Detection of high-risk macular edema using texture features and classification using SVM classifier

Abstract

In digital retinal images, positive means of texture feature detection around the macula region with specified radius is still an open issue. Diabetic macular edema is a complication caused due to Diabetic Retinopathy (DR) and is the true cause of blindness and visual loss. In this paper, we have presented a computerized method for texture feature extraction around the specified radius taking macula as the centre. By proper segmentation techniques, the region of 1DD (Disc Diameter) around the macula centre, was extracted out. The extracted region contained a great amount of abnormalities like micro-aneurysms, hard-exudates and hemorrhages, thereby texture features varied greatly. Unlike other well-known approaches of machine learning classifier techniques, we propose a combination of texture feature extraction from the region of interest around macula and grading using Support Vector Machine (SVM) classifier. The segmented region containing abnormalities differ greatly in texture and a promising “accuracy > 86%” was obtained between the “normal” and “abnormal” type classification. The performance evaluation of the automated system was determined by parameters, namely Sensitivity, Specificity and Accuracy with values obtained about 91%, 75% & 86 % respectively.