Diabetic Retinopathy Detection Using Deep Learning Techniques

Abstract

DiabeticRetinopathy(DR)diagnosisholdsimmensepotentialfortransformationthrough theapplicationofArtificialIntelligence(AI)inhealthcare. Thisstudyfocusesonutilizing deeplearningtechniquestoaddressthelimitationsofcurrentdiagnosticmethodsforDR. Byanalyzingretinalimages,AIcanaccuratelydetectearlysignsofDR,leadingtotimely interventionsandbetterpatientoutcomes. Thethesisalsoaddressesthechallengeoflowresolution DR images obtained from handheld fundus cameras. To enhance the quality of DR images, a comprehensive approach is adopted, integrating various deep-learning techniques. The Denoising Convolutional Autoencoder (DCAE) is used to reduce noise and artifacts in the images. Advanced models like Contrastive PredictiveCoding(CPC)andVisionTransformers,alongwithpre-processingtechniques, are employed for accurate classification. The comparative analysis shows that attention mechanisms,particularlyintheCPCandTransformersmodels,significantlyimproveDR classification accuracy. Compared to other models, CPC and Transformers outperform in terms of accuracy, sensitivity,andspecificity. Attentionmechanismsenablethesemodelstocapturerelevant features and focus on critical areas within the images. The research findings highlight the potential of AI-based systems, particularly attention-based models, in improving DR diagnosis. Implementing these advancements can lead to better patient outcomes, reduced healthcare costs, and prevention of DR progression to severe stages.