Diagnosis Of Eye Diseases Research Articles

Smart Vision Transparency: Efficient Ocular Disease Prediction Model Using Explainable Artificial Intelligence

The early prediction of ocular disease is certainly an obligatory concern in the domain of ophthalmic medicine. Although modern scientific discoveries have shown the potential to treat eye diseases by using artificial intelligence (AI) and machine learning, explainable AI remains a crucial challenge confronting this area of research. Although some traditional methods put in significant effort, they cannot accurately predict the proper ocular diseases. However, incorporating AI into diagnosing eye diseases in healthcare complicates the situation as the decision-making process of AI demonstrates complexity, which is a significant concern, especially in major sectors like ocular disease prediction. The lack of transparency in the AI models may hinder the confidence and trust of the doctors and the patients, as well as their perception of the AI and its abilities. Accordingly, explainable AI is significant in ensuring trust in the technology, enhancing clinical decision-making ability, and deploying ocular disease detection. This research proposed an efficient transfer learning model for eye disease prediction to transform smart vision potential in the healthcare sector and meet conventional approaches’ challenges while integrating explainable artificial intelligence (XAI). The integration of XAI in the proposed model ensures the transparency of the decision-making process through the comprehensive provision of rationale. This proposed model provides promising results with 95.74% accuracy and explains the transformative potential of XAI in advancing ocular healthcare. This significant milestone underscores the effectiveness of the proposed model in accurately determining various types of ocular disease. It is clearly shown that the proposed model is performing better than the previously published methods.

Comparison of tear Matrix Metalloproteinase 9 (MMP-9) estimation with Schirmer's test in Ocular Surface Disorders.

Ocular surface disorder (OSD) is a vexed eye problem and a diagnostic conundrum. Diagnosis has traditionally depended upon symptoms and tests like Schirmer's, TBUT, staining with dyes, and tear meniscus height. Schirmer's test is the most popular. However, the test strips irritate with reflex tearing - producing false high results. Matrix Metalloproteinase 9 (MMP) in the tear is believed to be expressed by stressed epithelial cells of the corneal surface - a key pathology in dry eye disease. This study attempts to compare the results of Schirmer's test and MMP-9 so that the test can individually or severally add to a more definite diagnosis of dry eye disease. 100 eyes of 50 symptomatic patients underwent MMP-9 estimation and were divided into two groups (MMP-9+ve and MMP-9-ve). They were then sub-grouped as per DEWS-2007 based on Schirmer test levels and Ocular Symptomatology Score (OSS). The two groups were compared for severity of dry eye based on Schirmer's test and OSS. Mean Schirmer's value was 12.85 (SD 7.07) for MMP-9+ve and 19.18 (SD 8.94) for MMP-9-ve patients. 80% of patients with severe dry eye and 55.6% of moderate dry eye patients were positive for MMP-9. 85% of the MMP-9 patients had OSS values of 2 or 3. A higher OSDI and positive MMP-9 were shown to be correlated in a statistically remarkable way (p<0.001). The OSDI values of 0-12 for 3/44 (6.8%) positive results, 13-22 for 2/8 (25%) positive results, 23-32 for 4/14 (28.6%) positive results, and 33-100 for 13/35 (37.1%) positive results all showed an increase in MMP-9 positivity along with a rise in the subjective severity of the illness. MMP-9 compares well with Schirmer's values and DED categories based on Schirmer's. The result pointed towards the usefulness of this test in diagnosing patients who may have not yet manifested symptoms.

Evaluation of tear film function by Oculus Keratograph 5M and IDRA ocular surface analyser.

The aim of this study was to evaluate the function of tear film with Oculus Keratograph 5M (Oculus K5M) and IDRA ocular surface analyser (IDRA), analyse their consistency and explore the potential of IDRA in the diagnosis of dry eye disease (DED). This cross-sectional study enrolled 36 participants (DED group, 14 eyes; non-DED group, 22 eyes). The parameters of tear film function, including the first noninvasive breakup time (fNIBUT), average NIBUT (aNIBUT), tear meniscus height (TMH), lipid layer thickness (LLT), lipid layer colour (LLC), lipid layer uniformity (LLU), morphology of meibomian glands (MGs) and MG loss, were obtained with Oculus K5M and IDRA. The consistency of parameter measurements between the two devices was evaluated. All the parameters except LLT, which can be measured only by IDRA, were not significantly different between the two instruments in DED eyes. However, IDRA reported lower values of fNIBUT, aNIBUT and TMH as well as higher MG loss scores in non-DED eyes than Oculus K5M did (p < 0.001, < 0.001, = 0.002, and = 0.002, respectively). Further regression analysis revealed that aNIBUT and LLT measured by IDRA were the optimal parameters for diagnosing DED (OR = 0.567 and 0.845, p = 0.057 and 0.043, respectively), and their combination had the strongest diagnostic potential (AUC = 0.841, sensitivity = 85.7%, and specificity = 77.3%). As a user-friendly noninvasive device, the tear film function parameters measured by IDRA were highly consistent with those measured by Oculus K5M in DED patients. The combination of aNIBUT and LLT measured by IDRA had the best diagnostic accuracy for DED.

Predictors of Discordance Between Dry Eye Symptoms and Signs: Insights From the Sjögren's International Collaborative Clinical Alliance.

The diagnosis and management of dry eye disease (DED) could be complicated by the discordance between DED-related symptoms and signs. We performed a cross-sectional study to investigate the factors of and develop predictive models for the discrepancy in DED symptomatology. We used data from 3455 participants, 21 to 89 years old, from the Sjögren's International Collaborative Clinical Alliance study. We performed a multivariable stepwise linear regression model with backward elimination and Bayesian information criteria to select predictors for the discordance in DES symptomatology, which was defined as the difference between the rank score of Ocular Surface Disease Index 6 (OSDI-6) and the rank score of ocular staining score (OSS). Ten predictors, such as "vitality," "immunomodulating drugs," sensory symptoms," and "ethnicity," remained in the final models, achieving an adjusted R2 (aR2) of 0.35 (95% confidence interval [CI], 0.32-0.39). Specifically, medication use explained 19% (95% CI, 0.17-0.22) of the variance in the outcome, followed by medical history (aR2 = 0.18; 95% CI, 0.15-0.21). Health-related quality of life contributed 16% to the variance in the outcome (95% CI, 0.13-0.19), and, last, demographics contributed 11% (95% CI, 0.09-0.13). Our results suggest that individuals of Asian descent and those using immunomodulating medications often present with severe ocular signs that necessitate regular ophthalmological evaluations, even in the absence of proportionate ocular symptoms. Additionally, ocular symptoms, when accompanied by abnormal sensations in other parts of the body, could indicate systemic conditions that require further investigation and medical care.

Ocular Imaging Challenges, Current State, and a Path to Interoperability: A HIMSS-SIIM Enterprise Imaging Community Whitepaper.

Office-based testing, enhanced by advances in imaging technology, is routinely used in eye care to non-invasively assess ocular structure and function. This type of imaging coupled with autonomous artificial intelligence holds immense opportunity to diagnose eye diseases quickly. Despite the wide availability and use of ocular imaging, there are several factors that hinder optimization of clinical practice and patient care. While some large institutions have developed end-to-end digital workflows that utilize electronic health records, enterprise imaging archives, and dedicated diagnostic viewers, this experience has not yet made its way to smaller and independent eye clinics. Fractured interoperability practices impact patient care in all healthcare domains, including eye care where there is a scarcity of care centers, making collaboration essential among providers, specialists, and primary care who might be treating systemic conditions with profound impact on vision. The purpose of this white paper is to describe the current state of ocular imaging by focusing on the challenges related to interoperability, reporting, and clinical workflow.

OCTA-ReVA: an open-source toolbox for comprehensive retinal vessel feature analysis in optical coherence tomography angiography.

Optical coherence tomography angiography (OCTA) has significantly advanced the study and diagnosis of eye diseases. However, current clinical OCTA systems and software tools lack comprehensive quantitative analysis capabilities, limiting their full clinical utility. This paper introduces the OCTA Retinal Vessel Analyzer (OCTA-ReVA), a versatile open-source platform featuring a user-friendly graphical interface designed for the automated extraction and quantitative analysis of OCTA features. OCTA-ReVA includes traditional established OCTA features based on binary vascular image processing, such as blood vessel density (BVD), foveal avascular zone area (FAZ-A), blood vessel tortuosity (BVT), and blood vessel caliber (BVC). Additionally, it introduces new features based on blood perfusion intensity processing, such as perfusion intensity density (PID), vessel area flux (VAF), and normalized blood flow index (NBFI), which provide deeper insights into retinal perfusion conditions. These additional capabilities are crucial for the early detection and monitoring of retinal diseases. OCTA-ReVA demystifies the intricate task of retinal vasculature quantification, offering a robust tool for researchers and clinicians to objectively evaluate eye diseases and enhance the precision of retinal health assessments.

Detection and diagnosis of diabetic eye diseases using two phase transfer learning approach.

Early diagnosis and treatment of diabetic eye disease (DED) improve prognosis and lessen the possibility of permanent vision loss. Screening of retinal fundus images is a significant process widely employed for diagnosing patients with DED or other eye problems. However, considerable time and effort are required to detect these images manually. Deep learning approaches in machine learning have attained superior performance for the binary classification of healthy and pathological retinal fundus images. In contrast, multi-class retinal eye disease classification is still a difficult task. Therefore, a two-phase transfer learning approach is developed in this research for automated classification and segmentation of multi-class DED pathologies. In the first step, a Modified ResNet-50 model pre-trained on the ImageNet dataset was transferred and learned to classify normal diabetic macular edema (DME), diabetic retinopathy, glaucoma, and cataracts. In the second step, the defective region of multiple eye diseases is segmented using the transfer learning-based DenseUNet model. From the publicly accessible dataset, the suggested model is assessed using several retinal fundus images. Our proposed model for multi-class classification achieves a maximum specificity of 99.73%, a sensitivity of 99.54%, and an accuracy of 99.67%.

The effect of type 2 diabetes mellitus on lid wiper epitheliopathy and ocular surface parameters

IntroductionThis cross-sectional study was conducted to investigate the impact of type 2 diabetes mellitus(DM) and its duration on indicators such as lid wiper epitheliopathy(LWE), and to assess the significance of LWE for early diagnosis of dry eye disease(DED) in DM patients. MethodsA total of 137 subjects with ocular surface disease index(OSDI) score ≥13 were divided into the non-DM group, the short-term DM group (duration <5 years), and the mid-to-long-term DM group(duration ≥5 years). Evaluations were conducted for LWE, OSDI, lipid layer thickness (LLT), partial blinking rate (PBR), fluorescein tear breakup time (FTBUT), corneal fluorescein staining score (CFS), eyelid margin score, and meibomian gland dropout (MGd). ResultsThe upper-LWE score and total LWE score in the mid-to-long-term group were higher than those in the non-DM group (p = 0.008 and p = 0.031, respectively). The lower-LWE scores were more severe than upper-LWE scores in the non-DM and short-term groups (p = 0.001 and p = 0.045, respectively).The confirmed diagnosis rate of DEWSII dry eye with LWE as the primary diagnostic indicator was significantly higher than that which utilize FTBUT<5s as the primary diagnostic indicator(p < 0.05). Compared to the non-DM group, the LLT was thinner and the MGd was more severe in the mid-to long-term group. The upper-LWE score was moderately positively correlated with the MGd, and the lower LWE score was moderately negatively correlated with LLT. ConclusionLWE, LLT, and MGd worsen with the progression of diabetes. Additionally, changes in LWE may precede the FTBUT, indicating that LWE could be considered as an important indicator for early diagnosis of DED in diabetic patients.

Applications of Artificial Intelligence in Diagnosis of Dry Eye Disease: A Systematic Review and Meta-Analysis.

Clinical diagnosis of dry eye disease is based on a subjective Ocular Surface Disease Index questionnaire or various objective tests, however, these diagnostic methods have several limitations. We conducted a comprehensive review of articles discussing various applications of artificial intelligence (AI) models in the diagnosis of the dry eye disease by searching PubMed, Web of Science, Scopus, and Google Scholar databases up to December 2022. We initially extracted 2838 articles, and after removing duplicates and applying inclusion and exclusion criteria based on title and abstract, we selected 47 eligible full-text articles. We ultimately selected 17 articles for the meta-analysis after applying inclusion and exclusion criteria on the full-text articles. We used the Standards for Reporting of Diagnostic Accuracy Studies to evaluate the quality of the methodologies used in the included studies. The performance criteria for measuring the effectiveness of AI models included area under the receiver operating characteristic curve, sensitivity, specificity, and accuracy. We calculated the pooled estimate of accuracy using the random-effects model. The meta-analysis showed that pooled estimate of accuracy was 91.91% (95% confidence interval: 87.46-95.49) for all studies. The mean (±SD) of area under the receiver operating characteristic curve, sensitivity, and specificity were 94.1 (±5.14), 89.58 (±6.13), and 92.62 (±6.61), respectively. This study revealed that AI models are more accurate in diagnosing dry eye disease based on some imaging modalities and suggested that AI models are promising in augmenting dry eye clinics to assist physicians in diagnosis of this ocular surface condition.

Application of whole exome sequencing technology in the diagnosis of hereditary eye diseases

Objective: To evaluate the application of whole exome sequencing (WES) in the diagnosis of hereditary eye diseases. Methods: A total of 24 patients who came to the Obstetrics and Gynecology Hospital of Fudan University for reproductive genetic counseling from December 2020 to December 2023 with the main complaint of congenital eye disorders were included in this study. All cases had no known infections or exposure to known teratogenic drugs, karyotype and chromosome microarray analysis (CMA) abnormalities. Genomic DNA was extracted from the peripheral blood of the probands and their family members and tested for WES. Among them, three individual WES and 21 Trio WES were performed. Potential pathogenic sites were screened and analyzed by Sanger sequencing. For RPGRIP1:c.1611+26G>A site, minigene vector was constructed and RT-qPCR was performed to detect the effect of mRNA splicing. Results: A total of 24 families were collected in this study, of which 20 yielded positive results, achieving a diagnosis rate of 83.3% (20/24). The results involved 21 genes and identified 30 distinct variants, 19 of which were new variants reported. Prenatal diagnostic analysis of family 3 revealed that the fetus carried a c.6970G>T heterozygous nonsense mutation in the PRPF8 gene. The results of RT-PCR with the minigene vector at the non-classical splice site in family 24 indicated that the transcription product of the mutant plasmid was partially retained 104 bp in intron 12, resulting in a p.Glu538Valfs*12 alteration of the protein. Conclusions: The high detection rate of WES in the diagnosis of hereditary eye diseases further supports the advantages of its application as an important molecular detection tool for determining the etiology of hereditary eye diseases.

Multiple Eye Diseases Detection using Convolutional Neural Network

Many eye cases that will arise in the next few years will require early diagnosis for rapid intervention. Polyophthalmia diagnostic methods such as physical examination, examination and diagnosis will be limited to medical and professional methods. Therefore, automated processes are needed. There have been some studies on computer-aided diagnosis (CAD) of polyocular disease using tools such as experts, but these are limited to their knowledge base and therefore not accurate. Early diagnosis of polyophthalmia allows rapid intervention and treatment. This application uses convolutional neural networks for computer-aided diagnosis of various eye diseases and can be used by a common person outside the clinic. This algorithm was trained using transformation learning on a dataset of 100 poliocular disease images from Google image searches for “normal human poliocular disease” and “human poliocular disease.” It leverages the ImageNet model built using a convolutional neural network classifier and transforms its knowledge using transfer learning to train a new model. The new model can classify polyocular disease images into “normal” and “diseases such as bulging polyophthalmia, glaucoma, uveitis, strabismus, dry polyophthalmia, and color blindness.” The system is designed to use deep convolutional neural networks to take images as input and classify the images such as human polyocular disease and “human polyocular disease.” Multiple eye disease detection system using a neural network developed by VS Code. Here we are using python 3.8 as frontend and Mysql Server as backend.

Artificial intelligence in ophthalmology: the present and the future

The medical industry is undergoing an active digital transformation, including the creation of electronic databases, cloud security systems, mobile health monitoring devices, and telemedicine tools. Artificial intelligence (AI), one of the most important technological achievements of the last decade, is gradually gaining momentum in various areas of practical medicine. The cutting edge of AI, neural networks, offers promising approaches to the improvement of clinical examination quality. The review presents data of studies focusing on the use of AI tools in the diagnosis of the most common ophthalmic diseases: diabetic retinopathy, macular degeneration, retinopathy of prematurity, glaucoma, cataracts, and ophthalmic oncology. We discuss both the advantages of neural networks in the diagnosis and monitoring of eye diseases, and outline the difficulties of their implementation, including ethical and legal conflicts.

Retinal Blood Vessels Segmentation With Improved SE‐UNet Model

ABSTRACTAccurate segmentation of retinal vessels is crucial for the early diagnosis and treatment of eye diseases, for example, diabetic retinopathy, glaucoma, and macular degeneration. Due to the intricate structure of retinal vessels, it is essential to extract their features with precision for the semantic segmentation of medical images. In this study, an improved deep learning neural network was developed with a focus on feature extraction based on the U‐Net structure. The enhanced U‐Net combines the architecture of convolutional neural networks (CNNs) with SE blocks (squeeze‐and‐excitation blocks) to adaptively extract image features after each U‐Net encoder's convolution. This approach aids in suppressing nonvascular regions and highlighting features for specific segmentation tasks. The proposed method was trained and tested on the DRIVECHASE_DB1 and STARE datasets. As a result, the proposed model had an algorithmic accuracy, sensitivity, specificity, Dice coefficient (Dc), and Matthews correlation coefficient (MCC) of 95.62/0.9853/0.9652, 0.7751/0.7976/0.7773, 0.9832/0.8567/0.9865, 82.53/87.23/83.42, and 0.7823/0.7987/0.8345, respectively, outperforming previous methods, including UNet++, attention U‐Net, and ResUNet. The experimental results demonstrated that the proposed method improved the retinal vessel segmentation performance.

Measurement of lacrimal gland tissue stiffness for the diagnosis of visual display terminal-associated dry eye disease using shear wave elastography

ObjectiveThis study aimed to evaluate whether lacrimal gland tissue stiffness can aid in diagnosing dry eye disease (DED) using shear wave elastography (SWE). We also aimed to assess the correlation between the subjective symptoms of ocular strain, SWE values, and other ocular examination findings (Schirmer's test and tear film breakup time [TBUT]) contributing to the diagnosis of DED. MethodsThis cross-sectional study recruited 300 participants who were engaged in video display terminal (VDT) work and had been diagnosed with DED by an ophthalmologist for more than one year, and 100 healthy participants without DED symptoms, from August 2020 to December 2021. The modified TFOS DEWS II was used for diagnosing DED. The ocular symptoms, Schirmer's test results, and TBUT of the participants were assessed using standardized methods. Eighty patients with DED and 40 sex- and age-matched healthy participants were selected as the observation and control group, respectively. The lacrimal gland tissue architecture was assessed using ultrasound. ResultThe SWE Emean value reflecting the lacrimal gland tissue stiffness of patients with DED was significantly higher than that of healthy participants (11.06 ± 1.26 for the DED group vs. 7.54 ± 1.27 for the control group, P < 0.001). In patients with DED, the Emean values of the lacrimal glands significantly correlated with the scores for subjective symptoms of ocular dryness (P < 0.001). No significant correlation was observed between the TBUT and Schirmer's test results. The area under the receiver operating characteristic (ROC) curve for SWE was 0.995. According to the ROC analysis results using an Emean cutoff value of 9.32 kPa for the diagnosis of DED, the sensitivity, specificity, positive predictive value（PPV）, and negative predictive （NPV）value were 97.5 %, 95 %, 100 %, and 95 %, respectively. Thus, SWE can be considered for diagnosing VDT-associated DED. ConclusionVDT-associated DED was associated with increased tear gland tissue stiffness. SWE aids in the quantitative detection of lacrimal gland tissue stiffness. Moreover, SWE significantly correlated with the subjective symptoms of DED.

WITHDRAWN: LSAC-Net: A lightweight scale-aware CNN with densely connected focal modulation for retinal blood vessel segmentation

The diagnosis of sight-threatening eye conditions typically involves segmenting retinal structures in fundus images. The segmentation process enables clinicians to identify alterations in retinal morphology, facilitating the detection and diagnosis of various eye diseases. Despite recent developments in image processing, current approaches often fail to partition delicate arteries precisely. Although deep learning has great potential for segmenting medical images, the accuracy of the segmentation process may be limited due to its dependence on repetitive convolution and pooling processes, which may impair the representation of edge information. In this study, we present LSAC-Net, a pixel-level convolutional neural network (CNN) designed specifically for retinal vessel segmentation. Moreover, it is lightweight, with only 37.5 K learnable parameters. LSAC-Net integrates a unique scale-aware feature extraction block (SAFEB) and a densely connected focal modulation block (DCFMB) to improve segmentation accuracy. In addition, careful selection of filter numbers to avoid overlap effectively shortens training times and boosts computational performance, both of which improve overall model efficiency. We performed extensive tests on different aspects of retinal images to evaluate the robustness and generalizability of LSAC-Net. In particular, a thorough evaluation of the model's precision in retinal artery segmentation, a crucial task for ophthalmological diagnosis and treatment, was carried out. We evaluated the performance and efficacy of LSAC-Net with a particular focus on RBV. The model obtains 98.65% specificity, 97.54% accuracy, and 98.88% area under the curve surpassing existing approaches. Experimental results confirm that LSAC-Net is robust, generalizable, and maintains a high level of segmentation accuracy. These features highlight LSAC-Net's potential as a useful instrument for accurate and timely retinal image segmentation in a variety of therapeutic contexts.

Computer vision for eye diseases detection using pre‐trained deep learning techniques and raspberry Pi

AbstractEarly diagnosis of eye diseases is very important to prevent visual impairment and guide appropriate treatment methods. This paper presents a unique approach that can detect numerous eye diseases automatically. Initially, this approach used the pre‐trained ImageNet models that provides various pre‐trained models for training the acquired data. The existing data sets are composed of 645 data images acquired clinically, represented by two groups of subjects as healthy and others holding the proposed eye defect like cataracts, foreign bodies, glaucoma, subconjunctival haemorrhage, and viral conjunctivitis. Followed by comparisons of the pre‐trained model's coefficients and prediction performance. Later, the first‐class execution model is integrated within the Raspberry Pi staging and the real‐time digital camera detection. The evaluation process used the confusion matrix, model accuracy, precision factor, recall coefficient, F1 score, and the Matthews Correlation Coefficient (MCC). Resulting in the performance of these pre‐trained ImageNet models used in this study represented by 93% (InceptionResNetV2), 90% (MobileNet), 86% (Residual Network ResNet50), 85% (InceptionV3), 78% (Visual Geometry Group VGG19), and 72% (Neural Architecture Search Network NASNetMobile). The results show that the InceptionResNetV2 achieved the highest performance. This proposed approach shows its efficiency and strength by early detection of the subject's unhealthy eyes through real‐time monitoring in the field of ophthalmology.

Allvar Gullstrand: The Only Ophthalmologist Who Won the Nobel Prize.

Allvar Gullstrand, the Swedish ophthalmologistand Nobel laureate, was a self-taught mathematician who applied mathematics and higher-order equations to understand the optic system. His inventions, the slit lamp, and the ophthalmoscopeareused in clinical practice for the diagnosis of eye diseases. With his efforts, he explained the accommodation, the process of changing the shape of the lens to focus on near or distant objects.In 1911, he was awarded the Nobel Prize in Physiology or Medicine. In 1913, he was elected as the first president of the Swedish Ophthalmological Society. In 1927, he was awarded the Graefe Medal of the Deutsche Ophthalmologische Gesellschaft.

Ocular Disease Recognition Based on Deep Learning: A Comprehensive Review

This review article represents a major advance in the field of medical imaging and ophthalmology by exploring the critical role of deep learning in the detection and diagnosis of eye diseases. Early and accurate diagnosis becomes essential due to the frequency of ocular disorders that pose a significant risk to vision, including diabetic retinopathy, age-related macular degeneration, glaucoma, and cataracts. The need for more reliable automated solutions is highlighted by the limitations of traditional methods, despite their benefits, which include reliance on small datasets and manual feature analysis. Deep learning, a subset of machine learning, is becoming evident as a powerful tool that can interpret complex medical images and improve diagnostic accuracy without the need to extract human features. This article explores the evolution of deep learning applications in ophthalmology, highlighting the difficulties encountered such as interpretability of models and data quality and the creative solutions that have been found to overcome them. We highlight the revolutionary impact of deep learning in eye disease detection through an in-depth analysis of recent developments, providing insight into potential future research avenues that may further improve patient care in ophthalmology.

Challenges and prospects in the application of artificial intelligence for ocular disease screening and diagnosis

In recent years, artificial intelligence (AI) technologies have experienced substantial growth across various sectors, with significant strides made particularly in medical AI through advancements such as large models. The application of AI within the field of ophthalmology can enhance the accuracy of eye disease screening and diagnosis. However, the deployment of AI and its large models in ophthalmology still encounters numerous limitations and challenges. This article builds upon the transformative achievements in the medical AI sector and discusses the ongoing challenges faced by AI applications in ophthalmology. It provides forward-looking insights from an ophthalmic perspective regarding the era of large models and anticipates research trends in AI applications in ophthalmology, so as to foster the continuous advancement of AI technologies, thereby significantly promoting eye health.

Vessel Segmentation in Fundus Images with Multi-Scale Feature Extraction and Disentangled Representation

Vessel segmentation in fundus images is crucial for diagnosing eye diseases. The rapid development of deep learning has greatly improved segmentation accuracy. However, the scale of the retinal blood-vessel structure varies greatly, and there is a lot of noise unrelated to blood-vessel segmentation in fundus images, which increases the complexity and difficulty of the segmentation algorithm. Comprehensive consideration of factors like scale variation and noise suppression is imperative to enhance segmentation accuracy and stability. Therefore, we propose a retinal vessel segmentation method based on multi-scale feature extraction and decoupled representation. Specifically, we design a multi-scale feature extraction module at the skip connections, utilizing dilated convolutions to capture multi-scale features and further emphasizing crucial information through channel attention modules. Additionally, to separate useful spatial information from redundant information and enhance segmentation performance, we introduce an image reconstruction branch to assist in the segmentation task. The specific approach involves using a disentangled representation method to decouple the image into content and style, utilizing the content part for segmentation tasks. We conducted experiments on the DRIVE, STARE, and CHASE_DB1 datasets, and the results showed that our method outperformed others, achieving the highest accuracy across all three datasets (DRIVE:0.9690, CHASE_DB1:0.9757, and STARE:0.9765).