Automated Microaneurysms Detection from Retinal Fundus Images using Pixel Intensity Rank Transform
P. R. Wankhede* and K. B. KhanchandaniElectronics and Telecommunication Engineering, Shri Sant Gajanan Maharaj College of Engineering, Shegaon, India, 444203
Corresponding Author E-mail: pravin.india1@gmail.org
Abstract: Computer aided design is in demand for diabetic retinopathy screening systems. Microaneurysms are the first observations as sign of Non-prolific diabetic retinopathy. Number of microaneurysms in retinal image helps to find severity of diabetic retinopathy. This paper presents a method to detect microaneurysms in retinal fundus images. Non illuminated and varying field of view images preprocess initially, so that non lesion and lesion features can be clearly visible for proper detection. Then negative pixel intensity rank transform (PIRT) is used to find smooth regions and edges in retinal images. In next step, optic disc and blood vessels are subtracted from rank transform image. Finally true microneurysms candidates are selected using histogram thresholding. The proposed method is evaluated on publicly available datasets DIARETDB1 and E-optha MA. The performance parameters sensitivity, specificity, and accuracy achieved are 98.79%, 83.33%, and 97.75% respectively for DIARETDB1 database and 94.59%, 96.56% and 95.80% respectively for E-optha MA database. The results show that proposed method is able to detect microaneurysms efficiently in retinal images for diagnosis of diabetic retinopathy.
Keywords: Computer Aided Diagnosis; Image Enhancement; Image Segmentation; Microaneurysm; Rank Transform; Retinal Image Back to TOC