EfficientNetB3-Based Model for High-Accuracy Retinal Disease Classification

Early identification and diagnosis of eye disorders are crucial for persons who are at a heightened risk, such as those with a familial predisposition to eye problems, elderly individuals, individuals with diabetes, and individuals with a history of eye injuries or surgery. Conditions including diabetic retinopathy, glaucoma, and cataracts may advance without apparent symptoms during their early stages, underscoring the importance of early identification to avoid permanent vision impairment. Conventional diagnostic techniques that utilize retinal fundus imaging are efficient but require a significant amount of effort and depend on the subjective knowledge of medical experts. This study aims to overcome these obstacles by introducing a resilient CNN model that is specifically built to automate the categorization of eye illnesses using retinal fundus pictures and provide healthy lives. The CNN model exhibited outstanding performance, with an overall accuracy of 99.85%. The confusion matrix and performance metrics revealed that the model achieved a perfect classification accuracy of 100% for diabetic retinopathy, while successfully identifying cataracts diseases with over 95% accuracy and glaucoma with 86-89% accuracy.