Automatic Classification of Pterygium-Non Pterygium Images Using Deep Learning

Abstract

Pterygium is an ocular disease caused by the invasion of a fibro-vascular tissue onto the cornea region. Several researches has been developed for automatic detection of pterygium in eyes images. In those researches, color and shape information of pterygium were explored using Digital Image Processing techniques and Machine Learning algorithms such as Artificial Neural Networks (ANN) and Support Vector Machine (SVM). More recently, Deep Learning techniques were applied for implementing a system for diagnosing multiple ocular diseases including pterygium, however no study have been developed on using Deep Learning focused on ptyregium detection only. We present a method for automatic classification of pterygium - non pterygium images using Convolutional Neural Networks (CNN). A dataset of positive (pterygium) and negative (non pterygium) images, previously used in early researches, was employed in order to train and to test a CNN model with one convolutional layer. The images were studied in two color formats, RGB and grayscale. The best result in the pterygium – non pterygium image classification task was attained using RGB format, getting an Area Under the ROC curve of 99.4%. The results obtained overcome the results found in literature.