A Real-Time Cataract Detection and Diagnosis Through Web-Based Imaging Analysis

Abstract

More than half of the cases of blindness in this industrialized world has their main cause as cataract. Loss of blindness and minimizing the suffering can be done by early diagnosis. However, due to the associated costs, clinical cataract detection and grading necessitate the expertise of qualified eye doctors, which may make everyone's early intervention difficult. The preset set of image features used in current analysis on automatic cataract detection and grading based on fundus pictures may give a redundant or even noisy conclusions. The effectiveness and efficiency of employing a deep convolutional neural network (DCNN) to automatically detect and grade cataract is investigated in this study. It also illustrates some of the feature maps at the pool 5 layer with their high-order empirical semantic meaning, explaining the feature representation extracted by the DCNN. Up to 5620 clinical retinal fundus images from hospitals, obtained as part of a population-based study, were used to cross-validate the proposed DCNN classification system. Two findings are proposed in this paper: First, interference from regionally inconsistent lighting and eye reflection was eliminated through the use of retinal fundus images after G-filter, which considerably enhances DCNN classification.