IntroductionTo identify independent risk factors for dry eye disease (DED) and to develop and validate a predictive model for DED among myopic schoolchildren aged 8–16 years in northern China.MethodsA cross-sectional study was conducted among myopic children in Zhangjiakou, Hebei Province. The children underwent comprehensive ocular surface evaluations, including corneal fluorescein staining, tear film break-up time (FBUT), Schirmer I test, lipid layer thickness (LLT), and partial blink rate (PBR). DED was diagnosed using the 2022 Chinese Expert Consensus criteria. Behavioral and environmental risk factors were assessed via validated questionnaires. Logistic regression identified independent factors, and a nomogram was constructed and validated for individualized DED risk estimation.ResultsA total of 1,303 myopic children were included for analysis, and the prevalence of objectively diagnosed DED was 31.2%. Tear film instability, reduced LLT, and increased PBR were the predominant ocular surface abnormalities. The children were divided into training and validation sets according to the community. Among the 912 children in the training set, multivariate analysis identified orthokeratology (Ortho-K) lens use (OR = 4.74), daily screen time ≥ 4 h (OR = 4.21), near work ≥ 4 h (OR = 3.53), BMI ≥ 24 (OR = 3.20), and sleep duration < 6 h (OR = 2.26) as independent risk factors (all p < 0.05). The risk prediction nomogram demonstrated acceptable discriminative ability (AUC: 0.74 in the training set and 0.70 in the validation set).ConclusionDry eye disease is common and under-recognized among myopic children in northern China, with risk closely linked to modifiable behavioral and lifestyle factors and Ortho-K lens use. The developed nomogram can facilitate early identification and targeted interventions for high-risk children.

Deep learning (DL) methods utilize large numbers of images and annotating them is very labor-intensive. In contrast, in clinical practice such large numbers are not needed. Under the DL framework, how to integrate knowledge and data is unknown. We established an ensemble deep learning system (EDLS) that could integrate knowledge and data to automatically detect Glaucomatous Optic Neuropathy (GON), pathologic myopia and diabetic retinopathy using fundus images. An EDLS for the classification of GON was developed using 4225 fundus images obtained from Beijing Tongren Hospital. The generalization of the EDLS was tested on three testing datasets. Two EDLSs for the classification of pathologic myopia and diabetic retinopathy were developed and tested on two datasets obtained from websites respectively. For the classification of GON, the area under the receiver operating characteristic curve (AUC) of EDLS was 0.998 (95

Standardization of clinical reports is crucial for improving the quality of healthcare and facilitating data integration. The lack of unified standards, including format, terminology, and style, is a great challenge in clinical fundus diagnostic reports, which increases the difficulty for large language models (LLMs) to understand the data. To address this, we construct a bilingual standard terminology, containing fundus clinical terms and commonly used descriptions in clinical diagnosis. Then, we establish two models, RetSTA-7B-Zero and RetSTA-7B. RetSTA-7B-Zero, fine-tuned on an augmented dataset simulating clinical scenarios, demonstrates powerful standardization behaviors. However, it encounters a challenge of limitation to cover a wider range of diseases. To further enhance standardization performance, we build RetSTA-7B, which integrates a substantial amount of standardized data generated by RetSTA-7B-Zero along with corresponding English data, covering diverse complex clinical scenarios and achieving report-level standardization for the first time. Experimental results demonstrate that RetSTA-7B outperforms other compared LLMs in bilingual standardization task, which validates its superior performance and generalizability. The checkpoints are available at https://github.com/AB-Story/RetSTA-7B.

Vision loss remains one of the most pervasive and preventable global health burdens, yet ophthalmology has yet to fully benefit from molecular precision medicine. Unlike oncology and cardiometabolic diseases, the early detection and subtyping of eye disorders are hindered by limited access to intraocular tissues and a prevailing belief that the blood-ocular barrier precludes systemic biomarker utility. To systematically evaluate the relevance of the plasma proteome to ocular phenotypes, we profiled 2,920 circulating proteins in 53,016 UK Biobank participants across 80 clinically defined eye diseases and 51 quantitative ocular traits. We integrated association analyses, protein-based prediction models, Mendelian randomization, and unsupervised clustering to uncover predictive, causal, and mechanistic insights. We identified >2,700 significant protein-disease and >3,100 protein-trait associations, revealing widespread links between systemic proteins and intraocular features—particularly in diabetic retinopathy, intraocular pressure, and ISOS-RPE thickness. Plasma proteins such as GDF15, VSIG4, and PLAUR were consistently associated across phenotypes, implicating inflammation, vascular leakage, and complement signaling as convergent mechanisms. Proteome-based models outperformed clinical predictors in multiple conditions (e.g., AUC = 0.913 for diabetic retinopathy) and identified distinct risk gradients. Mendelian randomization supported causal roles for 144 proteins, including therapeutically actionable targets. Hierarchical clustering of 80 ocular phenotypes revealed six proteome-defined ocular modules, linking anatomically diverse traits through shared systemic biology and suggesting a new molecular taxonomy of eye health. This study provides the first comprehensive map of systemic protein signatures across the ocular phenome. By revealing biologically coherent associations across anatomically diverse ocular traits, our findings underscore the relevance of circulating proteins in reflecting both local ocular pathology and broader systemic physiology. These insights support the use of plasma proteomics for early detection, disease subtyping, and therapeutic exploration in ophthalmology, and position the eye as a clinically informative site of systemic biological signaling.

Current retinal foundation models remain constrained by curated research datasets that lack authentic clinical context, and require extensive task-specific optimization for each application, limiting their deployment efficiency in low-resource settings. Here, we show that these barriers can be overcome by building clinical native intelligence directly from real-world medical practice. Our key insight is that large-scale telemedicine programs, where expert centers provide remote consultations across distributed facilities, represent a natural reservoir for learning clinical image interpretation. We present ReVision, a retinal foundation model that learns from the natural alignment between 485,980 color fundus photographs and their corresponding diagnostic reports, accumulated through a decade-long telemedicine program spanning 162 medical institutions across China. Through extensive evaluation across 27 ophthalmic benchmarks, we demonstrate that ReVison enables deployment efficiency with minimal local resources. Without any task-specific training, ReVision achieves zero-shot disease detection with an average AUROC of 0.946 across 12 public benchmarks and 0.952 on 3 independent clinical cohorts. When minimal adaptation is feasible, ReVision matches extensively fine-tuned alternatives while requiring orders of magnitude fewer trainable parameters and labeled examples. The learned representations also transfer effectively to new clinical sites, imaging domains, imaging modalities, and systemic health prediction tasks. In a prospective reader study with 33 ophthalmologists, ReVision's zero-shot assistance improved diagnostic accuracy by 14.8

Background: Minimally invasive glaucoma surgery has become a popular research topic over the past decade. However, no published studies have provided a systematic overview for this field. A bibliometric analysis is urgently required to characterise current international trends and provide an intuitive description of past and emerging trends. Methods: This study analysed minimally invasive glaucoma surgery-related studies by searching the Web of Science for relevant articles published between 1992 and 2023. All the retrieved titles and abstracts were screened for eligibility, and only articles and reviews written in English were included in the analysis. CiteSpace (version 6.1.6), VOSviewer (version 1.6.19), and the bibliometric package in RStudio were used to construct and visualise the results. Results: A total of 1533 publications were included in the analysis with 26072 citations. A total of 4482 authors from 1191 organizations in 57 countries and regions published papers in 139 journals. After 2010, the number of publications increased significantly, with the highest annual productivity occurring in 2022 (n = 229, 15 %). Most of these studies were published in ophthalmology journals. The journal “Ophthalmology” ranked first with 30 papers and 5275 citations. Among the 10 most productive countries, the United States had the largest share of publications (n = 423, 36 %) and Switzerland had the highest proportion of multiple-country publications (70 %). Neodymium was the first keyword discovered, appearing in 1992 and continuing for 21 years. Kahook dual-blade, progression, gonioscopy-assisted transluminal trabeculotomy, efficacy, minimally invasive glaucoma surgery, cataract extraction, and primary open-angle glaucoma were the most recent keywords since 2020. Conclusions: This was the first bibliometric analysis of minimally invasive glaucoma surgery and provides an overview of the developments in this field. Our results identified outstanding studies, countries, institutions, journals, and authors in the field to point the way forward for scientific research and clinical applications of minimally invasive glaucoma surgery.

Subtle semantic differences in retinal image and text data present great challenges for pre-training visual-language models. Moreover, false negative samples, i.e., image-text pairs having the same semantics but incorrectly regarded as negatives, disrupt the visual-language pre-training process and affect the model's learning ability. This work aims to develop a retinal foundation model, called ViLReF, by pre-training on a paired dataset comprising 451,956 retinal images and corresponding diagnostic text reports. In our vision-language pre-training strategy, we leverage expert knowledge to facilitate the extraction of labels and propose a novel constraint, the Weighted Similarity Coupling Loss, to adjust the speed of pushing sample pairs further apart dynamically within the feature space. Furthermore, we employ a batch expansion module with dynamic memory queues, maintained by momentum encoders, to supply extra samples and compensate for the vacancies caused by eliminating false negatives. Extensive experiments are conducted on multiple datasets for downstream classification and segmentation tasks. The experimental results demonstrate the powerful zero-shot and transfer learning capabilities of ViLReF, verifying the effectiveness of our pre-training strategy. Our ViLReF model is available at: https://github.com/T6Yang/ViLReF.

Purpose:This study aimed to propose a new deep learning (DL) approach to automatically predict the retinal nerve fiber layer thickness (RNFLT) around optic disc regions in fundus photography trained by optical coherence tomography (OCT) and diagnose glaucoma based on the predicted comprehensive information about RNFLT. Methods:A total of 1403 pairs of fundus photographs and OCT RNFLT scans from 1403 eyes of 1196 participants were included. A residual deep neural network was trained to predict the RNFLT for each local image in a fundus photograph, and then a RNFLT report was generated based on the local images. Two indicators were designed based on the generated report. The support vector machines (SVM) algorithm was used to diagnose glaucoma based on the two indicators. Results:A strong correlation was found between the predicted and actual RNFLT values on local images. On three testing datasets, we found the Pearson r to be 0.893, 0.850, and 0.831, respectively, and the mean absolute error of the prediction to be 14.345, 17.780, and 19.250 μm, respectively. The area under the receiver operating characteristic curves for discriminating glaucomatous from healthy eyes was 0.860 (95 % confidence interval, 0.799-0.921). Conclusions:We established a novel local image-based DL approach to provide comprehensive quantitative information on RNFLT in fundus photographs, which was used to diagnose glaucoma. In addition, training a deep neural network based on local images to predict objective detail information in fundus photographs provided a new paradigm for the diagnosis of ophthalmic diseases.

Anomaly detection is an important yet challenging task in medical image analysis. Most anomaly detection methods are based on reconstruction, but the performance of reconstruction -based methods is limited due to over -reliance on pixel -level losses. To address the limitation, we propose a patch -wise contrastive learningbased auto -encoder for medical anomaly detection. The key contribution is the patch -wise contrastive learning loss that provides supervision on local semantics to enforce semantic consistency between corresponding input- output patches. Contrastive learning pulls corresponding patch pairs closer while pushing non -corresponding ones apart between input and output, enabling the model to learn local normal features better and improve discriminability on anomalous regions. Additionally, we design an anomaly score based on local semantic discrepancies to pinpoint abnormalities by comparing feature difference rather than pixel variations. Extensive experiments on three public datasets (i.e., brain MRI, retinal OCT, and chest X-ray) achieve state-ofthe-art performance, with our method achieving over 99% AUC on retinal and brain images. Both the contrastive patch -wise supervision and patch -discrepancy score provide targeted advancements to overcome the weaknesses in existing approaches.

OBJECTIVE:To compare the efficacy and safety of ab interno canaloplasty (ABiC) with gonioscopy-assisted transluminal trabeculotomy (GATT) in patients with open-angle glaucoma (OAG). METHOD:This randomised clinical trial recruited eyes with OAG and no previous incisional ocular surgery, among which 38 were randomised to ABiC and 39 to GATT. Follow-ups were performed at 1, 3, 6 and 12 months postoperatively. The primary outcome measures were intraocular pressure (IOP) and use of glaucoma medication at 12 months postoperatively. The secondary outcome measure was complete surgical success (not requiring glaucoma surgery, IOP ≤21 mm Hg and non-use of glaucoma medications). RESULTS:Both groups had similar demographic and ocular characteristics. A total of 71 of the 77 subjects (92.2%) completed 12-month follow-up. At 12 months, mean IOP was 19.0±5.2 mm Hg in the ABiC group and 16.0±3.1 mm Hg in the GATT group (p=0.003). Overall, 57.2% of ABiC patients and 77.8% of GATT patients were medication free (p=0.06). The number of glaucoma medications was 0.9±1.3 in the ABiC group and 0.6±1.2 in the GATT group (p=0.27). The 12-month cumulative rate of complete surgical success was 56% in the ABiC group and 75% in the GATT group (p=0.09). Three eyes in the ABiC group and one eye in the GATT group required additional glaucoma surgery. Hyphema (87% vs 47%) and supraciliary effusion (92% vs 71%) were noted more often in the GATT group than in the ABiC group. CONCLUSIONS:The preliminary result showed that GATT had an advantage over ABiC in IOP reduction for OAG patients, accompanied by favourable safety at 12-month postoperatively. TRIAL REGISTRATION NUMBER:ChiCTR1800016933.

The Vision-Language Foundation model is increasingly investigated in the fields of computer vision and natural language processing, yet its exploration in ophthalmology and broader medical applications remains limited. The challenge is the lack of labeled data for the training of foundation model. To handle this issue, a CLIP-style retinal image foundation model is developed in this paper. Our foundation model, RET-CLIP, is specifically trained on a dataset of 193,865 patients to extract general features of color fundus photographs (CFPs), employing a tripartite optimization strategy to focus on left eye, right eye, and patient level to reflect real-world clinical scenarios. Extensive experiments demonstrate that RET-CLIP outperforms existing benchmarks across eight diverse datasets spanning four critical diagnostic categories: diabetic retinopathy, glaucoma, multiple disease diagnosis, and multi-label classification of multiple diseases, which demonstrate the performance and generality of our foundation model. The sourse code and pre-trained model are available at https://github.com/sStonemason/RET-CLIP.

Clinical lesions progress continuously but previous grading strategies are not fine-grained enough to model the continuously changing features of lesions. For lack of temporal sequential medical data to provide lesion progression information, we propose to use the severity ranking of disease lesions as spatial ranking label to represent temporal progression. Absolute ranking labels and relative ranking labels are calculated from severity ranking of datasets. A two-branch framework with spatial-temporal feature encoder is designed which using ranking labels to exploit the ranking relation between query and reference images. Furthermore, ranking loss is designed to enforce that sample features should be distributed in the feature space based on ranking scores. Our model achieves five-grade accuracy of 0.9204 on myopic maculopathy dataset. Compared with discrete grading, great improvement for automatic diagnosis is achieved. Experiments on B-mode fatty liver ultrasound dataset and glaucoma dataset also show generality of our algorithm.

Anomaly detection is an important task for medical image analysis, which can alleviate the reliance of supervised methods on large labelled datasets. Most existing methods use a pixel-wise self-reconstruction framework for anomaly detection. However, there are two challenges of these studies: 1) they tend to overfit learning an identity mapping between the input and output, which leads to failure in detecting abnormal samples; 2) the reconstruction considers the pixel-wise differences which may lead to an undesirable result. To mitigate the above problems, we propose a novel heterogeneous Auto-Encoder (Hetero-AE) for medical anomaly detection. Our model utilizes a convolutional neural network (CNN) as the encoder and a hybrid CNN-Transformer network as the decoder. The heterogeneous structure enables the model to learn the intrinsic information of normal data and enlarge the difference on abnormal samples. To fully exploit the effectiveness of Transformer in the hybrid network, a multi-scale sparse Transformer block is proposed to trade off modelling long-range feature dependencies and high computational costs. Moreover, the multi-stage feature comparison is introduced to reduce the noise of pixel-wise comparison. Extensive experiments on four public datasets (i.e., retinal OCT, chest X-ray, brain MRI, and COVID-19 ) verify the effectiveness of our method on different imaging modalities for anomaly detection. Additionally, our method can accurately detect tumors in brain MRI and lesions in retinal OCT with interpretable heatmaps to locate lesion areas, assisting clinicians in diagnosing abnormalities efficiently.

Purpose: To investigate the potential phases in myopic retinal vascular alterations for further elucidating the mechanisms underlying the progression of high myopia (HM). Methods: For this retrospective study, participants diagnosed with high myopia at Beijing Tongren Hospital were recruited. Based on bionic mechanisms of human vision, an intelligent image processing model was developed and utilized to extract and quantify the morphological characteristics of retinal vasculatures in different regions measured by papilla-diameter (PD), including vascular caliber, arteriole-to-venule ratio (AVR), tortuosity, the angle of the vascular arch (AVA), the distance of the vascular arch (DVA), density, fractal dimension, and venular length. In addition, the optic disc and the area of peripapillary atrophy (PPA) were also quantified. The characteristics of the overall population, as well as patients aged less than 25 years old, were compared by different genders. Univariate and multiple linear regression analyses were conducted to investigate the correlation of retinal vasculature parameters with PPA width, and detailed trends of the vascular indicators were analyzed to explore the potential existence of staged morphological changes. Findings: The study included 14,066 fundus photographs of 5775 patients (aged 41.2 +/- 18.6 years), of whom 7379 (61.2 %) were female. The study included 12,067 fundus photographs of 5320 patients (aged 41.2 +/- 18.6 years). Significant variations in the morphological parameters of retinal vessels were observed between males and females. After adjusting for age and sex, multiple linear regression analysis showed that an increased PPA width ratio was associated with lower AVA (1PD), DVA (1PD), vascular caliber (0.5-1.0 PD), tortuosity (0.5-1.0 PD), density and fractal dimension (all P < 0.001, Spearman's rho < 0). Overall, the changes in retinal vascular morphology showed two phases: tortuosity (0.5-1.0PD) and AVA (1PD) decreased rapidly in the first stage but significantly more slowly in the second stage, while vascular density and fractal dimension showed a completely opposite trend with an initial slow decline followed by a rapid decrease. Conclusions: This study identified two distinct phases of retinal vascular morphological changes during the progression of HM. Traction lesions were predominant in the initial stage, while atrophic lesions were predominant in the later stage. These findings provide further insight into the development mechanism of HM from the perspective of retinal vasculature.

BACKGROUND:More than 90% of vision impairment is avoidable. However, in China, a routine screening programme is currently unavailable in primary health care. With the dearth of economic evidence on screening programmes for multiple blindness-causing eye diseases, delivery options, and screening frequencies, we aimed to evaluate the costs and benefits of a population-based screening programme for multiple eye diseases in China. METHODS:We developed a decision-analytic Markov model for a cohort of individuals aged 50 years and older with a total of 30 1-year cycles. We calculated the cost-effectiveness and cost-utility of screening programmes for multiple major blindness-causing eye diseases in China, including age-related macular degeneration, glaucoma, diabetic retinopathy, cataracts, and pathological myopia, from a societal perspective (including direct and indirect costs). We analysed rural and urban settings separately by different screening delivery options (non-telemedicine [ie, face-to-face] screening, artificial intelligence [AI] telemedicine screening, and non-AI telemedicine screening) and frequencies. We calculated incremental cost-utility ratios (ICURs) using quality-adjusted life-years and incremental cost-effectiveness ratios (ICERs) in terms of the cost per blindness year avoided. One-way deterministic and simulated probabilistic sensitivity analyses were used to assess the robustness of the main outcomes. FINDINGS:Compared with no screening, non-telemedicine combined screening of multiple eye diseases satisfied the criterion for a highly cost-effective health intervention, with an ICUR of US$2494 (95% CI 1130 to 2716) and an ICER of $12 487 (8773 to 18 791) in rural settings. In urban areas, the ICUR was $624 (395 to 907), and the ICER was $7251 (4238 to 13 501). Non-AI telemedicine screening could result in fewer costs and greater gains in health benefits (ICUR $2326 [1064 to 2538] and ICER $11 766 [8200 to 18 000] in rural settings; ICUR $581 [368 to 864] and ICER $6920 [3926 to 13 231] in urban settings). AI telemedicine screening dominated no screening in rural settings, and in urban settings the ICUR was $244 (-315 to 1073) and the ICER was $2567 (-4111 to 15 389). Sensitivity analyses showed all results to be robust. By further comparison, annual AI telemedicine screening was the most cost-effective strategy in both rural and urban areas. INTERPRETATION:Combined screening of multiple eye diseases is cost-effective in both rural and urban China. AI coupled with teleophthalmology presents an opportunity to promote equity in eye health. FUNDING:National Natural Science Foundation of China.

Fundus images are widely used in the screening and diagnosis of eye diseases. Current classification algorithms for computer-aided diagnosis in fundus images rely on large amounts of data with reliable labels. However, the appearance of noisy labels degrades the performance of data-dependent algorithms, such as supervised deep learning. A noisy label learning framework suitable for the multiclass classification of fundus diseases is presented in this paper, which combines data cleansing (DC), adaptive negative learning (ANL), and sharpness-aware minimization (SAM) modules. Firstly, the DC module filters the noisy labels in the training dataset based on the prediction confidence. Then, the ANL module modifies the loss function by choosing complementary labels, which are neither the given labels nor the labels with the highest confidence. Moreover, for better generalization, the SAM module is applied by simultaneously optimizing the loss and its sharpness. Extensive experiments on both private and public datasets show that our method greatly promotes the performance for classification of multiple fundus diseases with noisy labels.

We present VisionFM, a foundation model pre-trained with 3.4 million ophthalmic images from 560,457 individuals, covering a broad range of ophthalmic diseases, modalities, imaging devices, and demography. After pre-training, VisionFM provides a foundation to foster multiple ophthalmic artificial intelligence (AI) applications, such as disease screening and diagnosis, disease prognosis, subclassification of disease phenotype, and systemic biomarker and disease prediction, with each application enhanced with expert-level intelligence and accuracy. The generalist intelligence of VisionFM outperformed ophthalmologists with basic and intermediate levels in jointly diagnosing 12 common ophthalmic diseases. Evaluated on a new large-scale ophthalmic disease diagnosis benchmark database, as well as a new large-scale segmentation and detection benchmark database, VisionFM outperformed strong baseline deep neural networks. The ophthalmic image representations learned by VisionFM exhibited noteworthy explainability, and demonstrated strong generalizability to new ophthalmic modalities, disease spectrum, and imaging devices. As a foundation model, VisionFM has a large capacity to learn from diverse ophthalmic imaging data and disparate datasets. To be commensurate with this capacity, in addition to the real data used for pre-training, we also generated and leveraged synthetic ophthalmic imaging data. Experimental results revealed that synthetic data that passed visual Turing tests, can also enhance the representation learning capability of VisionFM, leading to substantial performance gains on downstream ophthalmic AI tasks. Beyond the ophthalmic AI applications developed, validated, and demonstrated in this work, substantial further applications can be achieved in an efficient and cost-effective manner using VisionFM as the foundation.