Diabetic Retinopathy (DR) is one of the leading causes of blindness in the United States and other high-income countries. Early detection is key to prevention, which could be achieved effectively with a fully automated screening tool performing well on clinically relevant measures in primary care settings. We have built an artificial intelligence-based tool on a cloud-based platform for large-scale screening of DR as referable or non-referable. In this paper, we aim to validate this tool built using deep learning based techniques. The cloud-based screening model was developed and tested using deep learning techniques with 88702 images from the Kaggle dataset and externally validated using 1748 high-resolution images of the retina (or fundus images) from the Messidor-2 dataset. For validation in the primary care settings, 264 images were taken prospectively from two diabetes clinics in Queens, New York. The images were uploaded to the cloud-based software for testing the automated system as compared to expert ophthalmologists’ evaluations of referable DR. Measures used were area under the curve (AUC), sensitivity, and specificity of the screening model with respect to professional graders. The screening system achieved a high sensitivity of 99.21% and a specificity of 97.59% on the Kaggle test dataset with an AUC of 0.9992. The system was also externally validated in Messidor-2, where it achieved a sensitivity of 97.63% and a specificity of 99.49% (AUC, 0.9985). On primary care data, the sensitivity was 92.3% overall (12/13 referable images are correctly identified), and overall specificity was 94.8% (233/251 non-referable images). The proposed DR screening tool achieves state-of-the-art performance among the publicly available datasets: Kaggle and Messidor-2 to the best of our knowledge. The performance on various clinically relevant measures demonstrates that the tool is suitable for screening and early diagnosis of DR in primary care settings.

Age-related macular degeneration (AMD) is a leading cause of blindness in the developed world. In this study, we compare the performance of retinal fundus images and genetic-information-based machine learning models for the prediction of late AMD. Using data from the Age-related Eye Disease Study, we built machine learning models with various combinations of genetic, socio-demographic/clinical, and retinal image data to predict late AMD using its severity and category in a single visit, in 2, 5, and 10 years. We compared their performance in sensitivity, specificity, accuracy, and unweighted kappa. The 2-year model based on retinal image and socio-demographic (S-D) parameters achieved a sensitivity of 91.34%, specificity of 84.49% while the same for genetic and S-D-parameters-based model was 79.79% and 66.84%. For the 5-year model, the retinal image and S-D-parameters-based model also outperformed the genetic and S-D parameters-based model. The two 10-year models achieved similar sensitivities of 74.24% and 75.79%, respectively, but the retinal image and S-D-parameters-based model was otherwise superior. The retinal-image-based models were not further improved by adding genetic data. Retinal imaging and S-D data can build an excellent machine learning predictor of developing late AMD over 2–5 years; the retinal imaging model appears to be the preferred prognostic tool for efficient patient management.

In this paper, we propose an automated framework to extract three parameters (the area of newly appeared, disappeared and steady drusen) for tracking the changes in the retinal fundus images over the time. Since the retinal images of different times or visits are not necessarily exactly same, instead they could have different locations, scaling, and rotated but there are common regions (i.e., pathologies). We need to find the changes in these common regions of the images. For finding the common region, we propose an image registration algorithm based on vessel geometric shape. Following the registration of two images, we detect the optic disc based on an adaptive threshold, region growing and circle findings, and drusen based on surrounding intensity ratio. Finally, we extract three parameters for tracking the changes of drusen over time. We have tested our proposed method on 22 retinal red-free fundus images, which are randomly selected from the NAT-2 dataset and manually graded the drusen by an experienced grader. The Pearson Correlation Coefficient of 0.94 is achieved for all these three parameters for tracking the changes of the drusen by our proposed method which is better than the state-of-the-art methods. We envisage that the method will help researchers in finding the correlation between the progression of drusen and many eyes and retinal diseases.

Stroke is a leading cause of death and disability in developed countries. We validated an AI-based prediction model for incident stroke using sensors such as fundus cameras and ophthalmoscopes for retinal images, along with socio-demographic data and traditional risk factors. The model was trained on a proprietary dataset of over 6500 participants, including 171 with 5-year incident strokes and 242 with 10-year incident strokes. The model provides separate 5-year and 10-year risk scores. The model was externally validated on the UK Biobank dataset (3000 subjects with 5-year incident strokes). Using retinal imaging, our models identified individuals with 5-year incident strokes with 80% sensitivity, 82% specificity, and an AUC of 0.83, and predicted 10-year incidents with 72% sensitivity, 78% specificity, and an AUC of 0.79. In comparison, for the 10-year model, the AUC for the Framingham score was 0.73, and the CHADS2 score was 0.74. On the Biobank external dataset, our 5-year model (without retinal features) demonstrated moderate but lower sensitivity (69.3%) and specificity (66.4%) compared to its performance on the proprietary dataset (with retinal features). Using a multi-ethnic dataset, we developed and validated a prediction model that improves stroke risk identification for 5-year and 10-year incidences by incorporating retinal features.

Backgrounds & Objective Glaucomatous vision loss may be preceded by an enlargement of the cup-to-disc ratio (CDR). We propose to develop and validate an artificial intelligence based CDR grading system that may aid in effective glaucoma-suspect screening. Design, Setting & Participants 1546 disc-centered fundus images were selected including all 457 images from the Retinal Image Database for Optic Nerve Evaluation dataset, and images randomly selected from the Age-Related EyeDisease Study, and Singapore Malay Eye Study to develop the system. First, a proprietary semi-automated software was used by an expert grader to quantify vertical CDR. Then, using CDR below 0.5 (not suspect) and CDR above 0.5 (glaucoma-suspect), deep learning architectures were used to train and test a binary classifier system. Measurements The binary classifier, with glaucoma-suspect as positive, is measured using sensitivity, specificity, accuracy, and AUC. Results The system achieved an accuracy of 89.67% (sensitivity, 83.33%; specificity, 93.89%; AUC, 0.93). For external validation, the Retinal Fundus Image database for Glaucoma Analysis dataset, which has 638 gradable quality images, was used. Here the model achieved an accuracy of 83.54% (sensitivity, 80.11%; specificity, 84.96%; AUC, 0.85). Conclusions Having demonstrated an accurate and fully automated glaucoma-suspect screening system that can be deployed on telemedicine platforms, we plan prospective trials to determine the feasibility of the system in primary care settings.

Purpose: Soft drusen and subretinal drusenoid deposits (SDD) aka reticular macular disease (RMD) characterize two pathways to advanced age-related macular degeneration (AMD). We propose these pathways are distinct diseases, with distinct genetic risks, serum risks and associated systemic diseases. Methods: 126 Subjects with AMD had: retinal imaging for RMD status, serum risks, genetic testing, and histories of cardiovascular disease (CVD) and stroke. Results: 62 subjects had RMD, 64 were nonRMD (drusen only), 51 had CVD or Stroke. RMD correlated significantly with: ARMS2 risk allele (p= 0.019); lower mean serum HDL (61±18 vs. 69±22 mg/dl, p= 0.038, t test); CVD and troke (34/51 RMD, p= 0.001).NonRMD correlated/trended with APOE2 (p= 0.032) and CETP (p= 0.072) risk alleles. 97 subjects total had some drusen, which correlated with CFH risk (p= 0.016). Multivariate independent risks for RMD were: CVD and Stroke (p= 0.008), and ARMS2 homozygous risk (p= 0.038). Conclusion: The RMD and soft drusen AMD pathways have distinct systemic associations, serum and genetic risks. RMD is associated with CVD and stroke, ARMS2 risk, and lower HDL; drusen with CFH risk and two lipid risk genes. These pathways appear to be distinct diseases leading to advanced AMD. Summary Statement: Two phenotypes of age-related macular degeneration, soft drusen and reticular macular disease (the combination of subretinal drusenoid deposits and choriocapillaris insufficiency), are shown here to have distinct systemic vascular, serum, and genetic risks. These findings support the concept that these phenotypes actually represent distinct disease processes.