Classification of Corneal Pattern Based on Convolutional Neural Network

Abstract

The early discovery of the disease is a great achievement in management of the cornea. This paper presents an efficient approach for the classification of normal and abnormal corneal patterns based on deep learning. Convolutional Neural Networks (CNNs) are utilized for this purpose. The CNN model built for this purpose comprises5 layers. The classification process is achieved through two stages. Automatic feature extraction based CNN is applied in the first stage, followed by sequence of processing layers includes: pooling layer, dropout layer and fully connected layer resulted in a diagnosis of the condition of the patient in terms of normal or abnormal. The proposed technique was tested and evaluated based MATLAB environment on a set of corneal images. These images were collected for patients based on confocal microscopy. The CNN classification results on corneal fundus images recorded an accuracy of 100 %.