DIFFERING MATCHED FILTER RESPONSIVITY FOR THE DETECTION OF PROLIFERATIVE DIABETIC RETINOPATHY

摘要

Diabetic retinopathy (DR) is a retinal vascular disease andrnis one of the most common causes of blindness worldwide.rnProliferative diabetic retinopathy (PDR) is the most advancedrnstage of the disease and poses a high risk of severernvisual impairment. PDR is characterised by the growth ofrnabnormal new vessels known as neovascularisation. In thisrnpaper, we propose the use of the matched filter (MF) techniquernfor vessel segmentation with emphasis on using tworndifferent sets of parameters to allow for the detection ofrnnew vessels. Parameters are selected to first increase andrnthen decrease the MF response to new vessels, followed byrnthresholding to produce two separate binary vessel maps.rnThe difference image removes most normal vessels andrnretains all possible new vessels, therefore making furtherrnanalysis a much simpler task. Several steps are also includedrnto reduce the detection of non vessel objects (darkrnand bright lesions). Five local features associated with thernmorphology of the vasculature are used to create a featurernset. Based on these features, regions of the retina are categorizedrnas normal or abnormal using a k-nearest neighbourrnclassifier. Sensitivity and specificity results were 100% andrn70% respectively on a per image basis.