Drusen exudate lesion discrimination in colour fundus images

Abstract

Automatic screening and diagnosis of ocular disease through fundus images are in place and considered worldwide. One of the leading sight loosing disease known as age related macular degeneration (AMD) has many proposed automatic screening systems. These systems detect yellow bright lesion and through the number of lesion and their size the disease is graded as advance and earlier stage. It becomes difficult for these systems to differentiate drusens from exudates another bright lesion associated with Diabetic retinopathy. These two lesions look similar on retinal surface. Differentiating these two lesions can improve the performance of any automatic system. In this paper we proposed a novel approach to discriminate these lesions. The approach consists of two stage procedure. The first stage after pre-processing detects all bright pixels from the image. The suspicious pixels are removed from the detected region. On the second stage bright regions are classified as drusen and exudates through Support Vector Machine (SVM). Proposed method was evaluated on publically available dataset STARE. The system achieve 92% accuracy.