A framework for automatic analysis of digital fundus images

Abstract

The optic disc (OD), the blood vessels and the macula are the most important features in the retinal images. These features are used for automatic eye screening systems that provide an accurate and efficient tool for the early detection of many eye diseases. A method for detecting these features is presented in this paper. First, the blood vessels are detected by using the mathematical morphology. Then, based on the percentage of the brightest pixels in the OD, the temporal side is detected since it contains the brightest region in the OD. By combining the information from temporal side and blood vessels, the whole OD is segmented. Finally, the macula is extracted by using the spatial relationship with the OD. The proposed method is tested on two publicly databases DRIVE and DIARETDB1. The detection of the OD achieved a success rate of 97.5% and 95.5% for DRIVE and DIARETDB1 respectively while the macula is detected correctly with a success rate of 100% and 97.6% respectively.