Automatic Cataract Classification System Using Neural Network Algorithm Backpropagation

Based on data from the World Health Organization in 2001 Indonesia is one of countries with the highest blindness rates in the world with the addition of new sufferers reaching 210,000 people per year. Of the 250 million population, there are only 1160 opthalmologist with uneven distribution. Cataract is one of disease such as macula degeneration, diabetes retinopatty. In this paper, classification of cataracts is divided into 4 normal retina, mild cataract, medium and severe. the classifier-making procedure includes four parts: pre-processing, segmentation, feature extraction, and classification. pre-processing using HSV to search for the highest level of light intensity, GLCM is used on feature extraction to obtain features that will be used to classify using Network Backpropagation that has great potential to improve the diagnostic efficiency diagnostic accuracy. In this research use image processing in detecting cataract characteristic in fundus image based on opacity level of optic disc. The data used were 60 retinal fundus images consisting of 15 normal retinal images, 15 light cataract images, 15 medium cataract images and 15 severe cataract images. The result of simulation test using MATLAB R2014a software obtained the normal retinal grade accuracy value of 95.71% with 95.7% sensitivity and 96.15% specificity, mild cataract 69.97% with sensitivity 69.97% and specificity 89.47%. Accuracy of medium cataract class is 75.69% with sensitivity 75.69% and specificity 92.75%. The accuracy of severe cataract class is 87.13% with sensitivity 87.13% and specificity 98.56%. The average accuracy value of the cataract classification system was 82.14%.