Blood vessel segmentation for diabetic retinopathy

Abstract

DR or Diabetic Eye Disease is a medical condition which causes blindness in people with diabetes. It is found to be a proceeding cause of preventable blindness. The lack of conduction of retinal screening examination on all diabetic patients has let to many undiagnosed and thereby untreated cases of DR. Timely and accurate diagnoses can reduce the rate vision loss if patients with DR are referred to an ophthalmologist for evaluation & treatment. This study aims to bring about a robust diagnostic technology in order to automate DR screening. For the automated DR detection, a data-driven deep learning algorithm was developed and evaluated as a novel diagnostic tool. Colour fundus images were processed by this algorithm and classified them as having DR or healthy, identifying medically relevant cases for referral. For further clinical review, all the learned information from the automated method was readily visualized through automatically generated abnormality heat map, which highlighted sub-regions within each input fundus image. This study enables to identify cases that should be referred to an ophthalmologist for further evaluation and treatment, with use a fully data-driven artificial intelligence based grading algorithm which can screen fundus photographs from diabetic patients. On a global basis, the implications of such algorithm can drastically aid to reduce the rate of vision loss caused by DR. The model is executed in two phases with the purpose of strengthening the framework of Diabetic Retinopathy (DR) recognition