Automatic Segmentation of Optic Nerve Head by Median Filtering and Clustering Approach

The optic nerve head (ONH) is a sphere area with light-colored on the fundus image. It needs to be observed by an ophthalmologist to detect glaucoma. Glaucoma is an eye disease that may cause permanent blindness. It can be detected based on the cup-to-disk ratio (CDR) value. This value is generated by calculating the diameter length of the ONH. In order to perform these calculations, it is necessary to segment the ONH area. This study aims to develop an ONH area segmentation method that consists of four main processes: detection of the region of interest (ROI), pre-processing, segmentation and post-processing. ROI detection is implemented in the green channel using the OTSU method, followed by pre-processing using the median filtering, which aims to discard the blood vessel. Furthermore, K - Means is applied to the segmentation process, followed by post-processing using several morphological operations to remove the appearance noise. This method successfully achieves the F1
score
value of 0.941 with test data of 68 images.