SORTING OF RETINAL BLOOD VESSELS USING NEURAL NETWORKS

Abstract

In recent years, computation techniques are emerging with a good speed in the medical field. A significant branch of biomedical field is said to be Ophthalmology; it needs automated techniques for identification of pathology. In health care units, since new technologies are increasing, their goal is to reduce the visits to specialised doctors. They aim at efficiency of doctor which means how many cases a doctor can deal with proper diagnosis. Also, it might reduce the cost of overall procedure for both patients and clinic. Diabetic retinopathy being one of the diseases which if encountered with at later stage might cause serious problems like blindness. So, we need to detect it as soon as we can, and this is done with the help of segmentation of retinal images. There are many retinal diseases which are characterized by modification in retinal vessels. Retinal blood vessels are of two types: arteries and veins. Irregular wide veins are the symptom of Diabetic Retinopathy which leads to low Artery to Vein ratio(AVR). The technique used here is implementing a supervised classification for retinal vessel detection. These vessels go through few stages like pre-processing, feature extraction and hence classification is done. The classifier used is Neural Network which is fast, has good convergence speed and properties like generalization, non-linearity and more are discussed. Two types of feedforward Neural Network are used, Back propagation and Probabilistic Neural Network. It has to be found out which is better for this classification problem and why.