Optimizing Retinal Disease Detection: A Comparative Analysis of State-of-the-Art Deep Learning Models for Multiclass Classification

Retinal conditions such as diabetic retinopathy, glaucoma, and cataract are the most important reasons for vision loss, and early diagnosis is the key to successful treatment. Manual diagnosis takes time and involves expert knowledge, thus posing a requirement for automatic diagnostic systems. Transfer learning in the context of classifying retinal images into four classes: normal, diabetic retinopathy, glaucoma, and cataracts, is investigated in this paper. We employed a number of pre-trained deep learning models, such as EfficientNet, DenseNet121, VGG16 and modified them to classify retinal images. The models were fine-tuned to address the particular task of classifying retinal diseases. A comparative evaluation of the models was carried out to assess their performance on a number of different metrics, including accuracy, precision, and recall. EfficientNet emerged as the best model, with the highest classification accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 2. 2 6 \%}$</tex>. DenseNet121 and VGG16 also performed well, demonstrating the resilience of transfer learning in medical imaging. Both models surpass the conventional methods with more accurate and efficient diagnoses. The results of this research may be used in automated retinal disease detection systems and can be beneficial in remote locations or in primary care centers, enhancing early diagnosis and treatment outcomes.