A Comprehensive Texture Segmentation Framework for Segmentation of Capillary Non-Perfusion Regions in Fundus Fluorescein Angiograms

摘要: Capillary non-perfusion (CNP) in the retina is a characteristic feature used management of wide range retinal diseases. There no well-established computation tool for assessing extent CNP. We propose novel texture segmentation framework to address this problem. This comprises three major steps: pre-processing, unsupervised total variation segmentation, and supervised segmentation. It employs state-of-the-art multiphase model which enhanced by new kernel based region terms. The can be applied intensity-based problems. A step allows take expert knowledge into account, an AdaBoost classifier with weighted cost coefficient chosen tackle imbalanced data classification To demonstrate its effectiveness, we 48 images from malarial retinopathy 10 ischemic diabetic maculopathy. performance satisfactory when compared reference standard manual delineations: accuracy, sensitivity specificity are 89.0%, 73.0%, 90.8% respectively dataset 80.8%, 70.6%, 82.1% maculopathy dataset. In terms region-wise analysis, method achieved accuracy 76.3% (45 out 59 regions) 73.9% (17 26 comprehensive quantify capillary two distinct etiologies, has potential adopted wider applications.