Neural Computing BasedAbnormality Detection inRetinal Optical Images

Automated eyedisease identification oncolour features todifferentiate theabnorm:alities. systems facilitate theophthalmologists inaccurateIn(4), region growing techniqes on graylevel diagnosis andtreatment planning. Inthispaper,images areproposed fortheabnormality detection. anautomated systembasedonartificial neural Datamining systems areusedforthedecision support network isproposed foreyedisease classification, systemdevelopment in theeye abnormality Abnormalretinal imagesfromfourdifferent classification applications (5). Bloodvessel detection classesnamely non-proliferativ e diabeticbasedclassification approaches arealsocited inthe retinopathy (NPDDR), Centralretinalvein literature. Klandiraju etal(61haveproposed ablood occlusion (CRVO), Choroidal neo-vascularisation vesseldetection algorithm basedon recutrsive membrane (CNVM) and Centralserous hierarchical decomposition andquadtrees. The retiinopathy (CSR)areusedinthiswork.A matched filter technique isthewidely usedtemplate suitable feature setisextracted fromthepre- based technique forclassification applications (7). An processed imagesand fedto theclassifier, integrated automated analyzer hasbeenproposed by Classification ofthefoureyediseases isperformedCreeetal(8). Morphological operations based blood using thesupervised neural network namely back vessel detection isproposed intheliterature (9). But propagation neural network (BPN). Experimental these classifiers suffer fromn thedrawback ofnon- results showpromising results fortheback adaptive parameters. propagation neural network asadisease classifier.Another groupofresearchers highly dependon The results arecomparedwiththestatistical advanced computing techniques withintelligence for classifier namely min:imum distance classifier to retinal imageapplications. Gardner etal(10) usedthe justify thesuperior natureofneuralnetwork artificial netral networks forthedetection oflesions basedclassificat'ion. ingrayscale images. Abnormality classification using Index terms- Retinal images, Backpropagationperceptron havebeenimplemented byPovilas (111). neuralnetwork, Classification accuracyand Support vector machine basedabnonnality detection Sensitivity.