Segmentazione automatica dei vasi sanguigni in immagini wide-field della retina di neonati affetti da retinopatia del prematuro attraverso classificazione supervisionata

Retinopathy of prematurity (ROP) is a disease that can affect premature infants and which can lead to retinal detachment and visual loss if not correctly and promptly healed. It is characterized by an abnormal development of retinal vasculature, which doesn’t reach the periphery of the retina. In fact, the development of retinal network usually becomes complete around full term at 9 months, that is the reason why in premature infants the vessels are not completely developed. Our purpose is to create a suite of algorithms able to automatically detect the vascular network in the retina and, basing on that, calculate the main parameters of the vessels, as extension, tortuosity and dilatation. All the pre-existent algorithms we know have been created for the segmentation of adult image, taken with fundus camera. In our program we use images of the infant eye, that have a lower quality and that are taken with a RetCam. In order to achieve our target, we filter the images at our disposal and, with the results, we create a set of features that a support vector machine (SVM) classifier uses to make a prediction of the vessels in the images. The results that we obtained can be very useful for the follow up of the extension of the vessels in the retina, that is one of the principal hallmark of ROP; moreover some features can be used for the extraction of other parameters as tortuosity or dilatation of the vessels