Segmentation of Drusen in Retinal Images for the Detection of Diabetic Retinopathy

Abstract

Diabetic retinopathy (DR) is a most complicated eye disease that affects people with diabetes for an extended period. If necessary treatment is not given at the right time, DR leads to a severe loss of vision without prior symptoms. Therefore, patients with diabetes are recommended to undergo continuous screening for early detection of DR. In this paper, we proposed an automated detection process to detect the lesion called Drusen in retinal images. Drusen is not related to DR. The presence of Drusen does not indicate the disease DR. Still, it is used to identify the severity level of diabetes for the patient and to avoid the misdetection rate with other bright lesions (both exudates and cotton wool spots). This method is based on the Background image approach and inverse segmentation to detect the area affected by Drusen in retinal images. Inverse segmentation is used to segment the healthy areas based on regular texture rather than varying the texture of unhealthy areas. The segmented healthy areas are compared with the original image for segmenting Drusen. The segmentation process involved 40 images from the STARE database, producing better results based on accuracy and processing time.