Graph Adversarial Transfer Learning for Diabetic Retinopathy Classification

Diabetic Retinopathy caused by Diabetes Mellitus is a major vision threatening condition across the global population. Early detection and grading of diabetic retinopathy are pivotal to avoiding the associated vision impairment, and retinal image analysis serves as an effective method for the screening process. Analysing retinal images manually to detect diabetic retinopathy and grade them is a time-consuming pro cess that necessitates the involvement of experts to perform the classification. Computer-aided diagnosis of diabetic retinopathy through retinal image analysis is an effective tool to reduce the time involved in the screening process. This study proposed a convolutional neural network based Selective Feature Map Fusion architecture that leverages the fused feature maps of ResNet50 and EfficientNetV2L networks for the detection and grading of diabetic retinopathy. Feature set of ResNet50 and EfficientNetV2L captures different aspects of the underlying lesion distribution, resulting in a more comprehensive representation of the features. This approach helps the model generalize better to unseen data by incorporating diverse data aspects, thereby preventing overfitting to a specific feature set and developing a model that is more robust to noise. Selective feature fusion helps to reduce the computational overhead of the the feature fusion computation. Comparative analysis against the ResNet50 and EfficientNetV2L models individually revealed the superior performance of the fused model in both detection and grading of diabetic retinopathy. The proposed model attained an excellent level of accuracy of 95.2% in diabetic retinopathy detection with a sensitivity of 95.2% and a specificity of 95.2%, when tested on the IDRiD dataset. The model achieved a classification accuracy of 92% for diabetic retinopathy grading with a sensitivity of 88%, a specificity of 98% and an F1-score of 0.89. Moreover, the fused framework attained superior performance when compared against existing methodologies on the IDRiD dataset. The developed model exhibited robust performance also in the DeepDRiD dataset, and it can be employed successfully in the screening processes for the diagnosis and grading of diabetic retinopathy.

We described a challenge named "DRAC - Diabetic Retinopathy Analysis Challenge" in conjunction with the 25th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2022). Within this challenge, we provided the DRAC datset, an ultra-wide optical coherence tomography angiography (UW-OCTA) dataset (1,103 images), addressing three primary clinical tasks: diabetic retinopathy (DR) lesion segmentation, image quality assessment, and DR grading. The scientific community responded positively to the challenge, with 11, 12, and 13 teams submitting different solutions for these three tasks, respectively. This paper presents a concise summary and analysis of the top-performing solutions and results across all challenge tasks. These solutions could provide practical guidance for developing accurate classification and segmentation models for image quality assessment and DR diagnosis using UW-OCTA images, potentially improving the diagnostic capabilities of healthcare professionals. The dataset has been released to support the development of computer-aided diagnostic systems for DR evaluation.

Diabetic Retinopathy (DR) is an incurable eye disorder, caused due to prolonged diabetes mellitus and it leads to blindness on the go. As an effect, the veins in retina get impaired. Due to the lack of efficient automated DR screening systems, the diagnosis of DR and its grading are now performed by retinal experts. This paper proposes a technique intended for the automatic identification and grading of DR with more exact outcomes in contrast to the existing strategies. To this end, we adopted the ideas of transfer learning and ensemble learning. The pre-trained models used for DR identification are Xception, InceptionV3,and InceptionResnetV2 and that for grading purpose are ResNet-50, DenseNet-201, and DenseNet-169. IDRiD database is used to evaluate the proposed model. The accuracy obtained by the proposed model is 99.5% for DR identification and 99.6% for its grading.

Grader Variability and the Importance of Reference Standards for Evaluating Machine Learning Models for Diabetic Retinopathy

Objective To develop a remote diabetic retinopathy (DR) screening system and to evaluate the effectiveness of the screening system in community. Methods A cross-sectional study was carried out under the informed consent of subjects in Peking University People's Hospital and Beijing Xicheng District Desheng Community Health Service Center from June 2015 to December 2016.A remote DR screening system was established in Peking University People's Hospital and Beijing Xicheng District Desheng Community Health Service Center during June 2015 to December 2016.Based on non-mydriatic digital eye fundus camera photography and the internet transmission technology, anterior ocular segment and fundus images of 2 473 eyes from 1 355 community subjects with type 2 diabetes mellitus were transmitted from Beijing Xicheng District Desheng Community Health Service Center to the reading center of Peking University People's Hospital, and the results were provided to the subjects after analysis, including visual examination, diagnosis and follow-up rate of the subjects, the agreement between remote screening system and conventional screening method was analyzed and compared. Results The visual acuities of the 2 473 eyes of 1 355 subjects were obtained by trained community physician, and the visual acuity was ≤0.05 in 103 eyes (4.2%), >0.05-0.3 in 780 eyes (31.5%), >0.3 in 1 590 eyes (64.3%). A good consistency was found in the diagnosis and grading of DR (Kappa value=0.895) and in diagnosis of macular disorder (Kappa value=0.763) between the remote screening system and conventional screening method.In addition, the diagnosis results of retinal photocoagulation were consistent between the two methods (Kappa value=1.000). The mean duration of the remote screening system for one subject was 10 minutes, which was shorter than 23 minutes of conventional screening method.The follow-up rate of remote screening system was 75.2%. Conclusions There is a high consistency in the DR diagnosis and evaluation between the remote non-mydriatic screening system and conventional screening method.The screening program with follow-up requests has a satisfying follow-up rate, which could meet the demand of DR screening. Key words: Diabetic retinopathy/diagnosis; Mass screening; Health care services accessibility; Telemedicine; Cross-sectional studies

Diabetic retinopathy (DR) is the most common complication of diabetes mellitus and one of the leading causes of preventable blindness in the adult working population. Visual loss can be prevented from the early stages of DR, when the treatments are effective. Therefore, early diagnosis is paramount. However, DR may be clinically asymptomatic until the advanced stage, when vision is already affected and treatment may become difficult. For this reason, diabetic patients should undergo regular eye examinations through screening programs. Traditionally, DR screening programs are run by trained specialists through visual inspection of the retinal images. However, this manual analysis is time consuming and expensive. With the increasing incidence of diabetes and the limited number of clinicians and sanitary resources, the early detection of DR becomes non-viable. For this reason, computed-aided diagnosis (CAD) systems are required to assist specialists for a fast, reliable diagnosis, allowing to reduce the workload and the associated costs. We hypothesize that the application of novel, automatic algorithms for fundus image analysis could contribute to the early diagnosis of DR. Consequently, the main objective of the present Doctoral Thesis is to study, design and develop novel methods based on the automatic analysis of fundus images to aid in the screening, diagnosis, and treatment of DR. In order to achieve the main goal, we built a private database and used five retinal public databases: DRIMDB, DIARETDB1, DRIVE, Messidor and Kaggle. The stages of fundus image processing covered in this Thesis are: retinal image quality assessment (RIQA), the location of the optic disc (OD) and the fovea, the segmentation of RLs and EXs, and the DR severity grading. RIQA was studied with two different approaches. The first approach was based on the combination of novel, global features. Results achieved 91.46% accuracy, 92.04% sensitivity, and 87.92% specificity using the private database. We developed a second approach aimed at RIQA based on deep learning. We achieved 95.29% accuracy with the private database and 99.48% accuracy with the DRIMDB database. The location of the OD and the fovea was performed using a combination of saliency maps. The proposed methods were evaluated over the private database and the public databases DRIVE, DIARETDB1 and Messidor. For the OD, we achieved 100% accuracy for all databases except Messidor (99.50%). As for the fovea location, we also reached 100% accuracy for all databases except Messidor (99.67%). The joint segmentation of RLs and EXs was accomplished by decomposing the fundus image into layers. Results were computed per pixel and per image. Using the private database, 88.34% per-image accuracy (ACCi) was reached for the RL detection and 95.41% ACCi for EX detection. An additional method was proposed for the segmentation of RLs based on superpixels. Evaluating this method with the private database, we obtained 84.45% ACCi. Results were validated using the DIARETDB1 database. Finally, we proposed a deep learning framework for the automatic DR severity grading. The method was based on a novel attention mechanism which performs a separate attention of the dark and the bright structures of the retina. The Kaggle DR detection dataset was used for development and validation. The International Clinical DR Scale was considered, which is made up of 5 DR severity levels. Classification results for all classes achieved 83.70% accuracy and a Quadratic Weighted Kappa of 0.78. The methods proposed in this Doctoral Thesis form a complete, automatic DR screening system, contributing to aid in the early detection of DR. In this way, diabetic patients could receive better attention for their ocular health avoiding vision loss. In addition, the workload of specialists could be relieved while healthcare costs are reduced.

Diabetic retinopathy (DR) is an eye disorder triggered by diabetes that might result in loss of sight. Earlier diagnosis of DR is critical since it might cause loss of sight. Manual diagnoses of DR severity by ophthalmologists are time-consuming and challenging. As a result, there has been considerable attention on designing an automatic technique for DR detection using fundus photographs. In medical science, prognosis and diagnosis are the most challenging tasks due to the presence of fuzziness in medical images and the restricted subjectivity of the experts. Neutrosophic Set (NS) in medical image analysis provides an understanding of the NS concepts, together with knowledge of how to collect, handle, interpret, and analyze clinical images using NS techniques. The neutrosophic set (NS), which is a generality of fuzzy set, provides the overcoming prospect of the restriction of fuzzy-based models for the analysis of medical images. This manuscript develops a Robust Diabetic Retinopathy Detection and Grading using Neutrosophic Topological Vector Space (DRDG-NSTVS) technique on fundus images. The DRDG-NSTVS technique begins with Median Filter (MF) noise removal to optimize the clarity of fundus photographs by successfully eliminating noises. Later, the InceptionV3 is used to perform feature extraction for identifying complicated features and patterns related to DR. The parameter tuning is performed by the moth flame optimization (MFO) technique to ensure superior performance of the model. The final diagnoses and classification of DR are accomplished utilizing the NSTVS classifiers that easily perform the uncertainties inherent in medicinal statistics. The simulation was conducted on a benchmark dataset to examine the proposed model performance. This combined method gives a greatly reliable and accurate solution for the earlier diagnosis and detection of DR

Diabetic retinopathy (DR) is a dangerous eye condition that affects diabetic patients. Without early detection, it can affect the retina and may eventually cause permanent blindness. The early diagnosis of DR is crucial for its treatment. However, the diagnosis of DR is a very difficult process that requires an experienced ophthalmologist. A breakthrough in the field of artificial intelligence called deep learning can help in giving the ophthalmologist a second opinion regarding the classification of the DR by using an autonomous classifier. To accurately train a deep learning model to classify DR, an enormous number of images is required, and this is an important limitation in the DR domain. Transfer learning is a technique that can help in overcoming the scarcity of images. The main idea that is exploited by transfer learning is that a deep learning architecture, previously trained on non-medical images, can be fine-tuned to suit the DR dataset. This paper reviews research papers that focus on DR classification by using transfer learning to present the best existing methods to address this problem. This review can help future researchers to find out existing transfer learning methods to address the DR classification task and to show their differences in terms of performance.

Designing and developing automated systems to detect and grade Diabetic Retinopathy (DR) is one of the recent research areas in the world of medical image applications since it is considered one of the main causes of total blindness for people who have diabetes in the mid-age. In this paper, a complete pipeline for retinal fundus images processing and analysis has been described, implemented and evaluated. This pipeline has three main stages: (i) image pre-processing, (ii) features extraction and (iii) classification. In the first stage, the image has been pre-processed using different transformations to standardize the images and to enhance the images quality. It has been proven that Gaussian filtering is quite effective in this context to enhance the images contrast. In the second and third stage, the convolution neural network (CNN), one of the best neural network architecture for image analysis applications, has been used. The concept of transfer learning and fine tuning have been advocated in this paper and applied for ResNet18 using the publicly available Kaggle dataset. The problem of DR diagnosis has been handled as a multi-class classification problem where there are five levels of the disease severity (0 – No DR, 1 – Mild, 2 – Moderate, 3 – Severe, 4 – Proliferative DR). The final model has achieved accuracy of 70 %, recall of 50% and specificity of 88% outperforming other models built from scratch with less training time and proving the efficiency of transfer learning in this context. The training process has considered the problem of imbalanced dataset using two different ways and it has been discovered that using imbalanced dataset sampler is a very efficient solution. The final model developed in this research could be used as the main unit for a computer aided system to be hosted online for DR detection and diagnosis.

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.

Diabetic retinopathy (DR) is a gradual and severe consequence of diabetes that increases the risk of visual impairment and blindness globally. Early and precise identification of DR is crucial for successful care, but availability to specialist diagnostic tools is restricted, especially in low-resource settings. This study aims to improve the automated identification of diabetic retinopathy using cutting-edge deep learning models—specifically ViT, CLIP, BLIP, ViLT, Florence-2, and ResNet50—which are applied to the APTOS Diabetic Retinopathy dataset. Diabetic Retinopathy (DR) diagnosis is often hindered by poor image quality and subtle differences across varying severity levels. The grading of DR is typically categorized into five stages: No DR, Mild, Moderate, Severe, and Proliferative, as defined by the International Clinical Diabetic Retinopathy Disease Severity Scale. To address artifacts such as uneven illumination, we employ Gaussian preprocessing to smooth noise and Wavelet transforms to enhance edge details. This study evaluates the impact of Gaussian and Wavelet preprocessing techniques on the performance of Vision-Language Models (e.g., CLIP, BLIP) and CNNs (e.g., RESNET50). Results indicate that Gaussian Filtering significantly enhances classification accuracy, yielding a 37% improvement for RESNET50 and 36% for BLIP compared to the baseline. This study demonstrates that appropriate preprocessing is critical for maximizing the potential of deep learning in medical imaging. Specifically, Gaussian Filtering proved superior to Wavelet techniques, offering substantial accuracy gains for models like BLIP and RESNET50. Future work will explore hybrid preprocessing pipelines to further distinguish between the subtle early stages of the disease.

BackgroundWith the diabetes mellitus (DM) prevalence increasing annually, the human grading of retinal images to evaluate DR has posed a substantial burden worldwide. SmartEye is a recently developed fundus image processing and analysis system with lesion quantification function for DR screening. It is sensitive to the lesion area and can automatically identify the lesion position and size. We reported the diabetic retinopathy (DR) grading results of SmartEye versus ophthalmologists in analyzing images captured with non-mydriatic fundus cameras in community healthcare centers, as well as DR lesion quantitative analysis results on different disease stages.MethodsThis is a cross-sectional study. All the fundus images were collected from the Shanghai Diabetic Eye Study in Diabetics (SDES) program from Apr 2016 to Aug 2017. 19,904 fundus images were acquired from 6013 diabetic patients. The grading results of ophthalmologists and SmartEye are compared. Lesion quantification of several images at different DR stages is also presented.ResultsThe sensitivity for diagnosing no DR, mild NPDR (non-proliferative diabetic retinopathy), moderate NPDR, severe NPDR, PDR (proliferative diabetic retinopathy) are 86.19, 83.18, 88.64, 89.59, and 85.02%. The specificity are 63.07, 70.96, 64.16, 70.38, and 74.79%, respectively. The AUC are PDR, 0.80 (0.79, 0.81); severe NPDR, 0.80 (0.79, 0.80); moderate NPDR, 0.77 (0.76, 0.77); and mild NPDR, 0.78 (0.77, 0.79). Lesion quantification results showed that the total hemorrhage area, maximum hemorrhage area, total exudation area, and maximum exudation area increase with DR severity.ConclusionsSmartEye has a high diagnostic accuracy in DR screening program using non-mydriatic fundus cameras. SmartEye quantitative analysis may be an innovative and promising method of DR diagnosis and grading.

Objective To investigate the distribution patterns of diabetic macular edema (DME) based on optical coherence tomography (OCT), and explore its correlation with diabetic retinopathy (DR) stages and systemic factors. Methods A total of 135 patients (242 eyes) with type 2 diabetes were included in this retrospective study. There were 75 males (138 eyes) and 60 females (104 eyes), the ages were from 29 to 83 years, with an average age of (58.8±11.1) years. The general information such as height, weight, smoking history and blood glucose [such as glycosylated hemoglobin (HbA1c)], blood pressure, blood lipid, 24 hours urine protein and other examinations were collected. The diagnosis of DR and DME were made, and the staging of DR and typing of DME were performed based on fundus color imaging and OCT. DR were divided into mild non-proliferative DR (NPDR), moderate NPDR, severe NPDR and proliferative DR (PDR). DME were categorized into 4 types including sponge-like retinal swelling (SME), cystoid macular edema (CME), serous retinal detachment (SRD) and posterior hyaloid traction (PHT). The correlation between DME types and DR staging were analyzed by χ2 test and Fisher exact test. Multivariate logistic regression analysis was used to analyze the correlation between DME types and systemic factors. Results In 242 DR eyes the proportions of mild, moderate, severe NPDR and PDR were 30.99%, 32.64%, 23.14% and 13.23%, respectively. There were 199 eyes (82.23%) with DME. There were statistically significant differences in the proportion of DME in different stages of DR (χ2=21.077, P<0.01). In the 199 eyes with DME, There were 165 eyes (68.18%) of SME, 22 eyes (9.09%) of CME, 7 eyes (2.89%) of SRD and 5 eyes (2.07%) of PHT. The distribution of DME patterns in different stages of DR was statistically significant (χ2=156.273, P<0.01). Logistic regression analysis showed that the duration of diabetes, HbA1c and macroalbuminuria were independent risk factors for DME [odds ratio (OR)= 1.090, 1.510, 4.123; P<0.05], and were also independent for SME (OR=1.092, 1.445, 3.942; P<0.05); HbA1c was an independent risk factor for SRD (OR=2.337, P<0.05). Conclusions There are differences in the distribution of different DME types in each stage of DR. The duration of diabetes, HbA1c and macroalbuminuria were independent risk factors for DME and SME, and macroalbuminuria and HbA1c for CME and SRD. Key words: Macular edema/diagnosis; Risk factors; Diabetic retinopathy/diagnosis; Tomography, optical coherence

Diabetic retinopathy (DR) the prime cause of blindness, develops when glucose levels rise, causing retinal damage. DR can be prevented if the illness is detected early. As a result, early grading, categorization, and diagnosis of DR can help diabetic patients avoid visual loss. Several system methods assist in the classification of DR using high-performance criteria. This work proposes an efficient system-based DR classification. The purpose of efficient machine learning dabetic retinopatyy grading classification (EML-DRGC) design is to recognize DR impulsively with highest accuracy. The proposed technique employs preprocessing methods such as employing the Gaussian filtering approach for removing noise present in retinal fundus images. The segmentation process is followed using K-means segmentation algorithm which is used for segmenting the region of interest (ROI) from background. Moreover, Feature extraction process is done by using gray level co-occurrence matrix (GLCM) in which features are extracted bycapturing the image's visual content and features from acceerated segment test (FAST) design is used as extractor of features. Finally, multi support vector machine is utilized as classifier for detecting severity levels of DR. Performance metrics such as accuracy of 98.38% and specificity of 98.34% are obtained which are superior to existing designs.

To compare the detection rate of diabetic retinopathy (DR) lesions and the agreement of DR severity grading using the ultra-widefield color fundus photography (UWF CFP) combined with high-speed ultra-widefield swept-source optical coherence tomography angiography (UWF SS-OCTA) or fluorescein angiography (FFA). This prospective, observational study recruited diabetic patients who had already taken the FFA examination from November 2021 to June 2022. These patients had either no DR or any stage of DR. All participants were imaged with a 200° UWF CFP and UWF SS-OCTA using a 24 × 20 mm scan model. Images were independently evaluated for the presence or absence of DR lesions including microaneurysms (MAs), intraretinal hemorrhage (IRH), non-perfusion areas (NPAs), intraretinal microvascular abnormalities (IRMAs), venous beading (VB), neovascularization elsewhere (NVE), neovascularization of the optic disc (NVD), and vitreous or preretinal hemorrhage (VH/PRH). Agreement of DR severity grading based on UWF CFP plus UWF SS-OCTA and UWF CFP plus FFA was compared. All statistical analyses were performed using SPSS V.26.0. One hundred and fifty-three eyes of 86 participants were enrolled in the study. The combination of UWF CFP with UWF SS-OCTA showed a similar detection rate compared with UWF CFP plus FFA for all the characteristic DR lesions (p>0.05), except NPAs (p = 0.039). Good agreement was shown for the identification of VB (κ = 0.635), and very good agreement for rest of the DR lesions between the two combination methods (κ-value ranged from 0.858 to 0.974). When comparing the grading of DR severity, very good agreement was achieved between UWF CFP plus UWF SS-OCTA and UWF CFP plusr FFA (κ = 0.869). UWF CFP plus UWF SS-OCTA had a very good agreement in detecting DR lesions and determining the severity of DR compared with UWF CFP plus FFA. This modality has the potential to be used as a fast, reliable, and non-invasive method for DR screening and monitoring in the future.