Deep convolutional neural network based screening and assessment of age-related macular degeneration from fundus images

Abstract

In this paper, we provide a study on deep convolution neural networks for finding the appropriateness of using the transfer learning to screen an individual at risk of Age-related Macular Degeneration (AMD). We make use of the Age-Related Eye Disease Study (AREDS) dataset with over 150000 images which also provided qualitative grading information by expert graders and ophthalmologists. We use a modified VGG16 neural network with batch normalization in the last fully connected layers. For our study, we have conducted two experiments. First, we have categorized the images into two classes based on the clinical significance: No or early AMD and Intermediate or Advanced AMD. Second, we have categorized the images into four classes: No AMD, early AMD, Intermediate AMD and Advanced AMD. We have achieved the best accuracy with our modified VGG16 network which is 92.5% for the two class problem with more than one hundred thousand images. With accuracies ranging from 83% to 92.5%, we have demonstrated that the training of a deep neural network explicitly with a sufficient number of images fares better than using a pre-trained network, especially in AMD detection, and screening. We have also observed that the deeper neural network, i.e., VGG16 fares better than the other relatively shallower networks such as AlexNet for similar studies.