Automated Classification of Retinal Diseases in STARE Database Using Neural Network Approach

Abstract

The optics of the eye create a visual image of the visual world on the retina. Incoming light signal is converted into a neural signal, which in turn is processed by the visual cortex in brain. A healthy retina is crucial for reliable vision. It is vulnerable to organ-specific and systemic diseases as numerous imperative ailments manifest themselves in the retina. Retinal dystrophies and degenerations are often the cause of visual loss and complete blindness in severe cases, hence early diagnosis and appropriate treatment can avert the loss. Various retinal diagnostic techniques performed manually by the ophthalmologist are conventional procedures followed in numerous parts of the world. Since human intervention is highly prone to errors, these strategies don’t generally ensure high level of accuracy. Consequently, computerized procedures are significantly crucial for useful applications in the ophthalmology. The purpose of this research was to develop an automated diagnostic system that will be able to identify patients with retinal disorders from images using neural network. This study comprises of four main sections. Data related to retinal pathologies was taken from a publicly available fundus image database. Collected data was then pre-processed by applying exclusion and inclusion criteria on categorized diseases and then visualized using MATLAB. Neural network technique along with three different activation functions (Sigmoid, Gaussian and ArcTan) were used to classify multiple retinal diseases allowing timely detection of such ailments with high accuracy. Sigmoid and Gaussian function gave best performances across all performance metrics. Accuracy calculated for Sigmoid was 0.92, for Gaussian 0.90 and for ArcTan 0.46