Fundus image-based automatic cataract detection and grading system.

Cataract forms when too much protein develops in the eye, making the lens cloudy, which causes vision loss. Accurate and on-time diagnosis of cataracts is the best way to avoid blindness. This study used deep learning-based convolutional neural network architecture with fundus images to diagnose the cataract at the early stage. The proposed architecture is optimized with Adam optimizer and uses soft-max function as a classifier to grade cataract severity into 4-stages: Mild, Moderate, No, and Severe. The reason for using fundus images is its precision about the detailed information needed to diagnose eye diseases early. In this study, the fundus images are collected from various openly available datasets and graded into 4-classes with the assistance of an expert ophthalmologist. The proposed work achieved 4 -class accuracy of 92.7%, the sensitivity of 98%, specificity (recall) of 92.75%, the precision of 92.75%, and an F1 score of 92.5% that perform superior that previous states of the art methods. [ABSTRACT FROM AUTHOR]