Automated Cataract Detection from Eye Images Using Transfer Learning

Abstract

Automated cataract detection from eye images plays a crucial role in early diagnosis and prevention of vision impairment. This research proposes a deep learning model using transfer learning to detect cataracts from digital eye images. We utilize a pre-trained CNN backbone, fine-tuned on a balanced dataset containing both normal and cataract fundus images to improve generalization. Image preprocessing, augmentation, and normalization enhance classification performance. Experimental results demonstrate high accuracy, sensitivity, and specificity in distinguishing cataract cases from normal eyes, showing potential for clinical screening. The system is evaluated against baseline methods and state-of-the-art architectures. Our approach enables real-time detection suitable for deployment on clinical tools and mobile health platforms. Future work will focus on explainability and cross-dataset evaluation.