Median Filtering for Optic Disc Segmentation in Retinal Image

Abstract

One of diabetic complications is vision disturbance known as Diabetic Retinopathy (DR). Early detection becomes vital to prevent the development of DR by routine examination of the patient's eyes. This study developed a software application that can automatically segment the image of the eye retina, expected to assist eye specialists in performing DR detection using computer aids. The system implementation uses Visual C# 2010 software with 36 retinal images of the Drishti-GS1 dataset. The clinical features of the retina that can be used to detect DR are microaneurysms, hard and soft exudates, hemorrhages, neovascularization, and macular edema. Optic Disc (OD) segmentation is an important step to detect these features. This study aimed to detect OD using red channel, inversion, contrast enhancement, median filtering, and thresholding. The final result of the OD segmentation was validated by measuring the positive predictive value (PPV). The result of PPV for the detection of OD to ground truth image reached 90.924%.