• Version
• Download 27
• File Size 286.96 KB
• File Count 1
• Create Date 23 May 2025
• Last Updated 23 May 2025

DeepDiabetic: An Identification System of Diabetic Eye Diseases Using Deep Neural Networks

Authors:

KUNJETI VENKATA NANDA KUMAR ¹, K NARESH ²

¹ Post Graduate, Dept. of MCA, Annamacharya Institute of Technology and Sciences (AITS), Karakambadi, Tirupati, Andhra Pradesh, India.

Email:kunjatinanda@gmail.com

² Assistant Professor, Dept. of  MCA, Annamacharya Institute of Technology and Sciences (AITS), Karakambadi, Tirupati, Andhra Pradesh, India. Email:k.naresh1983@gmail.com

ABSTRACT: Deep Learning (DL) has demonstrated significant success and influence in medical imaging, particularly in diagnosis, image detection, and classification. Diabetes remains a major global health concern, and diabetic eye diseases are projected to become the leading cause of vision loss worldwide. In this study, we introduce a multi-class deep learning framework—DeepDiabetic—designed to diagnose and classify four types of diabetic eye diseases: Diabetic Retinopathy (DR), Diabetic Macular Edema (DME), glaucoma, and cataract. The model was evaluated using a dataset of 1,228 images collected from six publicly available sources: DIARETDB0, DIARETDB1, Messidor, HEI-MED, Ocular, and Retina. To enhance model performance, we employed two geometric data augmentation strategies—online and offline augmentation. We investigated the performance of five deep learning architectures: EfficientNetB0, VGG16, ResNet152V2, ResNet152V2 combined with a Gated Recurrent Unit (GRU), and ResNet152V2 combined with a Bidirectional GRU (Bi-GRU). A thorough evaluation and comparison were conducted using public fund us image datasets with four disease classes (DR, DME, Glaucoma, and Cataract). To the best of our knowledge, no existing study in the literature provides a comparative analysis of these specific models for the classification of these four diseases.

Keywords: public, Diabetic, evaluated, performance.