Steerable wavelet transform for atlas based retinal lesion segmentation

Abstract : Computer aided diagnosis and follow up can help in prevention and treatment of diabetes and its related complications. Screening of diabetes related disease in the eyes is done by a special low cost fundus camera. A follow up of the patients visiting at di fferent time intervals for screening brings us to the problem of image analysis for change detection and its cost per patient. It is very likely that human annotations for the lesions may be erroneous and often time consuming. Since the ethnic background plays a signi cant role in retinal pigment epithelium, visibility of the choroidal vasculature and overall retinal luminance in patients and retinal images, an ethnicity based atlas can provide a solution, simplify the image processing steps and increase the detection rate. In this article, we present a novel method of building a retinal atlas of a speci c ethnic group and use this atlas for exudates segmentation. To improve the detection accuracy, steerable filters are used to enhance the lesions. Experiments with the publicly available HEI-MED dataset show the good performance of the proposed method.