Extraction, Segmentation and Exudates Detection in Retinal Vessels using Morphological and Clustering with Neural Networks

Abstract

In ophthalmology, the crucial diagnostic factor is vascular network of human eye. The mutable dimensions of vessels and proportionately truncated contrast makes the task of fundus imaging segmental task a non-trivial one. We endorse a Deep Learning(DL)-based method to solve the hassle of detecting blood vessels in fundus imagery, a scientific imagery undertaking which has significant diagnostic relevance. The advantage of this method is that accuracy of segmentation and exudates detection is improved which enables us to detect many diseases like Diabetes in early stages. It can be used in bio-medical applications for Retinal image analysis, fundus, exudate detection and for applications like image fusion of CT Machine. This method not only uses both supervised and unsupervised methods but also employs the combination of both morphological and clustering methods. Thus, it provides better edge detection. The accuracy and sensitivity which are the metrics for performance will improve commendably. The existing methods have poor edge detection and the fundus exudates cannot be clustered. So, the proposed method overcomes these by combining a very efficient segmentation, extraction and multi-structured morphological processes for effective retinal vessel and exudates detection. Daubechies wavelet, GLCM features, Morphological process, K-means clustering are the basic methodologies used. The Neural Network classifier attains congruously high performance among the diverse practices related to the recognition of the patterns that are based on statistical supervision. The calculation of separation distance to the nearby training case classifies a new sample. Experimental result proves that by applying this method on retinal image, the detection of the blood vessels as well as exudates can be done productively with proposed sensitivity, accuracy, selectivity and ROC curve.