Automatic Detection of Microaneurysms in Retinal Image by Using Convolutional Neural Network

Abstract

One of the early findings of diabetic retinopathy is microaneurysm. Microaneurysms look like dark small dots in the retinal images, thus it is difficult for physicians to detect them visually. On the other hand, convolutional neural network (CNN) showed superior performance in image recognition studies. This paper describes about the automated microaneurysms detection by using CNN and the performance. First, retinal images were enhanced by using shape index based on Hessian matrix, double-ring filter and Gabor filter. Microaneurysm was then detected by using CNN. Finally, false positives were reduced by using another CNN and three-layer-perceptron with 48 features. By applying the proposed method to DIARETDB1, the performance was better than previous methods.