No abstract

Highly Reliable Automated medical diagnosis systems are of critical importance. Such systems aid in early detection of diseases and prevention of its further progression. Development of such a reliable and efficient software system is possible using a suitable system development life cycle (SDLC) model only. A SDLC model develops a system in a structured, deliberate and methodical mode and provides a very reliable and efficient system within limited resources and time. Macular edema is the blurring or loss of central vision which is caused as a result of Diabetic Retinopathy and Analysis of OCT images helps in identification of Macular Edema. The aim of this research is the successful detection of Macular Edema using Iterative Prototyping SDLC model. First the extraction of ILM layer has been done by using Active Contour based Segmentation and Curve Fitting Techniques then a new technique is proposed in this research for the successful localization of fovea in retinal ILM layer by using distance based method. Finally the detection of Macular edema has been done on the basis of analysis of fovea region. The system is evaluated using a local dataset of OCT images which is gathered with the help of Armed Forces institute of Ophthalmology. The dataset consists of 550 images and the developed system gives an accuracy of 84%.

Abstract-Highly

Abstract. Choroidal Neovascularization (CNV) is an age related disease which deals with the Degeneration of Macular tissue. This degeneration causes acute drop in central vision as the age progresses. Therefore it is necessary to identify the changes caused by CNV for the Successful detection of this disease. In CNV the Retinal Pigment Epithelial (RPE) layer encounters changes in different attributes which can be identified with the help of Optical Coherence Tomography (OCT) Images. This paper focuses on analyzing the changes caused in the RPE layer due to CNV. The proposed system segments out RPE layer and observes the changes in RPE layer by calculating different features like Euler Number, Energy, Homogeneity and Correlation. The system is tested on locally gathered dateset of 50 images from different patients and has achieved an accuracy of 98%.