Diabetic retinal imaging: methods in automatic processing

Abstract

AbstractFundus image based screening for diabetic retinopathy is offered to all diabetic patients aged 12 and older. This has proven to be an effective procedure for the early detection/diagnosis of diabetic retinopathy and forms the basis of current treatment plans. However, the increasing number of diabetic patients is putting a strain on the NHS. Computer based tools to aid detection of/grade diabetic pathologies are currently under development. In this MPhil a novel database of fundus is described. Many of whom also possess comorbidities such as glaucoma or hypertension. Retinal vessel masks were extracted by hand to establish accurate high-resolution images to test automatic vessel extraction algorithms on. Two previously published automatic vessel segmentation algorithms were tested on this database. This collection of images accurately represents the variety of fundus images a retinal grader can expect to encounter in a regional screening program in the UK. In addition retinal image quality can be significantly degraded by media opacities, limiting the diagnostic potential of retinal images. The amount of scattering increases with age and with some pathologies (e.g. cataract). Even though a large body of work exists on the enhancement of images recorded in poor visibility very little has been done on reducing the degradation of retinal images by intraocular scattering. In this thesis a defogging filter designed to enhance image clarity was applied to fundus images that had previously been graded as inadequate. 12% of these images were found to be assessable after filtering suggesting that a 'cataract filter' of this type may be beneficial in diabetic retinopathy screening programmes.

Date of Award	19 Aug 2015
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Niall Mcloughlin (Main Supervisor)