Classification and grading of cataracts using a deep convolutional neural network

Abstract

Cataract, a leading cause of partial impairment and blindness worldwide, is characterized by the clumping of protein in the transparent lenses, resulting in a white-colored covering. This eye condition is attributed to various factors, including developmental abnormalities, trachoma, metabolic disorders, genetics, drug-induced changes, and aging. Among these causes, genetics and aging are the primary contributors to cataracts that lead to blindness. Cataracts can be classified into three types based on the location of lens opacities nuclear, cortical, and posterior subcapsular. Additionally, ophthalmologists grade cataracts according to the severity of the disease. The grading system aids in early detection and timely treatment. Notably, previous research overlooked the influence of eye color, despite its impact on cataract development stages. Hence, this study aims to classify and grade cataracts while considering the effect of eye color, specifically within the Ethiopian population. This study encompasses various cataract classes, including normal, nuclear, cortical, and posterior subcapsular. Moreover, the severity of each cataract type is graded, encompassing cortical early, cortical advanced, nuclear early, nuclear advanced, posterior subcapsular early, and posterior subcapsular advanced stages. To achieve these objectives, an experimental research approach was employed, involving data collection, analysis, and processing. A model was designed to extract cataract features for effective classification and grading. The cataract classification and grading model demonstrated an accuracy of 74% using raw data, which significantly improved to 97% after implementing data preprocessing techniques. Furthermore, the model exhibited a 99% accuracy in grading cataract severity. Evaluation of the proposed model employed metrics such as accuracy, confusion matrix, precision, recall, and F1- score.