Segmentation of vessels through supervised classification in wide-field retina images of infants with Retinopathy of Prematurity

The earliest signs of Retinopathy of Prematurity (ROP) are tortuosity and dilation of retinal vessels, both considered of primary importance for the diagnosis and the follow-up of the disease. However, an accepted computerized system for their quantitative measurement is still missing. Vessel segmentation algorithms designed for adults´ fundus images do not work well in infants´ fundus images. In order to provide an accurate segmentation of the infant´s retina vascularity we propose a supervised classification approach. The extracted feature set comprises multi-scales vesselness and texture features. Their combination allows the discrimination both between vessels and background, and between choroidal and retinal vessels. In order to reduce the computational complexity, the most useful subset of features to be fed to the classifier is determined employing a wrapper greedy forward-selection method. The performance of the proposed method is assessed by cross-validation on a dataset of 20 images acquired with a RetCam fundus camera. Average accuracy and Matthews´ correlation coefficient are respectively 0.97 and 0.67 with respect to manual ground truth reference.