Graph Based Optic Disc Detection with Multi Threshold Segmentation using K-Means Clustering for Retinopathy Classification

Abstract

Listen

Translate article

A combination of Graph cut technique and k-means clustering with multi threshold medical image segmentation for retinopathy classification is presented. The main objective of medical image segmentation is to extract the anatomic structures and its characteristics with respect to some input features. There exist various methodologies for medical image segmentation and retinopathy classification but struggles with missing features and false exudates in the medical images. We propose a new technique to increase the resolution of the medical images to identify the features and edges of the medical images. The medical image is preprocessed to reduce the noise, and then multi level histogram is generated in the first phase of the process. In the second stage, with the initial segmentation obtained with gray level contours, and we generate the histogram then we construct the pixel adjacency graph in which each nodes represents the set of pixels of the image and edges links the neighbor pixels. We calculate the similarity of neighboring pixels; based on the distance value we cluster the similar pixels to a class. We use k-means clustering to group similar pixels and we used Euclidean distance measure to calculate the similarity between pixels. The optic disc is identified using the segmented results and the proposed method is carried out iteratively until the user gets satisfied. The method will be carried out repeatedly with the user defined threshold value. The threshold is used to group pixels with in the distance. With the proposed technique the features are maintained and the resolution of the image enhanced. The time complexity of the process is reduced.