Optic Disc Segmentation Based on Mask R-CNN in Retinal Fundus Images

Abstract

An optic disc is an object on the retina of the eye that has the characteristics of being brightly colored and round. Optical disc segmentation is the most commonstep taken before processing a retinal fundus image. The bright characteristics of the optic disc often interfere withthe detection of other objects in the retinal fundus image. Therefore, the optic disc is the first step before processingthe fundus image of the retina. With the help of digital image processing will help in the removal of the optic discon the fundus image of the retina. Many methods can be used in optical disc segmentation, one of which is the deep learning method. The deep learning method chosen is Mask R-CNN to produce a mask from the results of object detection on the retinal fundus image. There are 3 stages in the segmentation process using the Mask R-CNN. First, the data used in the training process will be labeled. thereis 1 label given, namely optic disc. Then the model is trained using the restnet50 backbone architecture and finally, the model will be evaluated. To evaluate the results obtained from the two methods, it uses Intersection over Union (IoU) by comparing directly the results of prediction and ground truth. The data used is an IDRiD dataset containing retinal fundus images taken from eye clinics across India. As the result, Mask R-CNN can segment the optical disc with an IoU value of 0.843. it is hoped that the results of this research can help the process in processing retinal fundus images in the future.