| Summary: | Cataracts are a significant global health concern, particularly among the elderly population. A cataract is an eye condition characterized by the accumulation of fluid in the lens, resulting in cloudiness and blurred vision. Therefore, the development of computer-aided diagnosis (CAD) systems is required to classify cataracts. In this paper, we proposed a convolutional neural network (CNN) model and Visual Geometry Group (VGG 19) for cataract detection in the eyes. Cataract disease detection is framed as a binary classification problem: normal and cataracts. This study explored the generalization performance of the CNN and VGG 19 for cataract classification, considering the challenge of overtraining with a limited number of training samples. This study presented the effectiveness of CNNs and VGG19 models in the context of cataract detection by utilizing a variety of approaches, including image augmentation, finetuning strategies, and various image resolutions. The experimental results demonstrated that the augmented VGG19 model slightly outperformed both the unaugmented VGG19 model (92.72%) and the unaugmented CNN (89.14%) in terms of average accuracy, particularly at a fine-tuning level of -2 with an accuracy of 93.57%. © 2024 IEEE.
|
|---|