Auto Detection and monitoring of diabetic retinopathy using super resolution technique on retinal images

Abstract

Diabetes is a rapidly increasing worldwide problem. The most common complication of diabetes is diabetic retinopathy, which is one of the primary causes of blindness. Treatments can prevent loss of vision and blindness in most cases. However, treatment cannot restore vision that is already lost. Therefore, if you have diabetes, it is vital that you have regular eye checks to detect retinopathy before your vision becomes badly affected. The rapid increase of diabetes pushes the limits on the current diabetic retinopathy screening capabilities for which the digital imaging of the eye fundus (retinal imaging), and Computer Aided Diagnosis (CAD) provide a potential solution. newlineThe present research work provides a system for auto-detection and monitoring of diabetic retinopathy. The current and past retinal images of a diabetic patient are used as inputs to the system. Initially both the images are pre-processed to enhance their quality. A local histogram equalization technique is modified so as to be time efficient and used for contrast enhancement of retinal images. The images are registered using a feature based registration technique. Optic disc, macula and vascular network are unique anatomic structures of retinal image. The proposed registration method uses bifurcation points of vascular network as control points. The distances of control points from informative points within optic disc and macula centre in respective image are used for obtaining corresponding points in two images which are to be registered. These images are compared using change detection algorithm. The contribution of the present work is that, the accuracy of optic disc detection, left / right eye retinal image detection and macula detection algorithms is excellent which is pre- newlinerequisite for proposed image registration technique. In addition, an affine transformation model is designed for retinal image registration. An inter pixel accuracy is achieved in proposed registration technique.