Glaucoma,a chronic eye disease affecting millions worldwide,poses a substantial threat to eyesight and can result in permanent vision loss if left untreated.Manual identification of glaucoma is a complicated and time-...Glaucoma,a chronic eye disease affecting millions worldwide,poses a substantial threat to eyesight and can result in permanent vision loss if left untreated.Manual identification of glaucoma is a complicated and time-consuming practice requiring specialized expertise and results may be subjective.To address these challenges,this research proposes a computer-aided diagnosis(CAD)approach using Artificial Intelligence(AI)techniques for binary and multiclass classification of glaucoma stages.An ensemble fusion mechanism that combines the outputs of three pre-trained convolutional neural network(ConvNet)models–ResNet-50,VGG-16,and InceptionV3 is utilized in this paper.This fusion technique enhances diagnostic accuracy and robustness by ensemble-averaging the predictions from individual models,leveraging their complementary strengths.The objective of this work is to assess the model’s capability for early-stage glaucoma diagnosis.Classification is performed on a dataset collected from the Harvard Dataverse repository.With the proposed technique,for Normal vs.Advanced glaucoma classification,a validation accuracy of 98.04%and testing accuracy of 98.03%is achieved,with a specificity of 100%which outperforms stateof-the-art methods.For multiclass classification,the suggested ensemble approach achieved a precision and sensitivity of 97%,specificity,and testing accuracy of 98.57%and 96.82%,respectively.The proposed E-GlauNet model has significant potential in assisting ophthalmologists in the screening and fast diagnosis of glaucoma,leading to more reliable,efficient,and timely diagnosis,particularly for early-stage detection and staging of the disease.While the proposed method demonstrates high accuracy and robustness,the study is limited by the evaluation of a single dataset.Future work will focus on external validation across diverse datasets and enhancing interpretability using explainable AI techniques.展开更多
Objective:Diabetic retinopathy(DR)screening using artificial intelligence(AI)has evolved significantly over the past decade.This study aimed to analyze research trends,developments,and patterns in AI-based fundus imag...Objective:Diabetic retinopathy(DR)screening using artificial intelligence(AI)has evolved significantly over the past decade.This study aimed to analyze research trends,developments,and patterns in AI-based fundus image DR screening from 2014 to 2024 through bibliometric analysis.Methods:The study used CiteSpace and Microsoft Excel to analyze 1,172 publications from the Web of Science Core Collection database.The analysis included publication trends over time,citation patterns,institutional collaborations,and the emergence of keywords.Results:From 2014-2022,there was a steady increase in the number of publications,reaching a peak in 2021.India(26%),China(20.05%),and the USA(9.98%)were the major contributors to research output in this field.Among the publication venues,IEEE ACCESS stood out as the leading one,with 44 articles published.The research landscape has evolved from traditional image processing techniques to deep learning approaches.In recent years,there has been a growing emphasis on multimodal AI models.The analysis identified three distinct phases in the development of AI-based DR screening:CNN-based systems(2014-2020),Vision Transformers and innovative learning paradigms(2020-2022),and large foundation models(2022-2024).Conclusions:The field has demonstrated a mature development in traditional AI approaches and is currently in the process of transitioning toward multimodal learning technologies.Future directions suggest an increased focus on the integration of telemedicine,innovative AI algorithms,and real-world implementation of these technologies in real-world settings.展开更多
Artificial Intelligence(AI)is being increasingly used for diagnosing Vision-Threatening Diabetic Retinopathy(VTDR),which is a leading cause of visual impairment and blindness worldwide.However,previous automated VTDR ...Artificial Intelligence(AI)is being increasingly used for diagnosing Vision-Threatening Diabetic Retinopathy(VTDR),which is a leading cause of visual impairment and blindness worldwide.However,previous automated VTDR detection methods have mainly relied on manual feature extraction and classification,leading to errors.This paper proposes a novel VTDR detection and classification model that combines different models through majority voting.Our proposed methodology involves preprocessing,data augmentation,feature extraction,and classification stages.We use a hybrid convolutional neural network-singular value decomposition(CNN-SVD)model for feature extraction and selection and an improved SVM-RBF with a Decision Tree(DT)and K-Nearest Neighbor(KNN)for classification.We tested our model on the IDRiD dataset and achieved an accuracy of 98.06%,a sensitivity of 83.67%,and a specificity of 100%for DR detection and evaluation tests,respectively.Our proposed approach outperforms baseline techniques and provides a more robust and accurate method for VTDR detection.展开更多
Diabetic retinopathy(DR) is one of the most important causes of visual impairment. Automatic recognition of DR lesions, like hard exudates(EXs), in retinal images can contribute to the diagnosis and screening of the d...Diabetic retinopathy(DR) is one of the most important causes of visual impairment. Automatic recognition of DR lesions, like hard exudates(EXs), in retinal images can contribute to the diagnosis and screening of the disease. To achieve this goal, an automatically detecting approach based on improved FCM(IFCM) as well as support vector machines(SVM) was established and studied. Firstly, color fundus images were segmented by IFCM, and candidate regions of EXs were obtained. Then, the SVM classifier is confirmed with the optimal subset of features and judgments of these candidate regions, as a result hard exudates are detected from fundus images. Our database was composed of 126 images with variable color, brightness, and quality. 70 of them were used to train the SVM and the remaining 56 to assess the performance of the method. Using a lesion based criterion, we achieved a mean sensitivity of 94.65% and a mean positive predictive value of 97.25%. With an image-based criterion, our approach reached a 100% mean sensitivity, 96.43% mean specificity and 98.21% mean accuracy. Furthermore, the average time cost in processing an image is 4.56 s. The results suggest that the proposed method can efficiently detect EXs from color fundus images and it could be a diagnostic aid for ophthalmologists in the screening for DR.展开更多
In recent years,there has been a significant increase in the number of people suffering from eye illnesses,which should be treated as soon as possible in order to avoid blindness.Retinal Fundus images are employed for...In recent years,there has been a significant increase in the number of people suffering from eye illnesses,which should be treated as soon as possible in order to avoid blindness.Retinal Fundus images are employed for this purpose,as well as for analysing eye abnormalities and diagnosing eye illnesses.Exudates can be recognised as bright lesions in fundus pictures,which can be thefirst indicator of diabetic retinopathy.With that in mind,the purpose of this work is to create an Integrated Model for Exudate and Diabetic Retinopathy Diagnosis(IM-EDRD)with multi-level classifications.The model uses Support Vector Machine(SVM)-based classification to separate normal and abnormal fundus images at thefirst level.The input pictures for SVM are pre-processed with Green Channel Extraction and the retrieved features are based on Gray Level Co-occurrence Matrix(GLCM).Furthermore,the presence of Exudate and Diabetic Retinopathy(DR)in fundus images is detected using the Adaptive Neuro Fuzzy Inference System(ANFIS)classifier at the second level of classification.Exudate detection,blood vessel extraction,and Optic Disc(OD)detection are all processed to achieve suitable results.Furthermore,the second level processing comprises Morphological Component Analysis(MCA)based image enhancement and object segmentation processes,as well as feature extraction for training the ANFIS classifier,to reliably diagnose DR.Furthermore,thefindings reveal that the proposed model surpasses existing models in terms of accuracy,time efficiency,and precision rate with the lowest possible error rate.展开更多
Cataract is the leading cause of visual impairment globally.The scarcity and uneven distribution of ophthalmologists seriously hinder early visual impairment grading for cataract patients in the clin-ic.In this study,...Cataract is the leading cause of visual impairment globally.The scarcity and uneven distribution of ophthalmologists seriously hinder early visual impairment grading for cataract patients in the clin-ic.In this study,a deep learning-based automated grading system of visual impairment in cataract patients is proposed using a multi-scale efficient channel attention convolutional neural network(MECA_CNN).First,the efficient channel attention mechanism is applied in the MECA_CNN to extract multi-scale features of fundus images,which can effectively focus on lesion-related regions.Then,the asymmetric convolutional modules are embedded in the residual unit to reduce the infor-mation loss of fine-grained features in fundus images.In addition,the asymmetric loss function is applied to address the problem of a higher false-negative rate and weak generalization ability caused by the imbalanced dataset.A total of 7299 fundus images derived from two clinical centers are em-ployed to develop and evaluate the MECA_CNN for identifying mild visual impairment caused by cataract(MVICC),moderate to severe visual impairment caused by cataract(MSVICC),and nor-mal sample.The experimental results demonstrate that the MECA_CNN provides clinically meaning-ful performance for visual impairment grading in the internal test dataset:MVICC(accuracy,sensi-tivity,and specificity;91.3%,89.9%,and 92%),MSVICC(93.2%,78.5%,and 96.7%),and normal sample(98.1%,98.0%,and 98.1%).The comparable performance in the external test dataset is achieved,further verifying the effectiveness and generalizability of the MECA_CNN model.This study provides a deep learning-based practical system for the automated grading of visu-al impairment in cataract patients,facilitating the formulation of treatment strategies in a timely man-ner and improving patients’vision prognosis.展开更多
Use of deep learning algorithms for the investigation and analysis of medical images has emerged as a powerful technique.The increase in retinal dis-eases is alarming as it may lead to permanent blindness if left untr...Use of deep learning algorithms for the investigation and analysis of medical images has emerged as a powerful technique.The increase in retinal dis-eases is alarming as it may lead to permanent blindness if left untreated.Automa-tion of the diagnosis process of retinal diseases not only assists ophthalmologists in correct decision-making but saves time also.Several researchers have worked on automated retinal disease classification but restricted either to hand-crafted fea-ture selection or binary classification.This paper presents a deep learning-based approach for the automated classification of multiple retinal diseases using fundus images.For this research,the data has been collected and combined from three distinct sources.The images are preprocessed for enhancing the details.Six layers of the convolutional neural network(CNN)are used for the automated feature extraction and classification of 20 retinal diseases.It is observed that the results are reliant on the number of classes.For binary classification(healthy vs.unhealthy),up to 100%accuracy has been achieved.When 16 classes are used(treating stages of a disease as a single class),93.3%accuracy,92%sensitivity and 93%specificity have been obtained respectively.For 20 classes(treating stages of the disease as separate classes),the accuracy,sensitivity and specificity have dropped to 92.4%,92%and 92%respectively.展开更多
The objective of the paper is to provide a general view for automatic cup to disc ratio(CDR)assessment in fundus images.As for the cause of blindness,glaucoma ranks as the second in ocular diseases.Vision loss caused ...The objective of the paper is to provide a general view for automatic cup to disc ratio(CDR)assessment in fundus images.As for the cause of blindness,glaucoma ranks as the second in ocular diseases.Vision loss caused by glaucoma cannot be reversed,but the loss may be avoided if screened in the early stage of glaucoma.Thus,early screening of glaucoma is very requisite to preserve vision and maintain quality of life.Optic nerve head(ONH)assessment is a useful and practical technique among current glaucoma screening methods.Vertical CDR as one of the clinical indicators for ONH assessment,has been well-used by clinicians and professionals for the analysis and diagnosis of glaucoma.The key for automatic calculation of vertical CDR in fundus images is the segmentation of optic cup(OC)and optic disc(OD).We take a brief description of methodologies about the OC and disc optic segmentation and comprehensively presented these methods as two aspects:hand-craft feature and deep learning feature.Sliding window regression,super-pixel level,image reconstruction,super-pixel level low-rank representation(LRR),deep learning methodologies for segmentation of OD and OC have been shown.It is hoped that this paper can provide guidance and bring inspiration to other researchers.Every mentioned method has its advantages and limitations.Appropriate method should be selected or explored according to the actual situation.For automatic glaucoma screening,CDR is just the reflection for a small part of the disc,while utilizing comprehensive factors or multimodal images is the promising future direction to furthermore enhance the performance.展开更多
Glaucoma is a prevalent cause of blindness worldwide.If not treated promptly,it can cause vision and quality of life to deteriorate.According to statistics,glaucoma affects approximately 65 million individuals globall...Glaucoma is a prevalent cause of blindness worldwide.If not treated promptly,it can cause vision and quality of life to deteriorate.According to statistics,glaucoma affects approximately 65 million individuals globally.Fundus image segmentation depends on the optic disc(OD)and optic cup(OC).This paper proposes a computational model to segment and classify retinal fundus images for glaucoma detection.Different data augmentation techniques were applied to prevent overfitting while employing several data pre-processing approaches to improve the image quality and achieve high accuracy.The segmentation models are based on an attention U-Net with three separate convolutional neural networks(CNNs)backbones:Inception-v3,visual geometry group 19(VGG19),and residual neural network 50(ResNet50).The classification models also employ a modified version of the above three CNN architectures.Using the RIM-ONE dataset,the attention U-Net with the ResNet50 model as the encoder backbone,achieved the best accuracy of 99.58%in segmenting OD.The Inception-v3 model had the highest accuracy of 98.79%for glaucoma classification among the evaluated segmentation,followed by the modified classification architectures.展开更多
This research focuses on the automatic detection and grading of microaneurysms in fundus images of diabetic retinopathy using artificial intelligence deep learning algorithms.By integrating multi-source fundus image d...This research focuses on the automatic detection and grading of microaneurysms in fundus images of diabetic retinopathy using artificial intelligence deep learning algorithms.By integrating multi-source fundus image data and undergoing a rigorous preprocessing workflow,a hybrid deep learning model architecture combining a modified U-Net and a residual neural network was adopted for the study.The experimental results show that the model achieved an accuracy of[X]%in microaneurysm detection,with a recall rate of[Y]%and a precision rate of[Z]%.In terms of grading diabetic retinopathy,the Cohen’s kappa coefficient for agreement with clinical grading was[K],and there were specific sensitivities and specificities for each grade.Compared with traditional methods,this model has significant advantages in processing speed and result consistency.However,it also has limitations such as insufficient data diversity,difficulties for the algorithm in detecting microaneurysms in severely hemorrhagic images,and high computational costs.The results of this research are of great significance for the early screening and clinical diagnosis decision support of diabetic retinopathy.In the future,it is necessary to further optimize the data and algorithms and promote clinical integration and telemedicine applications.展开更多
The prediction of fundus fluorescein angiography(FFA)images from fundus structural images is a cutting-edge research topic in ophthalmological image processing.Prediction comprises estimating FFA from fundus camera im...The prediction of fundus fluorescein angiography(FFA)images from fundus structural images is a cutting-edge research topic in ophthalmological image processing.Prediction comprises estimating FFA from fundus camera imaging,single-phase FFA from scanning laser ophthalmoscopy(SLO),and three-phase FFA also from SLO.Although many deep learning models are available,a single model can only perform one or two of these prediction tasks.To accomplish three prediction tasks using a unified method,we propose a unified deep learning model for predicting FFA images from fundus structure images using a supervised generative adversarial network.The three prediction tasks are processed as follows:data preparation,network training under FFA supervision,and FFA image prediction from fundus structure images on a test set.By comparing the FFA images predicted by our model,pix2pix,and CycleGAN,we demonstrate the remarkable progress achieved by our proposal.The high performance of our model is validated in terms of the peak signal-to-noise ratio,structural similarity index,and mean squared error.展开更多
A cataract is one of the most significant eye problems worldwide that does not immediately impair vision and progressively worsens over time.Automatic cataract prediction based on various imaging technologies has been...A cataract is one of the most significant eye problems worldwide that does not immediately impair vision and progressively worsens over time.Automatic cataract prediction based on various imaging technologies has been addressed recently,such as smartphone apps used for remote health monitoring and eye treatment.In recent years,advances in diagnosis,prediction,and clinical decision support using Artificial Intelligence(AI)in medicine and ophthalmology have been exponential.Due to privacy concerns,a lack of data makes applying artificial intelligence models in the medical field challenging.To address this issue,a federated learning framework named CDFL based on a VGG16 deep neural network model is proposed in this research.The study collects data from the Ocular Disease Intelligent Recognition(ODIR)database containing 5,000 patient records.The significant features are extracted and normalized using the min-max normalization technique.In the federated learning-based technique,the VGG16 model is trained on the dataset individually after receiving model updates from two clients.Before transferring the attributes to the global model,the suggested method trains the local model.The global model subsequently improves the technique after integrating the new parameters.Every client analyses the results in three rounds to decrease the over-fitting problem.The experimental result shows the effectiveness of the federated learning-based technique on a Deep Neural Network(DNN),reaching a 95.28%accuracy while also providing privacy to the patient’s data.The experiment demonstrated that the suggested federated learning model outperforms other traditional methods,achieving client 1 accuracy of 95.0%and client 2 accuracy of 96.0%.展开更多
AIM:To summarize the application of deep learning in detecting ophthalmic disease with ultrawide-field fundus images and analyze the advantages,limitations,and possible solutions common to all tasks.METHODS:We searche...AIM:To summarize the application of deep learning in detecting ophthalmic disease with ultrawide-field fundus images and analyze the advantages,limitations,and possible solutions common to all tasks.METHODS:We searched three academic databases,including PubMed,Web of Science,and Ovid,with the date of August 2022.We matched and screened according to the target keywords and publication year and retrieved a total of 4358 research papers according to the keywords,of which 23 studies were retrieved on applying deep learning in diagnosing ophthalmic disease with ultrawide-field images.RESULTS:Deep learning in ultrawide-field images can detect various ophthalmic diseases and achieve great performance,including diabetic retinopathy,glaucoma,age-related macular degeneration,retinal vein occlusions,retinal detachment,and other peripheral retinal diseases.Compared to fundus images,the ultrawide-field fundus scanning laser ophthalmoscopy enables the capture of the ocular fundus up to 200°in a single exposure,which can observe more areas of the retina.CONCLUSION:The combination of ultrawide-field fundus images and artificial intelligence will achieve great performance in diagnosing multiple ophthalmic diseases in the future.展开更多
Today, many eye diseases jeopardize our everyday lives, such as Diabetic Retinopathy (DR), Age-related Macular Degeneration (AMD), and Glaucoma.Glaucoma is an incurable and unavoidable eye disease that damages the vis...Today, many eye diseases jeopardize our everyday lives, such as Diabetic Retinopathy (DR), Age-related Macular Degeneration (AMD), and Glaucoma.Glaucoma is an incurable and unavoidable eye disease that damages the vision ofoptic nerves and quality of life. Classification of Glaucoma has been an active fieldof research for the past ten years. Several approaches for Glaucoma classification areestablished, beginning with conventional segmentation methods and feature-extraction to deep-learning techniques such as Convolution Neural Networks (CNN). Incontrast, CNN classifies the input images directly using tuned parameters of convolution and pooling layers by extracting features. But, the volume of training datasetsdetermines the performance of the CNN;the model trained with small datasets,overfit issues arise. CNN has therefore developed with transfer learning. The primary aim of this study is to explore the potential of EfficientNet with transfer learning for the classification of Glaucoma. The performance of the current workcompares with other models, namely VGG16, InceptionV3, and Xception usingpublic datasets such as RIM-ONEV2 & V3, ORIGA, DRISHTI-GS1, HRF, andACRIMA. The dataset has split into training, validation, and testing with the ratioof 70:15:15. The assessment of the test dataset shows that the pre-trained EfficientNetB4 has achieved the highest performance value compared to other models listedabove. The proposed method achieved 99.38% accuracy and also better results forother metrics, such as sensitivity, specificity, precision, F1_score, Kappa score, andArea Under Curve (AUC) compared to other models.展开更多
AIM： To report the surgical result of pars plana vitrectomy（PPV） with air tamponade for rhegmatogenous retinal detachment（RRD） by ultra-widefield fundus imaging system. METHODS： Of 25 consecutive patients（25 e...AIM： To report the surgical result of pars plana vitrectomy（PPV） with air tamponade for rhegmatogenous retinal detachment（RRD） by ultra-widefield fundus imaging system. METHODS： Of 25 consecutive patients（25 eyes） with fresh primary RRD and causative retinal break and vitreous traction were presented. All the patients underwent PPV with air tamponade. Visual acuity（VA） was examined postoperatively and images were captured by ultrawidefield scanning laser ophthalmoscope system（Optos）. RESULTS： Initial reattachment was achieved in 25 cases（100%）. The air volume was 〉60% on the postoperative day（POD） 1. The ultra-widefield images showed that the retina was reattached in all air-filled eyes postoperatively. The retinal break and laser burns in the superior were detected in 22 of 25 eyes（88%）. A missed retinal hole was found under intravitreal air bubble in 1 case（4%）. The air volume was range from 40% to 60% on POD 3. A doublelayered image was seen in 25 of 25 eyes with intravitreal gas. Retinal breaks and laser burns around were seen in the intravitreal air. On POD 7, small bubble without effect was seen in 6 cases（24%） and bubble was completely disappeared in 4 cases（16%）. Small oval bubble in the superior area was observed in 15 cases（60%）. There were no missed and new retinal breaks and no retinal detachment in all cases on the POD 14 and 1 mo and last follow-up. Air disappeared completely on a mean of 9.84 d postoperatively. The mean final postoperative bestcorrected visual acuity（BCVA） was 0.35 log MAR. Mean final postoperative BCVA improved significantly relative to mean preoperative（P〈0.05）. Final VA of 0.3 log MAR or better was seen in 13 eyes. CONCLUSION： PPV with air tamponade is an effective management for fresh RRD with superior retinal breaks. The ultra-widefield fundus imaging can detect postoperative retinal breaks in air-filled eyes. It would be a useful facility for follow-up after PPV with air tamponade. Facedown position and acquired visual rehabilitation may be shorten.展开更多
Adaptive optics scanning laser ophthalmoscopy（AOSLO） has been a promising technique in funds imaging with growing popularity. This review firstly gives a brief history of adaptive optics（AO） and AO-SLO. Then it co...Adaptive optics scanning laser ophthalmoscopy（AOSLO） has been a promising technique in funds imaging with growing popularity. This review firstly gives a brief history of adaptive optics（AO） and AO-SLO. Then it compares AO-SLO with conventional imaging methods（fundus fluorescein angiography, fundus autofluorescence, indocyanine green angiography and optical coherence tomography） and other AO techniques（adaptive optics flood-illumination ophthalmoscopy and adaptive optics optical coherence tomography）. Furthermore, an update of current research situation in AO-SLO is made based on different fundus structures as photoreceptors（cones and rods）, fundus vessels, retinal pigment epithelium layer, retinal nerve fiber layer, ganglion cell layer and lamina cribrosa. Finally, this review indicates possible research directions of AO-SLO in future.展开更多
Glaucoma,a leading cause of blindness,demands early detection for effective management.While AI-based diagnostic systems are gaining traction,their performance is often limited by challenges such as varying image back...Glaucoma,a leading cause of blindness,demands early detection for effective management.While AI-based diagnostic systems are gaining traction,their performance is often limited by challenges such as varying image backgrounds,pixel intensity inconsistencies,and object size variations.To address these limitations,we introduce an innovative,nature-inspired machine learning framework combining feature excitation-based dense segmentation networks(FEDS-Net)and an enhanced gray wolf optimization-supported support vectormachine(IGWO-SVM).This dual-stage approach begins with FEDS-Net,which utilizes a fuzzy integral(FI)technique to accurately segment the optic cup(OC)and optic disk(OD)from retinal images,even in the presence of uncertainty and imprecision.In the second stage,the IGWO-SVM model optimizes the SVM classification process,leveraging a gray wolf-inspired optimization strategy to fine-tune the kernel function for superior accuracy.Extensive testing on three benchmark glaucoma image databases DRIONS-DB,Drishti-GS,and Rim-One-r3 demonstrates the efficacy of our method,achieving classification accuracies of 97.65%,94.88%,and 93.2%,respectively.These results surpass existing state-of-the-art techniques,offering a promising solution for reliable and early glaucoma detection.展开更多
1 Introduction Retinal vessel analysis plays a crucial role in the detection and management of various systemic and ocular diseases,such as diabetic retinopathy,hypertension and cardiovascular disorders[1].Precise seg...1 Introduction Retinal vessel analysis plays a crucial role in the detection and management of various systemic and ocular diseases,such as diabetic retinopathy,hypertension and cardiovascular disorders[1].Precise segmentation of retinal vessels from fundus images enables clinicians to analyze vessel morphology,which can reveal disease progression or underlying conditions.Over recent years,deep learning methods have significantly advanced retinal vessel segmentation.展开更多
Diabetic retinopathy is a serious concern for people dealing with diabetes. Detecting diabetic retinopathy poses significant challenges, requiring skilled professionals, extensive manual image processing, and consider...Diabetic retinopathy is a serious concern for people dealing with diabetes. Detecting diabetic retinopathy poses significant challenges, requiring skilled professionals, extensive manual image processing, and considerable time investment. Fortunately, the integration of deep learning and transfer learning offers invaluable assistance to medical practitioners. This study introduces an ensemble classification framework to detect and grade diabetic retinopathy into 5 classes leveraging the concepts of transfer learning and data fusion. It utilizes three benchmark datasets on diabetic retinopathy: APTOS 2019, IDRiD, and Messidor-2. Initially, these datasets are merged, resulting in a total of 5922 fundus images. Then this fused dataset undergoes pre-processing. Firstly, the images are cropped to remove unwanted regions. Then, Contrast Limited Adaptive Histogram Equalization is applied to improve image quality and fine details. To tackle class imbalance issues, Synthetic Minority Over Sampling technique is employed. Additionally, data augmentation techniques such as flipping, rotation, and zooming are used to increase dataset diversity. The dataset is split into training, validation, and testing sets at a ratio of 70:10:20. For classification, three pre-trained CNN models, EfficientNetB2, DenseNet121, and ResNet50, are fine-tuned. After these models are trained, an ensemble model is constructed by averaging the predictions of each model. Results show that the ensemble model achieved the highest test accuracy of 96.96% in grading diabetic retinopathy into 5 classes outperforming the individual pre-trained models. Furthermore, the ensemble model’s performance is compared with previously published approaches where this model demonstrated superior result.展开更多
基金funded by Department of Robotics and Mechatronics Engineering,Kennesaw State University,Marietta,GA 30060,USA.
文摘Glaucoma,a chronic eye disease affecting millions worldwide,poses a substantial threat to eyesight and can result in permanent vision loss if left untreated.Manual identification of glaucoma is a complicated and time-consuming practice requiring specialized expertise and results may be subjective.To address these challenges,this research proposes a computer-aided diagnosis(CAD)approach using Artificial Intelligence(AI)techniques for binary and multiclass classification of glaucoma stages.An ensemble fusion mechanism that combines the outputs of three pre-trained convolutional neural network(ConvNet)models–ResNet-50,VGG-16,and InceptionV3 is utilized in this paper.This fusion technique enhances diagnostic accuracy and robustness by ensemble-averaging the predictions from individual models,leveraging their complementary strengths.The objective of this work is to assess the model’s capability for early-stage glaucoma diagnosis.Classification is performed on a dataset collected from the Harvard Dataverse repository.With the proposed technique,for Normal vs.Advanced glaucoma classification,a validation accuracy of 98.04%and testing accuracy of 98.03%is achieved,with a specificity of 100%which outperforms stateof-the-art methods.For multiclass classification,the suggested ensemble approach achieved a precision and sensitivity of 97%,specificity,and testing accuracy of 98.57%and 96.82%,respectively.The proposed E-GlauNet model has significant potential in assisting ophthalmologists in the screening and fast diagnosis of glaucoma,leading to more reliable,efficient,and timely diagnosis,particularly for early-stage detection and staging of the disease.While the proposed method demonstrates high accuracy and robustness,the study is limited by the evaluation of a single dataset.Future work will focus on external validation across diverse datasets and enhancing interpretability using explainable AI techniques.
基金supported by the National Natural Science Foundation of China(62402009)the Science and Technology Development Fund of Macao(0013-2024-ITP1).
文摘Objective:Diabetic retinopathy(DR)screening using artificial intelligence(AI)has evolved significantly over the past decade.This study aimed to analyze research trends,developments,and patterns in AI-based fundus image DR screening from 2014 to 2024 through bibliometric analysis.Methods:The study used CiteSpace and Microsoft Excel to analyze 1,172 publications from the Web of Science Core Collection database.The analysis included publication trends over time,citation patterns,institutional collaborations,and the emergence of keywords.Results:From 2014-2022,there was a steady increase in the number of publications,reaching a peak in 2021.India(26%),China(20.05%),and the USA(9.98%)were the major contributors to research output in this field.Among the publication venues,IEEE ACCESS stood out as the leading one,with 44 articles published.The research landscape has evolved from traditional image processing techniques to deep learning approaches.In recent years,there has been a growing emphasis on multimodal AI models.The analysis identified three distinct phases in the development of AI-based DR screening:CNN-based systems(2014-2020),Vision Transformers and innovative learning paradigms(2020-2022),and large foundation models(2022-2024).Conclusions:The field has demonstrated a mature development in traditional AI approaches and is currently in the process of transitioning toward multimodal learning technologies.Future directions suggest an increased focus on the integration of telemedicine,innovative AI algorithms,and real-world implementation of these technologies in real-world settings.
基金This research was funded by the National Natural Science Foundation of China(Nos.71762010,62262019,62162025,61966013,12162012)the Hainan Provincial Natural Science Foundation of China(Nos.823RC488,623RC481,620RC603,621QN241,620RC602,121RC536)+1 种基金the Haikou Science and Technology Plan Project of China(No.2022-016)the Project supported by the Education Department of Hainan Province,No.Hnky2021-23.
文摘Artificial Intelligence(AI)is being increasingly used for diagnosing Vision-Threatening Diabetic Retinopathy(VTDR),which is a leading cause of visual impairment and blindness worldwide.However,previous automated VTDR detection methods have mainly relied on manual feature extraction and classification,leading to errors.This paper proposes a novel VTDR detection and classification model that combines different models through majority voting.Our proposed methodology involves preprocessing,data augmentation,feature extraction,and classification stages.We use a hybrid convolutional neural network-singular value decomposition(CNN-SVD)model for feature extraction and selection and an improved SVM-RBF with a Decision Tree(DT)and K-Nearest Neighbor(KNN)for classification.We tested our model on the IDRiD dataset and achieved an accuracy of 98.06%,a sensitivity of 83.67%,and a specificity of 100%for DR detection and evaluation tests,respectively.Our proposed approach outperforms baseline techniques and provides a more robust and accurate method for VTDR detection.
基金Supported by the National High Technology Research and Development Program of China(863 Program)(No.2006AA020804)Fundamental Research Funds for the Central Universities(No.NJ20120007)+2 种基金Jiangsu Province Science and Technology Support Plan(No.BE2010652)Program Sponsored for Scientific Innovation Research of College Graduate in Jangsu Province(No.CXLX11_0218)Shanghai University Scientific Selection and Cultivation for Outstanding Young Teachers in Special Fund(No.ZZGCD15081)
文摘Diabetic retinopathy(DR) is one of the most important causes of visual impairment. Automatic recognition of DR lesions, like hard exudates(EXs), in retinal images can contribute to the diagnosis and screening of the disease. To achieve this goal, an automatically detecting approach based on improved FCM(IFCM) as well as support vector machines(SVM) was established and studied. Firstly, color fundus images were segmented by IFCM, and candidate regions of EXs were obtained. Then, the SVM classifier is confirmed with the optimal subset of features and judgments of these candidate regions, as a result hard exudates are detected from fundus images. Our database was composed of 126 images with variable color, brightness, and quality. 70 of them were used to train the SVM and the remaining 56 to assess the performance of the method. Using a lesion based criterion, we achieved a mean sensitivity of 94.65% and a mean positive predictive value of 97.25%. With an image-based criterion, our approach reached a 100% mean sensitivity, 96.43% mean specificity and 98.21% mean accuracy. Furthermore, the average time cost in processing an image is 4.56 s. The results suggest that the proposed method can efficiently detect EXs from color fundus images and it could be a diagnostic aid for ophthalmologists in the screening for DR.
文摘In recent years,there has been a significant increase in the number of people suffering from eye illnesses,which should be treated as soon as possible in order to avoid blindness.Retinal Fundus images are employed for this purpose,as well as for analysing eye abnormalities and diagnosing eye illnesses.Exudates can be recognised as bright lesions in fundus pictures,which can be thefirst indicator of diabetic retinopathy.With that in mind,the purpose of this work is to create an Integrated Model for Exudate and Diabetic Retinopathy Diagnosis(IM-EDRD)with multi-level classifications.The model uses Support Vector Machine(SVM)-based classification to separate normal and abnormal fundus images at thefirst level.The input pictures for SVM are pre-processed with Green Channel Extraction and the retrieved features are based on Gray Level Co-occurrence Matrix(GLCM).Furthermore,the presence of Exudate and Diabetic Retinopathy(DR)in fundus images is detected using the Adaptive Neuro Fuzzy Inference System(ANFIS)classifier at the second level of classification.Exudate detection,blood vessel extraction,and Optic Disc(OD)detection are all processed to achieve suitable results.Furthermore,the second level processing comprises Morphological Component Analysis(MCA)based image enhancement and object segmentation processes,as well as feature extraction for training the ANFIS classifier,to reliably diagnose DR.Furthermore,thefindings reveal that the proposed model surpasses existing models in terms of accuracy,time efficiency,and precision rate with the lowest possible error rate.
基金the National Natural Science Foundation of China(No.62276210,82201148,61775180)the Natural Science Basic Research Program of Shaanxi Province(No.2022JM-380)+3 种基金the Shaanxi Province College Students'Innovation and Entrepreneurship Training Program(No.S202311664128X)the Natural Science Foundation of Zhejiang Province(No.LQ22H120002)the Medical Health Science and Technology Project of Zhejiang Province(No.2022RC069,2023KY1140)the Natural Science Foundation of Ningbo(No.2023J390)。
文摘Cataract is the leading cause of visual impairment globally.The scarcity and uneven distribution of ophthalmologists seriously hinder early visual impairment grading for cataract patients in the clin-ic.In this study,a deep learning-based automated grading system of visual impairment in cataract patients is proposed using a multi-scale efficient channel attention convolutional neural network(MECA_CNN).First,the efficient channel attention mechanism is applied in the MECA_CNN to extract multi-scale features of fundus images,which can effectively focus on lesion-related regions.Then,the asymmetric convolutional modules are embedded in the residual unit to reduce the infor-mation loss of fine-grained features in fundus images.In addition,the asymmetric loss function is applied to address the problem of a higher false-negative rate and weak generalization ability caused by the imbalanced dataset.A total of 7299 fundus images derived from two clinical centers are em-ployed to develop and evaluate the MECA_CNN for identifying mild visual impairment caused by cataract(MVICC),moderate to severe visual impairment caused by cataract(MSVICC),and nor-mal sample.The experimental results demonstrate that the MECA_CNN provides clinically meaning-ful performance for visual impairment grading in the internal test dataset:MVICC(accuracy,sensi-tivity,and specificity;91.3%,89.9%,and 92%),MSVICC(93.2%,78.5%,and 96.7%),and normal sample(98.1%,98.0%,and 98.1%).The comparable performance in the external test dataset is achieved,further verifying the effectiveness and generalizability of the MECA_CNN model.This study provides a deep learning-based practical system for the automated grading of visu-al impairment in cataract patients,facilitating the formulation of treatment strategies in a timely man-ner and improving patients’vision prognosis.
文摘Use of deep learning algorithms for the investigation and analysis of medical images has emerged as a powerful technique.The increase in retinal dis-eases is alarming as it may lead to permanent blindness if left untreated.Automa-tion of the diagnosis process of retinal diseases not only assists ophthalmologists in correct decision-making but saves time also.Several researchers have worked on automated retinal disease classification but restricted either to hand-crafted fea-ture selection or binary classification.This paper presents a deep learning-based approach for the automated classification of multiple retinal diseases using fundus images.For this research,the data has been collected and combined from three distinct sources.The images are preprocessed for enhancing the details.Six layers of the convolutional neural network(CNN)are used for the automated feature extraction and classification of 20 retinal diseases.It is observed that the results are reliant on the number of classes.For binary classification(healthy vs.unhealthy),up to 100%accuracy has been achieved.When 16 classes are used(treating stages of a disease as a single class),93.3%accuracy,92%sensitivity and 93%specificity have been obtained respectively.For 20 classes(treating stages of the disease as separate classes),the accuracy,sensitivity and specificity have dropped to 92.4%,92%and 92%respectively.
基金supported by the National Natural Science Foundation of China under Grant No.61772118.
文摘The objective of the paper is to provide a general view for automatic cup to disc ratio(CDR)assessment in fundus images.As for the cause of blindness,glaucoma ranks as the second in ocular diseases.Vision loss caused by glaucoma cannot be reversed,but the loss may be avoided if screened in the early stage of glaucoma.Thus,early screening of glaucoma is very requisite to preserve vision and maintain quality of life.Optic nerve head(ONH)assessment is a useful and practical technique among current glaucoma screening methods.Vertical CDR as one of the clinical indicators for ONH assessment,has been well-used by clinicians and professionals for the analysis and diagnosis of glaucoma.The key for automatic calculation of vertical CDR in fundus images is the segmentation of optic cup(OC)and optic disc(OD).We take a brief description of methodologies about the OC and disc optic segmentation and comprehensively presented these methods as two aspects:hand-craft feature and deep learning feature.Sliding window regression,super-pixel level,image reconstruction,super-pixel level low-rank representation(LRR),deep learning methodologies for segmentation of OD and OC have been shown.It is hoped that this paper can provide guidance and bring inspiration to other researchers.Every mentioned method has its advantages and limitations.Appropriate method should be selected or explored according to the actual situation.For automatic glaucoma screening,CDR is just the reflection for a small part of the disc,while utilizing comprehensive factors or multimodal images is the promising future direction to furthermore enhance the performance.
文摘Glaucoma is a prevalent cause of blindness worldwide.If not treated promptly,it can cause vision and quality of life to deteriorate.According to statistics,glaucoma affects approximately 65 million individuals globally.Fundus image segmentation depends on the optic disc(OD)and optic cup(OC).This paper proposes a computational model to segment and classify retinal fundus images for glaucoma detection.Different data augmentation techniques were applied to prevent overfitting while employing several data pre-processing approaches to improve the image quality and achieve high accuracy.The segmentation models are based on an attention U-Net with three separate convolutional neural networks(CNNs)backbones:Inception-v3,visual geometry group 19(VGG19),and residual neural network 50(ResNet50).The classification models also employ a modified version of the above three CNN architectures.Using the RIM-ONE dataset,the attention U-Net with the ResNet50 model as the encoder backbone,achieved the best accuracy of 99.58%in segmenting OD.The Inception-v3 model had the highest accuracy of 98.79%for glaucoma classification among the evaluated segmentation,followed by the modified classification architectures.
文摘This research focuses on the automatic detection and grading of microaneurysms in fundus images of diabetic retinopathy using artificial intelligence deep learning algorithms.By integrating multi-source fundus image data and undergoing a rigorous preprocessing workflow,a hybrid deep learning model architecture combining a modified U-Net and a residual neural network was adopted for the study.The experimental results show that the model achieved an accuracy of[X]%in microaneurysm detection,with a recall rate of[Y]%and a precision rate of[Z]%.In terms of grading diabetic retinopathy,the Cohen’s kappa coefficient for agreement with clinical grading was[K],and there were specific sensitivities and specificities for each grade.Compared with traditional methods,this model has significant advantages in processing speed and result consistency.However,it also has limitations such as insufficient data diversity,difficulties for the algorithm in detecting microaneurysms in severely hemorrhagic images,and high computational costs.The results of this research are of great significance for the early screening and clinical diagnosis decision support of diabetic retinopathy.In the future,it is necessary to further optimize the data and algorithms and promote clinical integration and telemedicine applications.
基金supported in part by the Gusu Innovation and Entrepreneurship Leading Talents in Suzhou City,grant numbers ZXL2021425 and ZXL2022476Doctor of Innovation and Entrepreneurship Program in Jiangsu Province,grant number JSSCBS20211440+6 种基金Jiangsu Province Key R&D Program,grant number BE2019682Natural Science Foundation of Jiangsu Province,grant number BK20200214National Key R&D Program of China,grant number 2017YFB0403701National Natural Science Foundation of China,grant numbers 61605210,61675226,and 62075235Youth Innovation Promotion Association of Chinese Academy of Sciences,grant number 2019320Frontier Science Research Project of the Chinese Academy of Sciences,grant number QYZDB-SSW-JSC03Strategic Priority Research Program of the Chinese Academy of Sciences,grant number XDB02060000.
文摘The prediction of fundus fluorescein angiography(FFA)images from fundus structural images is a cutting-edge research topic in ophthalmological image processing.Prediction comprises estimating FFA from fundus camera imaging,single-phase FFA from scanning laser ophthalmoscopy(SLO),and three-phase FFA also from SLO.Although many deep learning models are available,a single model can only perform one or two of these prediction tasks.To accomplish three prediction tasks using a unified method,we propose a unified deep learning model for predicting FFA images from fundus structure images using a supervised generative adversarial network.The three prediction tasks are processed as follows:data preparation,network training under FFA supervision,and FFA image prediction from fundus structure images on a test set.By comparing the FFA images predicted by our model,pix2pix,and CycleGAN,we demonstrate the remarkable progress achieved by our proposal.The high performance of our model is validated in terms of the peak signal-to-noise ratio,structural similarity index,and mean squared error.
基金Deputyship for Research&Innovation,Ministry of Education in Saudi Arabia,for funding this research work through Project Number 959.
文摘A cataract is one of the most significant eye problems worldwide that does not immediately impair vision and progressively worsens over time.Automatic cataract prediction based on various imaging technologies has been addressed recently,such as smartphone apps used for remote health monitoring and eye treatment.In recent years,advances in diagnosis,prediction,and clinical decision support using Artificial Intelligence(AI)in medicine and ophthalmology have been exponential.Due to privacy concerns,a lack of data makes applying artificial intelligence models in the medical field challenging.To address this issue,a federated learning framework named CDFL based on a VGG16 deep neural network model is proposed in this research.The study collects data from the Ocular Disease Intelligent Recognition(ODIR)database containing 5,000 patient records.The significant features are extracted and normalized using the min-max normalization technique.In the federated learning-based technique,the VGG16 model is trained on the dataset individually after receiving model updates from two clients.Before transferring the attributes to the global model,the suggested method trains the local model.The global model subsequently improves the technique after integrating the new parameters.Every client analyses the results in three rounds to decrease the over-fitting problem.The experimental result shows the effectiveness of the federated learning-based technique on a Deep Neural Network(DNN),reaching a 95.28%accuracy while also providing privacy to the patient’s data.The experiment demonstrated that the suggested federated learning model outperforms other traditional methods,achieving client 1 accuracy of 95.0%and client 2 accuracy of 96.0%.
基金Supported by 1.3.5 Project for Disciplines of Excellence,West China Hospital,Sichuan University(No.ZYJC21025).
文摘AIM:To summarize the application of deep learning in detecting ophthalmic disease with ultrawide-field fundus images and analyze the advantages,limitations,and possible solutions common to all tasks.METHODS:We searched three academic databases,including PubMed,Web of Science,and Ovid,with the date of August 2022.We matched and screened according to the target keywords and publication year and retrieved a total of 4358 research papers according to the keywords,of which 23 studies were retrieved on applying deep learning in diagnosing ophthalmic disease with ultrawide-field images.RESULTS:Deep learning in ultrawide-field images can detect various ophthalmic diseases and achieve great performance,including diabetic retinopathy,glaucoma,age-related macular degeneration,retinal vein occlusions,retinal detachment,and other peripheral retinal diseases.Compared to fundus images,the ultrawide-field fundus scanning laser ophthalmoscopy enables the capture of the ocular fundus up to 200°in a single exposure,which can observe more areas of the retina.CONCLUSION:The combination of ultrawide-field fundus images and artificial intelligence will achieve great performance in diagnosing multiple ophthalmic diseases in the future.
文摘Today, many eye diseases jeopardize our everyday lives, such as Diabetic Retinopathy (DR), Age-related Macular Degeneration (AMD), and Glaucoma.Glaucoma is an incurable and unavoidable eye disease that damages the vision ofoptic nerves and quality of life. Classification of Glaucoma has been an active fieldof research for the past ten years. Several approaches for Glaucoma classification areestablished, beginning with conventional segmentation methods and feature-extraction to deep-learning techniques such as Convolution Neural Networks (CNN). Incontrast, CNN classifies the input images directly using tuned parameters of convolution and pooling layers by extracting features. But, the volume of training datasetsdetermines the performance of the CNN;the model trained with small datasets,overfit issues arise. CNN has therefore developed with transfer learning. The primary aim of this study is to explore the potential of EfficientNet with transfer learning for the classification of Glaucoma. The performance of the current workcompares with other models, namely VGG16, InceptionV3, and Xception usingpublic datasets such as RIM-ONEV2 & V3, ORIGA, DRISHTI-GS1, HRF, andACRIMA. The dataset has split into training, validation, and testing with the ratioof 70:15:15. The assessment of the test dataset shows that the pre-trained EfficientNetB4 has achieved the highest performance value compared to other models listedabove. The proposed method achieved 99.38% accuracy and also better results forother metrics, such as sensitivity, specificity, precision, F1_score, Kappa score, andArea Under Curve (AUC) compared to other models.
文摘AIM： To report the surgical result of pars plana vitrectomy（PPV） with air tamponade for rhegmatogenous retinal detachment（RRD） by ultra-widefield fundus imaging system. METHODS： Of 25 consecutive patients（25 eyes） with fresh primary RRD and causative retinal break and vitreous traction were presented. All the patients underwent PPV with air tamponade. Visual acuity（VA） was examined postoperatively and images were captured by ultrawidefield scanning laser ophthalmoscope system（Optos）. RESULTS： Initial reattachment was achieved in 25 cases（100%）. The air volume was 〉60% on the postoperative day（POD） 1. The ultra-widefield images showed that the retina was reattached in all air-filled eyes postoperatively. The retinal break and laser burns in the superior were detected in 22 of 25 eyes（88%）. A missed retinal hole was found under intravitreal air bubble in 1 case（4%）. The air volume was range from 40% to 60% on POD 3. A doublelayered image was seen in 25 of 25 eyes with intravitreal gas. Retinal breaks and laser burns around were seen in the intravitreal air. On POD 7, small bubble without effect was seen in 6 cases（24%） and bubble was completely disappeared in 4 cases（16%）. Small oval bubble in the superior area was observed in 15 cases（60%）. There were no missed and new retinal breaks and no retinal detachment in all cases on the POD 14 and 1 mo and last follow-up. Air disappeared completely on a mean of 9.84 d postoperatively. The mean final postoperative bestcorrected visual acuity（BCVA） was 0.35 log MAR. Mean final postoperative BCVA improved significantly relative to mean preoperative（P〈0.05）. Final VA of 0.3 log MAR or better was seen in 13 eyes. CONCLUSION： PPV with air tamponade is an effective management for fresh RRD with superior retinal breaks. The ultra-widefield fundus imaging can detect postoperative retinal breaks in air-filled eyes. It would be a useful facility for follow-up after PPV with air tamponade. Facedown position and acquired visual rehabilitation may be shorten.
基金Supported by National Key Scientific Instrument and Equipment Development Project of China (No.2012YQ12008005)
文摘Adaptive optics scanning laser ophthalmoscopy（AOSLO） has been a promising technique in funds imaging with growing popularity. This review firstly gives a brief history of adaptive optics（AO） and AO-SLO. Then it compares AO-SLO with conventional imaging methods（fundus fluorescein angiography, fundus autofluorescence, indocyanine green angiography and optical coherence tomography） and other AO techniques（adaptive optics flood-illumination ophthalmoscopy and adaptive optics optical coherence tomography）. Furthermore, an update of current research situation in AO-SLO is made based on different fundus structures as photoreceptors（cones and rods）, fundus vessels, retinal pigment epithelium layer, retinal nerve fiber layer, ganglion cell layer and lamina cribrosa. Finally, this review indicates possible research directions of AO-SLO in future.
基金Researchers Supporting Project number(RSP2025R314),King Saud University,Riyadh,Saudi Arabia.
文摘Glaucoma,a leading cause of blindness,demands early detection for effective management.While AI-based diagnostic systems are gaining traction,their performance is often limited by challenges such as varying image backgrounds,pixel intensity inconsistencies,and object size variations.To address these limitations,we introduce an innovative,nature-inspired machine learning framework combining feature excitation-based dense segmentation networks(FEDS-Net)and an enhanced gray wolf optimization-supported support vectormachine(IGWO-SVM).This dual-stage approach begins with FEDS-Net,which utilizes a fuzzy integral(FI)technique to accurately segment the optic cup(OC)and optic disk(OD)from retinal images,even in the presence of uncertainty and imprecision.In the second stage,the IGWO-SVM model optimizes the SVM classification process,leveraging a gray wolf-inspired optimization strategy to fine-tune the kernel function for superior accuracy.Extensive testing on three benchmark glaucoma image databases DRIONS-DB,Drishti-GS,and Rim-One-r3 demonstrates the efficacy of our method,achieving classification accuracies of 97.65%,94.88%,and 93.2%,respectively.These results surpass existing state-of-the-art techniques,offering a promising solution for reliable and early glaucoma detection.
基金supported by the Shuangchuang Ph.D award,Jiangsu,China(No.JSSCBS20210804)the National Natural Science Foundation of China(Grant No.62201460)the Basic Research Programs of Taicang(No.TC2023JC22).
文摘1 Introduction Retinal vessel analysis plays a crucial role in the detection and management of various systemic and ocular diseases,such as diabetic retinopathy,hypertension and cardiovascular disorders[1].Precise segmentation of retinal vessels from fundus images enables clinicians to analyze vessel morphology,which can reveal disease progression or underlying conditions.Over recent years,deep learning methods have significantly advanced retinal vessel segmentation.
文摘Diabetic retinopathy is a serious concern for people dealing with diabetes. Detecting diabetic retinopathy poses significant challenges, requiring skilled professionals, extensive manual image processing, and considerable time investment. Fortunately, the integration of deep learning and transfer learning offers invaluable assistance to medical practitioners. This study introduces an ensemble classification framework to detect and grade diabetic retinopathy into 5 classes leveraging the concepts of transfer learning and data fusion. It utilizes three benchmark datasets on diabetic retinopathy: APTOS 2019, IDRiD, and Messidor-2. Initially, these datasets are merged, resulting in a total of 5922 fundus images. Then this fused dataset undergoes pre-processing. Firstly, the images are cropped to remove unwanted regions. Then, Contrast Limited Adaptive Histogram Equalization is applied to improve image quality and fine details. To tackle class imbalance issues, Synthetic Minority Over Sampling technique is employed. Additionally, data augmentation techniques such as flipping, rotation, and zooming are used to increase dataset diversity. The dataset is split into training, validation, and testing sets at a ratio of 70:10:20. For classification, three pre-trained CNN models, EfficientNetB2, DenseNet121, and ResNet50, are fine-tuned. After these models are trained, an ensemble model is constructed by averaging the predictions of each model. Results show that the ensemble model achieved the highest test accuracy of 96.96% in grading diabetic retinopathy into 5 classes outperforming the individual pre-trained models. Furthermore, the ensemble model’s performance is compared with previously published approaches where this model demonstrated superior result.
