Cataract is one of the major causes of visual loss worldwide. It refers to the gradual clouding of the lens inside the eye that changes the clarity of the person’s vision and significantly deteriorates their life quality. Cataracts especially affect the elderly. Detecting cataracts early and accurately is very important for their treatments. Currently, professionals are the main ones who diagnose this problem, and they use traditional ways of detecting that restrict those without resources to effectively find them. In order to offer solutions to such problems, the paper introduces a dependable computer-aided diagnostic system for the automatic detection and severity measurement of cataracts from only the fundus retinal images. It points out the system’s strategy of multi-scale feature fusion as a key method to dealing with issues like lack of data, unbalanced class distribution, and poor generalization. First, deep features are derived through pretrained architectures, e.g. Inception and Xception. After that, they are classified by CNN, ANN, and MLP based models. To the end, PCA-based feature fusion approach is used for dimensionality reduction without loss of discriminative information. The system is trained and tested by a set of 1835 annotated fundus images divided into four different levels of the disease. The study shows that the proposed system is able to achieve an accuracy of 98.75%, far exceeding existing methods. This article represents the effectiveness, scalability, and clinical use of the proposed system for accurate and early screening of cataracts.
Gautam et al. (Tue,) studied this question.