Computer aided approach for proliferative diabetic retinopathy detection in color retinal images

Abstract

Diabetic Retinopathy (DR) and its variants are the leading cause of vision loss among people with diabetics and working-age adults. Proliferative Diabetic Retinopathy (PDR) is the advanced stage of DR, which involves the formation of new vessels in the retinal surface. In this paper an algorithm is proposed for the patch wise extraction of texture and vesselness features. The vessel map for the feature extraction was obtained using gabor wavelet transform. The feature set obtained from the image patches are subjected to a random forest classification. The classifier outputs the result label and based on a rule based decision system, the images are classified into the corresponding stages: PDR, NPDR (Non Proliferative Diabetic Retinopathy) and Normal. The system shows higher specificity compared to the existing literatures and can be used as tool for screening of diabetic retinopathy.