Retinal Vessel Analysis for Early Detection of Retinopathy of Prematurity Using Image Processing

Abstract

Retinopathy of Prematurity is an eye disease that occurs in premature babies[1] where the major risk factors include birth before thirty-four weeks of gestation, low birth weight (less than 2 kg), and high exposure to oxygen. Early detection of these abnormalities can reduce the rate of blindness worldwide. The main objective of this work is to develop an algorithm using image processing techniques to analyse important retinal blood vessel features such as vessel width and tortuosity for the detection of Retinopathy of Prematurity (ROP). The process begins by converting retinal fundus images into grayscale and resizing them to a fixed size so that all images can be processed in a consistent manner. After that, a Gaussian filter is applied to reduce noise, and adaptive histogram equalization is used to improve the image contrast and make the blood vessels more visible. Once the pre-processing stage is completed, the optic disc is removed because it can affect the vessel analysis. The blood vessels are then enhanced using a bottom-hat transform, and global thresholding is applied to obtain a binary image of the vessel network. Small unwanted branches and noise are removed so that only the major vessel structures remain. Skeletonization and geodesic path analysis are then carried out to segment the vessels and extract the required features for further analysis. The vessel width is estimated by measuring the distance between each vessel centreline pixel and the vessel boundary using a distance transform technique. Tortuosity is calculated by comparing the actual length of a vessel with the straight-line distance between its starting and ending points.it just display the values of vessel width and tortuosity it does not directly classify whether ROP is present or not. Based on this, it assists doctors in making quicker decisions and may help reduce the chances of vision-related complications in premature infants through early detection.