The automated classification of retinal diseases, such as diabetic retinopathy, cataract, glaucoma, and normal retinal conditions, is critical for early diagnosis and timely intervention. This study employs an EfficientNet-based convolutional neural network (CNN) to classify these diseases using a dataset of fundus images, achieving an accuracy of 93.8%. By leveraging transfer learning and fine-tuning techniques, our model is both computationally efficient and highly accurate, making it suitable for clinical applications. In comparison to other CNN-based approaches and ensemble methods, as explored in recent studies, EfficientNet offers a balanced performance in terms of precision, recall, and inference speed, which is crucial in ophthalmic diagnostics. Previous research on real-time object detection in medical imaging, such as Billah et al.’s studies on YOLO models and ensemble methods, has demonstrated the effectiveness of CNNs for accurate classification across various medical imaging tasks. This paper situates EfficientNet within these advancements, underscoring its potential for high-accuracy, multi-class classification in retinal disease detection. Our findings suggest that EfficientNet can play a significant role in automating retinal disease screening, contributing to improved patient outcomes and facilitating the integration of AI in ophthalmology.
Keywords: cataract detection, deep learning in ophthalmology, diabetic retinopathy, efficientNet, fundus images, glaucoma detection, retinal disease classification
