Feature extraction from retinal fundus image for early detection of diabetic retinopathy

Abstract

Automated detection of lesions in retinal fundus image can be aid in the detection of diabetic retinopathy. Exudates are the early sign of diabetic retinopathy so the proper detection of these lesions is an essential task in an automatic retinal screening. On the research work leading to automatic analysis of exudate detection, the knowledge of Optic Disk (OD) location is very useful. An efficient algorithm is presented to detect the OD and exudate which are the most important features for early detection of diabetic retinopathy. From a retinal fundus image, the proposed method first preprocesses and estimates the histogram of retinal background, then filters out the bright pixels in intensity image. They include OD, and non-OD (exudates and noise). Next, an OD boundary is determined and eliminated after applying blob boundary measurement and morphological reconstruction. Finally, exudates are extracted by applying the maximum entropy thresholding to filter out the bright pixels from the green component of retinal image which OD region inside is eliminated. The proposed technique has been tested first on 100 images from hospital. Experimental results show that 93% and 89% of OD and exudate were detected correctly, respectively.