| Popis: |
Retinopathy of prematurity (ROP) is a disorder of the eye that may develop in preterm-born infants. If left untreated, ROP may lead to retinal detachment and ultimately blindness. A warrant for treatment of ROP is the detection of plus disease, which is clinically diagnosed by the presence of certain levels of increase in the tortuosity and thickness of retinal vessels. The openness of the major temporal arcade (MTA), the thickest branch of the venules, has also been observed to decrease in the presence of plus disease. It has been shown that there is interexpert disagreement on diagnosis of plus disease, even among expert ophthalmologists and retinal specialists, which calls for development of image processing techniques to detect and diagnose plus disease more accurately by analysis of retinal fundus images of preterm infants. This work presents image processing methods for the extraction of diagnostic features from retinal fundus images of preterm infants, for the purpose of computer-aided diagnosis (CAD) of plus disease. The features include measures of the thickness and openness of the MTA as well as tortuosity of retinal vessels. The methods include directionally sensitive Gabor filters for the detection and extraction of vessels, as well as morphological techniques for segmentation, binarization, and skeletonization of the vasculature. Methods are proposed for tracking and segmentation of only the MTA. Using geometrical modeling and analysis, the openness and thickness of the MTA are quantified. Tortuosity of vessels is quantified by analyzing the variation in the dominant orientation of vessels obtained by the application of Gabor filters. Receiver operating characteristic (ROC) analysis of the diagnostic performance of the measures of thickness, openness, and tortuosity resulted in area under the ROC curves of 0.75, 0.70, and 0.98, respectively. The methods may be used in a clinical or teleophthalmological setting for CAD of plus disease. |
