Classification of Retinal Images Using Image Processing Techniques

In the framework of computer assisted diagnosis of retinal diseases, an algorithm for classification of retinal images for certain medical applications such as screening is presented and discussed. Automated detection of lesions/exudates in retinal images can assist in the early diagnosis of some common retinal abnormalities such as Diabetic Retinopathy (DR). In this paper, we employ different image processing techniques to automatically detect the presence of abnormalities in the retinal images. The proposed method consists of two stages. In the first stage, the bright optic disc region is located and removed from the image. This helps in avoiding the confusion between bright lesions and optic disc. The second step is to detect bright lesions in a retinal image. Depending on the existence of exudates, retinal images are classified into normal and abnormal categories. The proposed work has been tested on retinal images from different databases (DRIVE, STARE, DIARETDB1) and images obtained from a local eye hospital (Sankaradeva Netralaya). The proposed method successfully classifies all the normal images and shows a sensitivity of 98.2% in classifying the abnormal images.