalisation under domain shift. A quantitative explainability analysis found Grad-CAM more wound-localised (energy-in-mask 0.871 versus 0.102) but attention rollout significantly more faithful (p = 0.038, n = 200), showing the two are complementary. Predicted and expert wound areas agreed strongly (Pearson r = 0.944), and the model is lightweight (8.79 M parameters).

Keywords:  Diabetic foot ulcer segmentation; Explainable deep learning (Grad-CAM); TransUNet; U-Net; Vision transformer (ViT); Wound assessment

DOI:  https://doi.org/10.1186/s12880-026-02793-3
Front Endocrinol (Lausanne). 2026 ;17 1977377

From risk classification to clinical action: a prespecified paired pilot benchmark of public large language model interfaces for diabetes-related foot ulcer prevention.

Yang Wen, Liyuan Chen, Xin Deng, Jiaping Lan, Xin Huang, Yang Liu, Lin Chen, Lei Li.

   Background: Large language models (LLMs) may assist in the prevention of diabetes-related foot ulcers; however, their performance in classification may not translate effectively to context-dependent decisions.
Objective: This study aimed to evaluate the accuracy, clinical actionability, reproducibility, and safety of five public LLM web interfaces in the context of the International Working Group on the Diabetic Foot (IWGDF) 2023 risk stratification and preventive management.
Methods: A prespecified, paired, blinded, noninterventional pilot benchmark was conducted using 20 translated, de-identified, curated cases structured in a fixed clinical sequence and evenly distributed across the four IWGDF risk categories. These input conditions represent an idealized, best-case benchmark rather than routine clinical documentation. Each interface assessed all cases under standardized no-search conditions. The primary outcome was exact agreement with a frozen multidisciplinary consensus risk category. Additional outcomes included screening frequency, need for referral, specialty and urgency of referrals, completeness, interreviewer reliability, repeated-generation stability, and safety.
Results: Each interface classified 20/20 cases in concordance with the frozen IWGDF reference (100%; Wilson 95% CI 83.9%-100.0%). The 16.1-percentage-point interval below the observed ceiling indicates limited precision and remains compatible with clinically meaningful error in new cases. Of 100 primary outputs, agreement was observed in 89/100 (89.0%) for referral need, 68/75 (90.7%) for referral specialty, and 92/100 (92.0%) for urgency. Median information-item and mandatory-measure coverage was 100.0% for both measures; however, completeness scoring had limited inter-reviewer reliability and should be interpreted cautiously. In an exploratory, hypothesis-generating four-case repeated-generation substudy, all-three-generation consensus concordance was observed in 16/20 interface-case combinations for referral need, 14/15 eligible combinations for specialty, and 19/20 combinations for urgency. These descriptive counts are not estimates of failure probability or tail behavior. Under the prespecified curated no-search benchmark conditions, no major safety errors were observed among 140 outputs; this absence of observed events does not establish safety in routine clinical use.
Conclusions: Performance was highest for structured guideline mapping, though reliability diminished in referral and individualized management across repeated generations. These findings highlight the necessity for auditable, clinician-supervised decision support instead of autonomous deployment.

Keywords:  artificial intelligence; clinical decision support; diabetic foot; large language models; preventive care; risk stratification

DOI:  https://doi.org/10.3389/fendo.2026.1977377