An efficient clustering algorithm for image segmentation of retinal OCT images

Successful automated diagnosis of diabetes inflicted retinal disease is based on a fast and precise clustering algorithm to improve the view of retinal layers with respect to background. This study proposes an improved k-means clustering algorithm for segmentation of retinal scans, which is able to improve analysis and diagnosis of retinal pathology. The novelty of this work is in the application of the structural similarity feature with a clustering algorithm. The algorithm provides four clusters of pixels as a result, out of which the most suitable clustering arrangement is selected for segmentation with respect to the background. The algorithm benefits most in cases where the pathology-generated features tend to lie in different clusters in different OCT images. This algorithm provides more precise segmentation as compared to using the clustering algorithm alone without increasing the computational complexity or time required for segmentation significantly.