Cataract Detection Using Textural Features and Machine Learning Algorithms

Abstract

The World Health Organization report indicates that one of the world's leading causes of blindness is reported to be due to cataracts. Even though cataract majorly affects the elderly population; however, now they can be seen among minors too. Among the various types, the prominently three types of cataracts affect masses in high numbers which are nuclear, cortical, and post-subcapsular cataracts. Conventional methods of cataract diagnoses include slit lamp image tests by doctors which do not prove to be effective in classifying cataracts in the early stages and can also have inaccuracies in identifying the correct type of cataract. Existing work to automate the process has worked on classification based upon binary detection only or has considered only one type of cataract among the mentioned types for further expanding the system. Our system works on the detection of cataracts in an attempt to reduce errors of manual detection of cataracts in the early ages. From the literature, we analyzed that textural features with improved pre-processing showed satisfactory improvement in detection rate with different classifiers. Our proposed system was successful to classify images as cataract affected or as a normal eye with an accuracy of 96% using combined feature vectors from the SIFT-GLCM algorithm applied to classifier models of SVM, Random Forest, and Logistic Regression. The effect of using SIFT and GLCM separately has also been studied which leads to comparatively lesser accuracies in the model trained.