Automated Cataract Detection Using Deep Learning and Pre-Trained CNN Models

Abstract

Cataracts are a leading cause of vision impairment worldwide, and early detection is crucial for preventing severe vision loss. In this paper, we propose an automated cataract detection system utilizing deep learning models, specifically pre-trained Convolutional Neural Networks (CNNs) including Mobile Net, VGG-16, VGG-19, ResNet-50, Inception-v3, and DenseNet-121. The system is designed to classify cataract and non-cataract fundus images with high accuracy and efficiency. A dataset of 1130 fundus images was augmented to 4746 images to improve model generalization. Experimental results show that DenseNet-121 outperforms all other models, achieving an accuracy of 92%, with a precision of 91%, recall of 90%, and an F1-score of 90.5%. The system also incorporates data augmentation and attention mechanisms to enhance its robustness and scalability. Our proposed model, CatCNNNet, offers a practical solution for real-time cataract detection and can be deployed in both clinical and mobile health applications. Future work will focus on further improving the model’s scalability and exploring interpretability techniques for clinical use.