Keratoconus Detection using Hybrid Density Supervision model with Clustering and Classification Techniques

Abstract

Main challenges lie in detecting the keratoconus in the corneal layer of the human eye. The border region of the image is difficult to extract the edges present in the image. Existing drawbacks could be overcome using the proposed method in this paper, which combines the unsupervised clustering techniques, for predicting the intensity point of the input image and supervised linear clustering algorithm, for estimating the regions characterized with disorders.  The proposed work overcomes the problem in detecting the edges based on density and intensity of the layer, in an image. The method combines the unsupervised clustering technique and collects the intensity points on the boundary surface of corneal region. The intensity level may be the unknown label for the clustering process. Clustering is an effective tool used in such cases where grouping into similar subsets is desired which is characteristic of classification problems.  The endpoint is a set of clusters, where each cluster is distinct from each other cluster. The supervised learning model helps to collect the behavior from the database information. The density based threholding predicts to group the K-Means clustering techniques while Naïve Bayes helps in classifying the layers of cornea. The proposed method is implemented using the MATR2014a software and results were discussed..