Cataract Detection and Grading with Retinal Images Using SOM-RBF Neural Network

A cataract is the prevailing cause of visual impairment in the modern world. The detection of cataract at early stages can lessen the risk of blindness. This study presents an automated system for cataract detection and grading based on retinal images. The system is comprised of image acquisition, preprocessing, feature extraction, classifier building, and cataract detection and grading. The preprocessing steps such as green channel extraction, histogram equalization, and top-bottom hat transformation are used to improve the quality of retinal images. The wavelet and texture features are extracted from the fundus image for building a classifier. A combination of SOM (Self-Organizing Maps) and RBF (Radial Basis Function) neural network has been taken to obtain better prediction accuracy of cataract. SOM-RBF neural network is evaluated on Tongren dataset with 8030 subjects categorized into four classes: Normal, Mild, Mature, and Severe. The proposed method achieved 95.3% and 91.7% of accuracy for cataract detection and grading tasks, respectively. The experimental results indicate that the proposed method performs better than the traditional RBF and other baseline methods.